SEO & Digital Marketing News Round-Up 5 October to 19 October 2026

On By DaveIn Screaming Frog, seo, wordpressLeave a comment

TL;DR

The September spam update finished on 8 October after 13 days and 16 hours. Three waves of ranking volatility hit sites along the way. Google then quietly tightened its helpful content guidance, published its first official crawl, index and recovery timing ranges, and added more inline links, carousels and creator profiles to AI Mode. OpenAI shipped GPT 6 with Intelligent UI. Google followed with Gemini 4 Argon. Bing opened a Topics Portfolio beta that may be the best free AI visibility report currently available.

New since Friday: Fresh ranking volatility spiked over the weekend of 17 to 19 October. Google denies a new update. The industry suspects one. OpenAI confirmed a ChatGPT bug that dropped UTM tags on outbound links from 8 to 15 October. Nozzle revealed that roughly 8% of Google SERPs served to rank trackers since mid September were poisoned with fake results. Google also pushed Gemini 4 Argon into Antigravity and added a Follow button to Discover videos.

Three jobs this week: date impact analysis from 24 September to 8 October, mark 8 to 15 October in analytics as a ChatGPT data glitch, and claim the Bing Topics Portfolio beta.

What’s new since Friday 16 October

  • 15 October: Google rolled more inline links, source blocks, article carousels and creator profile links into AI Mode and AI Overviews.
  • 16 October: A ranking volatility spike hit the SERPs. Barry Schwartz flagged an unconfirmed update. Google’s dashboard still shows nothing.
  • 14 to 15 October: OpenAI confirmed and fixed a bug that stripped UTM tags from ChatGPT outbound links between 8 and 15 October. Adobe tracked roughly 31,000 ChatGPT visits per day misclassified as direct traffic across one 500 page panel.
  • 10 October: Nozzle published research showing 8% of SERPs served to rank trackers since 13 September were poisoned, versus a baseline of 0.17%. The share peaked near 27% between 27 and 29 September.
  • 16 October: Gemini 4 Argon reached Google Antigravity for AI Pro subscribers. Google Discover added a Follow button to its video content viewer.
  • Over the weekend of 17 to 19 October: WebmasterWorld threads filled with reports of near zero traffic, stale News and Discover feeds, and the strong suspicion of a core update that Google has not confirmed.

Fourteen days, three waves of volatility, two major AI launches, one giant data glitch. This round up covers 5 October to 19 October 2026, what it means, and what to do about it.

What To Actually Know

Six developments shape planning for the rest of Q4:

  1. The September 2026 spam update finished on 8 October at 04:37 ET. Rollout time: 13 days and 16 hours. Normal spam updates in 2026 took two days. Google called this one “a normal spam update.” Barry Schwartz, Glenn Gabe and Lily Ray all described effects closer to a core update.
  2. Fresh volatility hit on 16 October. It kept running through the weekend. Google’s Search Status Dashboard shows no new update. Forum chatter leans heavily towards one being in progress.
  3. Google published official crawl, index and recovery timing ranges at Search Central Live Deep Dive Europe on 2 October. Core update recovery typical: 3 to 6 months. Slowest: 6 months to a year.
  4. Helpful content guidance tightened on authorship and AI on 1 October and again on 5 October. Fake author profiles are now called deception. Unreviewed AI content gets called out by name. Spam policy did not change. The quality lens did.
  5. OpenAI shipped GPT 6 and Intelligent UI on 7 and 8 October. Google followed with Gemini 4 Argon on 13 October. Answers now contain interactive tools, forms and charts. Plain text answers are becoming the exception.
  6. Rank tracking data is unreliable. Nozzle found Google serving fake results to tracker tools on 8% of SERPs since mid September. Combined with the ChatGPT UTM bug from 8 to 15 October, nothing outside Search Console or Bing Webmaster Tools can be fully trusted right now.

1. The September 2026 Spam Update finished, but the aftershocks haven’t

Google Search Central confirmed completion on 8 October. The quote: “The rollout was complete as of October 8, 2026.” Rollout time: 13 days and 16 hours. That’s longer than any 2026 spam update by an order of magnitude.

The 2026 spam update timeline

Update Started Finished Rollout time
March 2026 24 Mar 25 Mar About 19.5 hours
June 2026 24 Jun 26 Jun About 2 days 1 hour
August 2026 18 Aug 21 Aug About 2 days 16 hours
September 2026 24 Sep 8 Oct About 13 days 16 hours

The three waves of volatility

Relative volatility across the rollout. Aggregated signal from Semrush Sensor, Mozcast, AccuRanker and WebmasterWorld threads.

24 Sep
Launch, moderate

25 to 27 Sep
Wave 1, very high

28 to 29 Sep
Quiet

30 Sep to 1 Oct
Wave 2, high

2 to 3 Oct
Elevated

4 to 6 Oct
Wave 3, strongest signal

7 Oct
Winding down

8 Oct
Rollout complete

16 to 19 Oct
New spike, source unconfirmed

Sources: Barry Schwartz (Search Engine Roundtable), Glenn Gabe (G Squared Interactive), aggregated via Numinix and Distribb.

What got hit, what came back

Lily Ray flagged a representative casualty. A local services site had 143 county landing pages built from a single template. Traffic rose fast, held, then evaporated. The pattern repeated across several site categories:

  • Programmatic or templated pages with little unique value.
  • Unreviewed AI content at scale.
  • Doorway pages that funnel traffic to a different destination.
  • Parasite SEO setups piggybacking on domains with trust they did not earn.

On the recovery side, Glenn Gabe reported Reddit’s AI translated threads climbing again in Italy, Germany, Brazil and France after being crushed by the May and June updates. Edward Sturm posted that Reddit’s estimated organic traffic in Semrush hit an all time high. One site hit by the August 2026 update went from close to zero to roughly 8,000 clicks per day during this rollout and reappeared in AI Overviews and AI Mode.

What the community is saying

“Google called this ‘a normal spam update’ on LinkedIn. Taking two weeks to roll out is not normal, and the impact felt core update adjacent for a lot of sites.”

Barry Schwartz, Search Engine Roundtable, 8 October

“Several site owners emailed me about large drops starting 30 September. This update rolled out in stages. The industry just saw them as one update.”

Glenn Gabe, G Squared Interactive

“Every single website I monitor or have analytics access to is seeing barely any traffic today.”

WebmasterWorld member, 16 October

“Sigh, was hoping that the end of the update would stabilize things, but it’s just made them worse.”

WebmasterWorld member, 16 October

“Seems like a Core update on the way.”

WebmasterWorld member, 19 October

Charles Floate flagged a different problem on LinkedIn. The major SEO suites look out of sync with Search Console. Google’s anti scraping changes are hiding the real picture from third party tools. Ahrefs posted a notice that some SERP data may be inconsistent during the rollout.

2. Google published its first official crawl, index and recovery times

At Search Central Live Deep Dive Europe in Barcelona on 2 October, Google’s Gary Illyes shared the first on record “typical” and “slowest” timing ranges for the full search pipeline. Search Engine Journal published them on 4 October. Think of these as the numbers finally safe to quote to a client.

Crawling and indexing

Event Typical Slowest
New URL discovered About 20 hours Weeks to never
Known URL refreshed About 30 days Weeks to never
Sitemap processing About 24 hours Up to 14 days, or never
End to end indexing About 1.5 hours Months, or never (quality)
Canonicalisation change 1 to 3 weeks Months
Site move 1 to 3 months 6 months to a year or more

Search results and recovery

Event Typical Slowest
Title change picked up 1 to 2 days Several weeks to months
Snippet change picked up 1 to 2 days Several weeks to months
GSC URL removal (owner) About 2 hours 24 hours
Manual action removal 1 to 2 weeks 4 to 6 weeks, longer for dormant sites
Core update recovery 3 to 6 months 6 months to a year

Illyes was clear that these are reference points, not deadlines. The typical core update recovery of 3 to 6 months is the single most useful stat Google has volunteered this year. It settles plenty of nervous conversations.

3. Google quietly tightened its helpful content and AI guidance

No new algorithm was announced. Instead, Google edited two Search Central documents on 1 October and again on 5 October. The pages: helpful content, and generative AI content. Barry Schwartz surfaced the authorship change on 6 October.

Three changes that matter

  1. Fake author profiles are now explicitly deception. The updated page calls out AI headshots, invented names and false credentials. Deception makes a page untrustworthy to users and to automated quality systems. The last phrase is new.
  2. Unreviewed AI content gets named directly. The generative AI guide tells teams to manually fact check AI output before publishing. Titles, meta descriptions, structured data and image alt text all included.
  3. Main Content got elevated. The helpful content page now defines Main Content as the primary thing that answers the user. Four quality markers follow: effort, originality, skill and experience, and accuracy.

At a separate European event on 5 October, Google reportedly told attendees that spam updates now use AI to detect AI spam. The most 2026 sentence of the year, and a strong hint at what the September rollout targeted.

Second order effects came fast. On 2 and 3 October, affiliate networks began asking publishers for proof of human authorship. By week two, some unionised newsrooms reportedly signed contracts banning synthetic bylines. Any content supply chain that includes third party writers now faces a compliance headache for Q4.

4. AI search: GPT 6, Gemini 4 Argon, and the end of plain text answers

The biggest week for AI search product news in 2026. Six releases in nine days:

Date Release Why it matters
5 Oct OpenAI visual ads in ChatGPT Native ad units alongside image generation. First US advertisers testing this month.
7 Oct GPT 6 and Intelligent UI for paid tiers Answers now contain interactive tools, forms, charts and mini apps.
8 Oct GPT 6 extends to Free and Go tiers About 1.2 billion weekly users now on Intelligent UI.
13 Oct Gemini 4 Argon rolls out in AI Mode AI Pro and Ultra get it worldwide. More direct answers, less hedging.
13 Oct Bing Topics Portfolio beta The first free report naming AI topic clusters where competitors show and other sites don’t.
15 Oct AI Mode and AI Overviews get more inline links, source blocks, article carousels and creator profiles Google says preferred source usage grew 7x this year.
16 Oct Gemini 4 Argon reaches Google Antigravity AI Pro subscribers only. Agentic coding workflows get the latest model.

Intelligent UI matters most

For most of ChatGPT’s history the output has been text. GPT 6’s Intelligent UI changes that. The model now picks the format. Plain text, buttons, forms, calculators, labelled diagrams, cooking timelines, playable games. Ask for a dinner for twelve and a quantity calculator appears. Ask about splitting a bill and a bill splitter builds itself in the conversation.

OpenAI also claims GPT 6 Instant starts answering web search questions 44% sooner than GPT 5.6 Instant on average, with better instruction following and less hedging. The Chat tab moved to GPT 6 Sol (paid) and GPT 6 Luna (free) across 7 and 8 October. Work and Codex are unchanged for now.

The SEO problem Intelligent UI creates

When the answer is a built in tool rather than a sentence with a citation, less surface area remains for a link to appear. Expect a two to three month lag before anyone has solid data on how much this reduces citations. Any brand actively tracking AI visibility in ChatGPT should assume the citation to tool ratio shifts against sites, not in favour of them.

Bing Topics Portfolio is the sleeper hit

For practical value, the quietest release of the fortnight was Microsoft’s. Bing Topics Portfolio sits inside Bing Webmaster Tools’ AI performance reports. It groups the grounding queries where a site appears into thematic clusters. Each topic shows:

  • Topic Size: total citations across all sites for that topic.
  • Sites Cited: how many sites appear in that topic cluster.
  • Queries: how many distinct AI queries sit within the topic.

The killer feature: Top Opportunity Queries. A list of strong AI queries where competitors appear and a given site does not. One free report, one claim, one Q4 content brief half written.

What Google added to AI Mode on 15 October

Google rolled a fresh set of changes into AI Mode and AI Overviews:

  • More inline links. Links appear inside words and phrases within AI responses, and higher up than before.
  • Source blocks. When only a few official sites answer a query, a dedicated section at the top lists them with a short explanation.
  • Article carousels. Shown in more locations, more languages and for a wider range of queries. Preferred sources get the spotlight.
  • Related articles with thumbnails. Images and card style layouts aim to feel more clickable.
  • Creator profile links in AI Mode videos. Clicking a creator’s name sends traffic to a Google Search profile page, not the creator’s own site.

The headline figure: Google reports that preferred source usage grew 7x this year. The caveat: no click data has been published, and Barry Schwartz suspects the overall effect on publisher clicks is small.

5. The measurement layer broke twice this fortnight

Two separate stories, same conclusion. Trusting anything outside first party data right now is risky.

Nozzle: 8% of SERPs served to rank trackers were fake

Nozzle published research on 10 October analysing 17.8 million SERPs since 13 September. The headline numbers:

Metric Baseline 13 Sep onwards
Share of SERPs poisoned 0.17% 8.0% average, peaking at 27% on 29 Sep
Organic results kept from previous SERP 88.5% 20.7%
Tracked site falls out of top 10 2.7% 58.6%

In short, Google appears to be serving fake pages to rank trackers rather than refusing their requests. Social sites and forums fill the gap. Reddit visibility rose by 8.0 points on poisoned pages. 77.9% of poisoned SERPs returned to normal within three days. Nozzle credits Google’s broader fight against AI scraping, with rank trackers caught in the collateral damage.

OpenAI’s UTM bug misclassified ChatGPT traffic as direct from 8 to 15 October

Between 8 and 15 October, ChatGPT stopped adding its utm_source=chatgpt.com tag to outbound links. Because ChatGPT also normally strips the HTTP referrer, GA4 and Adobe recorded those visits as direct traffic.

OpenAI acknowledged the error on 14 October and finished the fix roughly 24 hours later. Tagged links are now back to about 95 to 100% of normal levels.

Adobe’s Cedric Huesler analysed 500 pages where ChatGPT normally drives at least 5% of visits. Reported ChatGPT traffic fell by around 31,000 visits per day. Direct traffic rose by around 26,000. Two in every three pages that lost ChatGPT traffic also saw direct traffic rise.

The practical fix: mark 8 to 15 October as a known data glitch in GA4 or Adobe Analytics. Do not change attribution models or channel rules based on that week. Keep UTM parameters on anything AI driven.

6. Google Discover is becoming an “AI concierge,” plus the quick hits

On 12 October, Search Engine Land, PPC Land and Search Engine Journal all picked up a Google job listing for a Principal UX lead on Discover. Salary: up to $382,000. The brief: turn Discover into a proactive, hyper personalised concierge with tighter AI integration. The signal: Discover is being rebuilt as the feed side entry point to AI Mode, in the same way Search was rebuilt as the query side one.

Any publisher relying heavily on Discover traffic should watch this more carefully than any upcoming core update. Previous Discover rebuilds have been brutal on traffic. A Follow button quietly appeared in Discover’s video content viewer on 16 October, which hints at the direction of travel.

Quick hits (dated)

Date Story Action
5 Oct Google appeals the US search monopoly ruling No near term SEO impact. Watch Q1 2027.
5 Oct Dark mode confirmed for Google Search Console Cosmetic. Enjoy.
7 Oct Review rich results disappearing for healthcare pages YMYL health sites should re check schema and expect lower CTR.
8 Oct UGC Fresh Data Program opens applications Only relevant to UGC platforms. Paywall and login content is out.
10 to 12 Oct GSC Performance report lagged about 60 hours Longest delay since Dec 2025. Back to normal by 12 Oct.
13 Oct Google: site migrations can run beyond 4 months Factor into migration planning. Old URLs may still rank for months.
16 Oct Google Ads: online gambling category certification rollout starts Affected advertisers should begin applications.
16 Oct Google Ads: lead forms support lead scoring (up to 12 qualifying answers per question) Review lead form questions for high intent filters.
16 Oct Google Ads API moves to twice monthly releases Factor into dev team sprint planning.
16 Oct Google Dataset Search outage (restored same day) No lasting effect reported.
16 to 19 Oct Fresh ranking volatility, unconfirmed update rumours Check Search Console. Do not act on third party data alone.

What To Actually Do

A short list for the next ten working days, in rough priority order:

  1. Date impact analysis from 24 September to 8 October. Pull clicks, impressions and average position by page type for that window. Compare against the previous four weeks. Pages that lost traffic across both waves are the real spam signal casualties. Pages that dropped in wave one and recovered are noise.
  2. Mark 8 to 15 October as a ChatGPT data glitch in GA4 and Adobe Analytics. Any apparent ChatGPT traffic drop that week was the UTM bug. Direct traffic rose by roughly the same amount it fell. Do not change attribution models based on it.
  3. Run a fake author and AI hallucination audit on the top 20 pages. For each: is the author real and verifiable? If AI assisted, is a documented human review step on record? Flag any page that fails either test.
  4. Claim the Bing Topics Portfolio beta inside Bing Webmaster Tools. Pull the Top Opportunity Queries list. That is a Q4 AI visibility content brief, half written.
  5. Set recovery expectations using Google’s own numbers. 3 to 6 months typical for core update recovery, 6 months to a year worst case. 1 to 2 weeks for manual action removal. Up to 14 days for sitemap processing. All on record now.
  6. Add a GPT 6 citation check to the AI visibility tracker. Run the fixed prompt set after 8 October and compare citation share to the pre GPT 6 baseline. Falling citations alongside rising tool based answers signal what happened.
  7. Do not trust rank tracking data alone for the next 3 to 4 weeks. Nozzle’s research puts 8% of SERPs served to trackers as fake. Cross check against Search Console before any ranking based decision.
  8. Diversify away from Discover. The move to an AI concierge model is coming, and prior rebuilds have hurt traffic badly.
  9. Watch the Search Status Dashboard this week. If Google confirms the 16 October volatility as an update, impact analysis should run from that date, not the end of the spam update.

The fortnight’s summary in one line: Google finished the harshest spam update of the year, raised the bar on authorship and AI content on paper, broke the measurement layer twice, and watched OpenAI and its own Gemini team rewire what an answer looks like.

Next round up in two weeks. Comments welcome on recovery playbooks for the spam update, GPT 6’s effect on AI citations, or how to use the Bing Topics Portfolio properly.

Sources: Google Search Central (@googlesearchc, 8 October); Google Search Status Dashboard, checked 19 October 2026; Search Engine Roundtable (Barry Schwartz, 8, 13 and 16 October); Search Engine Land (crawl timing ranges reporting; AI Mode and AI Overviews changes, 15 October; Bing Topics Portfolio, 13 October; ChatGPT UTM story, 16 October); Search Engine Journal (Google timing ranges, 4 October); Search Engine Watch (unconfirmed October update, 16 to 19 October); 9to5Google (Gemini 4 Argon, 13 October); Unite.AI (GPT 6 Intelligent UI, 7 and 8 October); TechCrunch (OpenAI visual ads, 5 October); PPC Land (fake author profiles; Google Discover job listing); SearchHerald (October 2026 archive); DesignRush SEO Roundup (8 October week); Distribb (September 2026 spam update retrospective); AEOSoup (helpful content guidance updates); WebmasterWorld (September and October 2026 observations threads); Numinix (October 2026 SEO news); Nozzle (Poisoned Google SERPs research, 10 October); Marketing4eCommerce (OpenAI UTM bug analysis).

SEO & Digital Marketing News Round-Up Late September to Early October 2026

On By DaveIn seoLeave a comment

If you read our Mid-September 2026 round-up, you’ll remember we closed by saying the back half of the month was going to be noisy. It was noisier than expected. Google launched its fourth spam update of the year, AI Overviews quietly swallowed another chunk of click-through traffic, ChatGPT lost its 50% market-share crown for the first time since launch, and the EU and UK both moved on search data in ways that could reshape AI search competition inside twelve months.

Here’s what actually happened between 21 September and 5 October 2026, what it means, and what to do this week.

1. The headline event: Google September 2026 Spam Update

Google began rolling out the September 2026 Spam Update at 09:15 PDT on 24 September. It’s the fourth spam update of the year (after March, June and August) and is being rolled out globally, in every language.

Two things make this one different:

  • The two-week window. Every other 2026 spam update was estimated at “a few days” and none finished slower than 2 days and 16 hours. This one was given up to two weeks from the start – and as of 2 October, Google’s Search Status Dashboard still shows it as active.
  • It arrived in phases. The first wave of volatility hit between 25–27 September. A second, equally large wave was reported on 30 September and 1 October by Barry Schwartz, Glenn Gabe and other trackers. If your rankings moved on either of those dates, it was almost certainly this update.

How this update compares to the rest of 2026

Spam Update Start date Rollout duration
March 2026 24 March 19 hours 30 minutes
June 2026 24 June 2 days 1 hour
August 2026 18 August 2 days 16 hours
September 2026 24 September Ongoing (expected ~8 October)

What to do: Don’t finalise an impact analysis until Google confirms the rollout is complete. Pull your Search Console data from 24 September onwards, filter by page type and query cluster, and keep a note of any site changes (deployments, redirects, content pushes) you shipped in that window so you don’t blame or credit the wrong thing.

2. AI search is now genuinely eating Google – and ChatGPT

Three numbers landed in the last fortnight that I’d pin to the wall if I were running an in-house SEO team:

  • Google AI Overviews now appear on 80% of branded queries, up from 26% earlier in the year.
  • Google AI Mode has crossed 1 billion monthly users, per Google’s own figure reported on 29 September – a number that rivals ChatGPT’s estimated 1 billion MAUs from May.
  • ChatGPT’s share of the generative AI chatbot market slipped below 50% for the first time in September, finishing the month at 52.1%, with Gemini climbing to 26.9%.

ChatGPT market share through 2026

Monthly share of the generative AI chatbot market. ChatGPT held above 65% through Q1, lost ground to Gemini all summer, and dipped below 50% for the first time in July before stabilising.

Jan 26
67.1%

Feb 26
65.7%

Mar 26
62.6%

Apr 26
61.8%

May 26
59.1%

Jun 26
58.6%

Jul 26
51.3%

Aug 26
51.5%

Sep 26
52.1%

Source: First Page Sage / Sensor Tower, October 2026 report.

Where that lost market share went

AI assistant Market share (Sep 2026) Quarterly user growth
ChatGPT 52.1% +5%
Google Gemini 26.9% +10% (approx.)
Microsoft Copilot 1.32% Highest reading of 2026
Perplexity 1.9% Down from 5.3% in Jan
Claude (not listed in top table) +15% (fastest growing)

The pattern is clear: ChatGPT is still the market leader by a comfortable margin, but the “AI = ChatGPT” era is over. Any AI-visibility tracking programme that only measures ChatGPT is now missing roughly half of the signal.

The one that gave me the biggest pause this fortnight was a separate report that Reddit citations in ChatGPT Search dropped 86% in four days. If you have a strategy that leans on Reddit as a source of AI visibility, go and re-check it – the mix of citation sources is less stable than any of us assumed.

3. Google Search Console: three real changes in two weeks

Search Console had more product news in September than it had in most of H1. Three things worth knowing:

a) Generative AI performance report now worldwide

Google’s Generative AI report in Search Console – showing impressions and clicks from AI Overviews and AI Mode – rolled out globally. If you don’t see it yet, it should appear in the next few days. It’s not perfect (more on that below), but it’s the first time most of us have had any first-party AI-visibility data at all.

b) New multimodal filter for image-based searches

A new filter – sitting under the Web search type – now splits out performance for searches where someone used an image (think Google Lens, Circle to Search). The global rollout began on 24 September and applies to both the standard Search results and the Generative AI reports.

c) Google is fixing a long-running impressions bug

Google has confirmed a logging error that inflated impression counts since 13 May 2025. The fix is now rolling out. Clicks are unaffected. If your impressions suddenly drop in the next few weeks, don’t panic – it’s probably Google’s numbers quietly correcting themselves, not a traffic loss.

A caveat on the AI reporting

An important caveat on this new data: Search Console treats an entire AI Overview as a single block and assigns the same ranking position to every link inside it, regardless of where the link actually appears in the answer. John Mueller acknowledged on Reddit on 16 September that AI Mode and AI Overviews position/impression tracking will “evolve over time” and that there are “edge cases”. Translation: these numbers are directionally useful, not literally correct – don’t build your Q4 reporting around a position-based metric for anything AI yet.

4. The AI Overviews / AI Mode drift continues

The product story keeps moving in one direction: AI Overviews behaving more like AI Mode, and both of them keeping users inside Google rather than sending them to publishers.

  • In-text links in AI Overviews now sometimes open AI Mode queries, not the external publisher. Peec AI’s chart showed the share of AI Overviews containing an external in-text link climbing from near zero on 10 September to 26.2% by 20 September. Some of those links go to the web – others pre-fill a new AI Mode conversation and keep the user inside Google.
  • AI Mode’s click-out rate is now around 4.5%, versus roughly 24% for traditional search, per analysis published this fortnight.
  • Gemini 3.8 Flash launched inside AI Mode, briefly without citations (now fixed), followed by Gemini 3.8 Live.
  • Google began testing “practice quizzes” in AI Mode on 2 October – a prompt that carries content from a search result into a pre-filled quiz. Another reason for the user to stay in AI Mode rather than click out.
  • Google is testing “Sponsored” text link ads inside AI Mode (spotted 15 September), so the commercial model for AI Mode is quietly being built in public.

Growth of external in-text links in AI Overviews

Daily share of AI Overviews containing an in-text external link, 25 August – 20 September 2026. Source: Peec AI.

25 Aug
~0%

5 Sep
~1%

10 Sep
~3%

15 Sep
~14%

20 Sep
26.2%

Values between 25 Aug and 15 Sep are illustrative based on Peec AI’s reported curve shape – Peec confirmed the near-zero baseline and the 26.2% peak on 20 September.

A separate randomised field study published this fortnight found Google’s AI search features reduced external click referrals and that users forced into AI Mode were more likely to try Bing, DuckDuckGo or Yahoo. Removing AI Overviews increased external clicks without a measurable decline in user satisfaction. That’s a difficult data point for Google to carry around.

5. Regulators actually moved

Two regulatory stories that matter more than they look:

  • The EU has ordered Google to share search data with rival search engines – and explicitly with AI-powered search tools and chatbots. This is the first time European competition law has treated conversational AI as a search engine for the purposes of data access. Expect challengers (Perplexity, You.com, Mistral-based tools and others) to put in requests before year-end.
  • The UK CMA’s consultation on search choice screens closes on 9 October. The proposals would show Android and Chrome users a choice screen at setup and annually, and AI assistants such as ChatGPT and Perplexity would be eligible for listing. The CMA also wants to cover fair attribution of publisher content. A final decision is expected by the end of 2026.

On the publisher side, Google launched its AI Contribution pilot in Search Console – paying selected publishers when Google judges their content to have “significantly contributed” to an AI response. The payouts are reportedly small. The precedent is not.

6. Quick hits: everything else worth knowing

Date Story Why it matters
22 Sep Google AI Mode vs AI Overviews explainer wave Industry is finally treating them as two different surfaces to optimise for.
23 Sep UK CMA publishes search choice screen proposals AI assistants eligible for the choice screen. Consultation closes 9 October.
24 Sep September 2026 Spam Update begins Fourth of the year. Still rolling as of 2 October.
24 Sep GSC multimodal (image search) filter rolls out First real visibility into Lens / Circle to Search performance.
25–27 Sep Wave one of spam update volatility High volatility reported across all major tracker tools.
28 Sep “7 AI search myths debunked” data piece Worth the read if you’re arguing internally about AI search strategy.
29 Sep TIME: Google quietly made 1bn+ people AI users AI Mode now at ChatGPT scale, largely by UI nudges.
30 Sep Yext acquires Flamel.ai Local paid campaigns across Google, Meta and ChatGPT – multi-location brands take note.
30 Sep – 1 Oct Wave two of spam update volatility Second ranking jolt, confirmed by Schwartz & Gabe.
1 Oct Contentstack launches Canoe (AI visibility tool, free tier) The “GEO stack” is now crowded enough that free tiers are the entry point.
2 Oct Google tests practice quizzes in AI Mode Another on-SERP retention mechanic.
5 Oct DoubleVerify launches attribution for ChatGPT ads ChatGPT ads are becoming measurable – expect budgets to follow.

7. What to do this week

A short, honest to-do list for the next seven days:

  1. Date-stamp the spam update impact. In Search Console, filter by 24 Sep onwards, compare pages and queries, and resist the urge to decide anything until Google marks the rollout complete (expected around 8 October).
  2. Note the GSC impressions correction. Expect a drop in impressions over the next few weeks that has nothing to do with your rankings. Tell your stakeholders now.
  3. Build a fixed AI-visibility prompt set. 20–30 queries, run across ChatGPT, Gemini, Claude and Perplexity. Not just ChatGPT. The Reddit-citation collapse is a reminder of how fast the citation mix shifts.
  4. Audit your in-SERP presence, not just your rankings. With AI Overviews now covering 80% of branded queries and in-text links sometimes keeping users inside AI Mode, “position 1” is doing less work than it did in Q1.
  5. If you’re in the UK, respond to the CMA consultation before 9 October. If you publish anything, the attribution rules matter directly to you.

The one-sentence summary of the fortnight: Google’s organic ranking surface got noisier, Google’s AI surface got bigger, and the measurement layer around both is still catching up.

We’ll be back with the next round-up in two weeks. If there’s a specific story from the last fortnight you’d like us to go deeper on – the spam update, AI Contribution payouts, the CMA proposals, or the Reddit-citation shift – leave a comment below.

Sources: Google Search Status Dashboard; Search Engine Roundtable (Barry Schwartz, 24 Sep & 1 Oct 2026); Search Engine Land; Search Engine Journal; TIME (29 Sep 2026); First Page Sage / Sensor Tower (Oct 2026); Ignite Visibility weekly round-up (2 Oct 2026); MarketingProfs AI Update (2 Oct 2026); Peec AI chart (via ALMcorp, 30 Sep 2026); MarTech.org (1 Oct 2026); AgileBrandGuide (1–2 Oct 2026).

How to Make AI Posters for Business & Events – that aren’t obvs AI Posters

On By DaveIn UncategorizedLeave a comment

TL;DR

  • Adverts look “AI” when the AI makes every decision. Let it make the picture; you make the choices.
  • Brief first: who it’s for, the one message, where it’ll be seen, how it should feel.
  • Prompt like a photographer: real setting, light, lens, film stock and empty space. Drop “stunning 8k masterpiece”.
  • Fix the tells: hands, gibberish text, waxy skin, glow, melting edges, dead-centre framing.
  • Never let AI write your words. Set the type yourself, two typefaces maximum.
  • Keep the real product in shot, crop off-centre, tone the colour down, add grain, then squint-test it at phone size.
  • Event posters: the date and details are the design, make 4:5 and 9:16 versions, use one real idea instead of a silhouette crowd, and build a reusable template.
  • Short on time? Jump to the hack near the end for copy-and-paste prompts.

AI can produce a poster in seconds, but most people can spot one just as fast. The glossy skin, the orange-and-teal glow, the gibberish lettering and the perfectly centred everything give it away before anyone reads a word.

The fix isn’t a magic prompt. It’s treating the AI as one tool in a normal design process: you brief it, you direct it, then you finish the work by hand. Follow the seven steps below and your posters and adverts will read as designed, not generated. There’s a finished example poster at the end.

Step 1: Start with a brief, not a prompt

The biggest giveaway of an AI advert is that it isn’t about anything. Someone typed “cool poster for a bakery” and accepted the first result, so the image is a generic average of every bakery picture ever made.

Before you open any image tool, answer four questions in plain words: who it’s for, the single message, where it will be seen, and how it should feel. Collect two or three real photos or adverts you admire as references.

Step 1: answer four questions before you promptWho is it for?e.g. first-time buyers, 25 to 35What is the one message?e.g. fresh bread, baked at 6amWhere will it be seen?e.g. A3 shop window, Instagram 4:5What should it feel like?e.g. 3 real reference imagesThen write the promptA specific brief gives a specific image.A vague one gives the AI average.
Step 1: the four-question brief

The format question matters more than people think. A shop-window A3 poster, a 4:5 Instagram post and a 16:9 web banner need different compositions, so decide the shape first and generate at that ratio.

Step 2: Prompt like a photographer, not a fan

Words like “stunning”, “8k”, “hyper-realistic” and “masterpiece” are the fastest route to the AI look. They push the model towards over-sharpened, over-lit, over-saturated images.

Describe the shot the way a photographer or art director would brief a crew: the subject in a real setting, the light, the lens, the film stock and where the empty space should go.

Ask for empty space on purpose. Real adverts leave room for a headline and a logo, and AI images tend to fill every corner. Generate several variations and pick the one that looks least “perfect”, not the most impressive.

Step 3: Hunt down the AI tells

Viewers don’t consciously analyse an image; they notice that something feels off. Almost always it’s one of six details, so check each one before you go any further.

Most modern tools let you select a small area and regenerate just that part (often called inpainting or generative fill). Use it for hands, edges and background signs. If a detail still won’t come right, crop it out; a tighter crop usually looks more intentional anyway.

Step 4: Never let the AI do your typography

Even the best image models still wobble on letters, and they love a big centred headline. That one habit gives away more AI adverts than anything else.

Generate the image with no text at all (add “no text, no lettering, no logos” to your prompt), then set the words yourself in Canva, Figma, Affinity, Photoshop or whatever you use.

Step 4: generate the picture, set the words yourselfAI letteringwarped, misspelt, centredremove text, re-set itType set by handBakedat 6am.Real sourdough, every morning.YOUR LOGO
Step 4: AI lettering versus type set by hand

Use no more than two typefaces, align text to an edge rather than the centre, and keep the headline short. Your real logo, a price and a web address in a sensible corner do a lot to make a poster feel like it came from an actual business.

Step 5: Composite, crop and edit by hand

The most convincing “AI” adverts are only partly AI. Photograph your real product on your phone, cut it out, and place it into an AI-generated scene. Customers trust the thing they’re buying looking exactly like the thing they’ll receive.

Then crop. AI tools centre everything, so move your subject onto a thirds line, trim the edges and let one element bleed off the frame. When you composite, match the direction of light and add a soft shadow under the product so it sits in the scene instead of floating on it.

Step 6: Add real-world texture and a consistent grade

AI images are too clean: perfect gradients, no noise, everything sharp and bright at once. Real photos and printed posters have grain, softness and a colour “personality”.

A few small moves in any photo editor (Lightroom, Snapseed, Photoshop or your phone’s built-in editor) remove most of that sheen.

Go gently: the aim is “nobody notices”, not a heavy Instagram filter. You can also overlay a subtle paper or print texture on posters, which makes them feel physical even on screen. Using the same grade and your brand colours on every advert makes the set look like a campaign, and campaigns look human-made.

Step 7: Do the squint test, then run the checklist

Shrink your advert to the size it will appear in a phone feed, then squint or blur it. A good advert still shows one clear headline and one clear subject. AI images often turn to busy mush at this point because they’re full of detail nobody asked for.

Finally, show it to someone who hasn’t seen it and ask what it’s advertising, not whether it looks like AI. If they answer quickly, you’re done. Before you publish, tick off the list:

  • Made from a written brief, at the right size and ratio
  • Prompt describes light, lens and setting, with no “stunning 8k” filler
  • Hands, edges, skin and background text checked and fixed
  • All words set by hand, no more than two typefaces, not centred by default
  • Real product and real logo in the image
  • Cropped off-centre with deliberate empty space
  • Saturation and sharpness toned down, fine grain added
  • Passes the squint test at phone size

The finished example

Here’s what all seven steps look like together, on a poster for a (fictional) neighbourhood bakery.

Notice what it does: a single message (“Baked at 6am.”), two typefaces set by hand and aligned left, a real photo of real bread fading up into the dark so the headline has room, a restrained palette, and fine grain over the top. The practical details (price, hours, address) sit where a real shop would put them. Here the loaf is a stock photo; in your version, step 5 would swap in a photo of your own bread.

Bonus: event posters for social media

Scroll any local events page and you’ll see the same poster over and over: a glowing title, a crowd of silhouettes with their hands up, light beams and a purple-orange sky. Quiz nights, craft fairs and gigs all end up looking identical, so none of them stands out.

The steps above still apply, but event posters have their own traps. This is the look to avoid.

The real problem is point 6. A good event poster could only be for your event: your venue, your people, your date. Everything below is about getting that back.

Event step 1: Make the information the design

Someone scrolling gives your poster about a second. In that second they need four answers, in this order: what it is, when, where (and how much), and what to do next. AI posters get this backwards, spending all the space on atmosphere and squeezing the facts into a strip at the bottom.

Write those four lines first, then design around them. On a type-led poster, the date or the event name can be the image.

Event step 2: Design for the feed, not a noticeboard

Make two versions from the start: a 4:5 feed post (1080 x 1350 px) and a 9:16 story (1080 x 1920 px). In stories, the account name at the top and the reply bar at the bottom cover roughly the first and last 250 px, so keep your text in the middle.

Check it at actual phone size. If the date isn’t readable at thumbnail size, make it bigger and cut something else.

Event step 3: Use one real idea instead of a generic crowd

Pick a single idea that could only belong to this event. A photo of the actual venue, last year’s crowd, the band’s own press shot, or one object that sums it up (a pint glass for a pub quiz, a record for a vinyl fair). If you have none of those, go type-led: huge date, bold colour, nothing else.

If you do use AI, use it for a background texture or a single prop, then follow steps 3 to 6 above. Ban the words “epic”, “neon”, “cinematic” and “crowd” from your prompts.

Event step 4: Build a template, not a one-off

Regular events (monthly quiz, weekly open mic, a festival line-up) should share one layout: same grid, same typefaces, same logo position. Change only the colour, the date and the image. After three or four posts, people recognise yours before they read it, which is exactly what the identical AI posters can’t do.

Event step 5: Let the caption and alt text do the work too

Repeat the key details in the caption: what, when, where, price and how to book. Add alt text describing the poster (on Instagram it’s under Advanced settings when you post). It helps people using screen readers, and it means your details are searchable text rather than pixels.

The finished example: an event poster (Looks better with a real photorgraph in the foreground)

Here’s a 4:5 feed poster for a (fictional) night market, built with the five event steps.

The date is the hero and readable from across the room. The only picture is a string of festoon lights, which says “evening market” in a way a silhouette crowd never could. All four answers (what, when, where and price, what next) sit in their own clear spot, and there’s no glow, no gradient and no lens flare. Swap the colour and the date and it becomes next month’s poster.

ai event poster example

The pitch itself is the image: mown stripes, the halfway line and the centre circle, pushed off to the right and cropped by the edge. The two club names sit either side of the halfway line, like the teams at kick-off, each with a scarf stripe in its kit colours. Date, kick-off, ground and prices sit in a clear footer, readable at phone size. For the next fixture, change the away team and its colours, and it’s done.

The hack: copy-and-paste prompts

These prompts build every step above into one go. Replace each [BRACKETED PLACEHOLDER] with your own details (examples are after “e.g.”), delete the examples, and paste the result into your image tool. Each one asks for no text, so you add your words yourself afterwards (step 4).

1. Product or advert poster

Editorial product photograph of [PRODUCT, e.g. a sourdough loaf] on [REAL SURFACE, e.g. a flour-dusted oak counter] in [SETTING, e.g. a small bakery kitchen]. [TIME OF DAY, e.g. early morning] light from [DIRECTION, e.g. a window on the left], soft natural shadows. Shot on a [LENS, e.g. 50mm] lens at f/2.8, [FILM STOCK, e.g. Kodak Portra 400] colour, fine film grain. [PRODUCT] placed [POSITION, e.g. in the lower right third], with plain empty space [WHERE, e.g. across the top half] for a headline. Muted, true-to-life colours, small natural imperfections. [ASPECT RATIO, e.g. 4:5 vertical]. No text, no lettering, no logos.

2. Event poster with a real scene

Documentary-style photograph of [VENUE OR SCENE, e.g. a village hall set up for a quiz night], with [DETAILS THAT MAKE IT REAL, e.g. trestle tables, answer sheets and paper bunting]. [TIME, e.g. evening], lit by [REAL LIGHT SOURCE, e.g. warm ceiling lights], shot on a 35mm lens, natural colour, visible film grain. Wide shot with [EMPTY AREA, e.g. the top third] kept plain and uncluttered for text. [ASPECT RATIO, e.g. 4:5 vertical]. No text, no signs, no logos, no crowd silhouettes, no lens flare, no light beams.

3. Type-led event poster background

Flat, straight-on photograph of [TEXTURE, e.g. recycled card / painted brick wall / worn felt] in [COLOUR, e.g. deep forest green], even soft lighting, no shadows, subtle natural texture and small imperfections. Plain and uncluttered, to be used as a poster background. [ASPECT RATIO, e.g. 9:16 vertical]. No text, no objects, no gradient, no glow.

4. Sports match poster

Top-down aerial photograph of a [SPORT, e.g. football] pitch at [GROUND TYPE, e.g. a small non-league ground], freshly mown [STRIPE DIRECTION, e.g. vertical] stripes, crisp white line markings. [LIGHT, e.g. flat overcast afternoon light]. The [PITCH FEATURE, e.g. centre circle] sits [POSITION, e.g. right of centre, partly cropped by the edge]. Realistic grass texture, muted natural greens, no players, no crowd. [ASPECT RATIO, e.g. 4:5 vertical]. No text, no logos, no stadium lights, no lens flare.

5. Fixing one detail (inpainting)

Select just the problem area in your tool’s edit or generative-fill mode, then use:

Regenerate only the selected area: [WHAT IT SHOULD BE, e.g. a hand holding a coffee cup] with [CORRECT DETAIL, e.g. five fingers and a relaxed grip]. Match the existing light direction, colour and film grain exactly. No text.

Generate four to six versions of each, pick the least “perfect” one, then carry on from step 3.

The takeaway: use AI for the draft, not the decisions

Adverts look “AI” when the AI made every decision. Brief it properly, direct it like a photographer, fix the tells, set your own type, composite in the real product and finish it by hand. The result is simply good design that happened to use a fast tool.

A quick note on honesty

Making an advert look polished is fine; misleading people isn’t. Don’t use AI to show a product doing something it can’t, invent customers or testimonials, or create fake photos of real people. Some platforms, such as Meta and TikTok, ask you to label realistic AI-generated content, and UK advertising rules still require ads to be accurate. Check the rules for wherever you’re posting, and when in doubt, disclose

SEO, PPC & Digital Marketing News: The Only Things Worth Acting On This Week -Late September 2026

On By DaveIn seo1 Comment

SEO, PPC & Digital Marketing News: The Only Things Worth Acting On This Week (Late September 2026)

Another fortnight, another spam update, and a batch of platform changes that are easy to miss while you’re getting on with actual work. Here’s what genuinely changed, why it matters, and what to do about it before Monday morning.

Photo of a laptop, notebook and mug on a wooden desk, morning light.
Photo by Unsplash.

TL;DR: The Practical Takeaways

Skim this table. If a row looks relevant, jump to the section.

# The Change The Practical Takeaway Priority
1 Google September 2026 spam update rolling out (started 24 Sept) SEO: Don’t touch anything for 2 weeks. Wait for the rollout to complete around 8 Oct before diagnosing traffic changes. 🔴 Critical
2 Search Console multimodal filter launched (24 Sept) SEO: Open Search Console this weekend. Set Search Type, then Web: multimodal, to see Google Lens and Circle to Search traffic you couldn’t see before. Especially useful for anyone with product photos. 🟠 High
3 ChatGPT Ads: Sponsored Agents, Shopify app, 5-way platform targeting (16 Sept, global 23 Sept) PPC: If you’re already testing ChatGPT Ads, switch to oCPM bidding and set separate platform targeting. Shopify merchants should install the ChatGPT Ads app for direct feed integration. 🟠 High
4 Chartbeat: publisher search traffic down 40.2% YoY (reported 24 Sept) Digital marketing: Diversify. Any business plan that assumes organic traffic will grow needs rewriting. Direct traffic, email lists and branded search are where the survivors are landing. 🟠 High
5 Google research paper on R4T-Diffusion query fan-out (15 Sept) SEO: No immediate action. Long-term signal that AI answers will keep pulling from more sub-queries per prompt. Write pages that answer clusters of related questions, not single keywords. 🟡 Medium
6 Google Business Profile spam review alerts now sending Local SEO: Nothing to do proactively. Watch your inbox for GBP notifications about spam review spikes so you can flag them fast. 🟡 Medium

1. Google’s September 2026 Spam Update: Sit on Your Hands for Two Weeks

Google confirmed the September 2026 spam update on Wednesday 24 September at 09:15 Pacific. It applies globally, in every language, and Google says the rollout may take up to two weeks. That means the earliest it finishes is around 8 October.

Two things stand out about this one.

First, the length of the stated window. The three previous 2026 spam updates each ran a few days at most. March finished in 19 hours and 30 minutes. This one has been given a fortnight. That is the same window Google typically hands to its core updates, which is worth noting even if the update itself is not a core one.

Second, the pattern. This is the fourth spam update in 2026, following March, June and August. Google runs these to improve its automated spam detection systems, not to introduce new policies. Nothing has been added to the rulebook.

What to do

  • Nothing dramatic. Day one to fourteen of a spam update is for diagnosis, not for changes.
  • Note the start date (24 Sept) somewhere you’ll remember. When you look at Search Console in October, line up dips or spikes with it.
  • If you were hit by the August spam update, the September one is not a second chance to recover. Recovery from spam updates is measured in months, sometimes multiple update cycles.
  • Don’t panic-rewrite AI-assisted content unless you already had real reasons to. Independent analysis of the August update showed the sites hit hardest were the ones publishing mass-generated content, not the ones using AI as a writing aid.

2. Search Console’s Multimodal Filter: A Weekend Job Worth Doing

On the same day the spam update landed, Google quietly rolled out something more useful: a new filter in Search Console’s Performance report called Web: multimodal.

It shows how your pages perform when someone searches using an image instead of typed text. That covers Google Lens, Circle to Search on Android, image uploads to Google Search, and Chrome’s right-click “Search this image”. All four are grouped together, so you can’t split them out, but this is data that was completely invisible before.

One catch: the Queries tab doesn’t work with multimodal. You see clicks, impressions, pages, countries and devices, but not what image or what text the searcher used. Which makes sense; there is no text query behind a Google Lens photo of a lamp.

Why it matters more than it looks

Anyone with product photography now has a measurable channel that used to be a mystery. If someone points their phone at your competitor’s product in a shop and Google Lens surfaces your page, that click has always existed. You just couldn’t count it. Now you can.

What to do (10 minutes on Sunday)

  1. Sign in to Search Console.
  2. Open Performance, then Search results.
  3. Click the Search type filter (defaults to Web).
  4. Choose Web: multimodal.
  5. Set the date range to include 10 Sept onwards. That’s the earliest date multimodal data is available.
  6. Export the pages tab as CSV. That’s your baseline.

If your top multimodal pages differ from your top text-based pages, you’ve just found which pages are quietly earning visual-search traffic. Optimise their images, alt text and structured data accordingly.

3. ChatGPT Ads Grew Up in One Week (16 Sept)

OpenAI shipped a lot on 16 September, and then rolled the biggest piece globally on 23 September. If you’re advertising on ChatGPT, or planning to, this is where the platform actually became viable.

Update What it does Who it’s for
Sponsored Agents (in test) Users can click an ad and start a labelled conversation with an advertiser’s own AI agent High-consideration purchases, complex products, B2B
Shopify app Direct product feed integration for ChatGPT Ads. US on 16 Sept, rest of the world 23 Sept Any Shopify merchant
HubSpot integration Build and manage ChatGPT campaigns from inside HubSpot HubSpot users doing paid
Ads Manager plugin Create, update and analyse campaigns through natural-language prompts in ChatGPT Work Agencies, in-house teams
oCPM billing Optimises for conversions from both clicks and views, billed on impressions. Now generally available and OpenAI’s recommended bid model Most performance advertisers
5-way platform targeting Separate targeting for Android app, Android web, desktop web, iOS app, iOS web Anyone whose funnel behaves differently by device
Adjustable attribution windows 7, 14 or 30-day click windows; 0 or 1-day view window Sensible measurement setups

The practical takeaway

If you’re already spending on ChatGPT Ads, three changes matter this week. Switch to oCPM unless you have a specific reason to use CPC. Set separate platform targeting if your mobile web experience differs from your app. And if you’re a Shopify merchant, install the app so your feed pipes through directly.

If you’re not spending yet, this is still not the time to jump in aggressively. But it is the time to install the OAIQ pixel and set up conversions, so you have baseline data ready when you do test.

4. The Publisher Traffic Data That Should Change How You Plan

Chartbeat published its 2026 Publisher Playbook, and the numbers are worth staring at for a minute. Reported on 24 September:

  • Google Search traffic to publishers fell 40.2% year-on-year between July 2025 and July 2026 across Chartbeat’s network. The year before, the drop was 21.9%. The decline is accelerating, not slowing.
  • Google Discover traffic fell 34.3% in the same period. Which matters because Discover was the channel a lot of publishers hoped would offset search losses.
  • AI referrals grew 8x in two years, but are still only 0.01% of publisher traffic. In other words, they are not filling the gap. Not yet, probably not soon.
  • The smaller the site, the harder the hit:
Chart showing publisher traffic decline by site size. Small sites down 60%, medium down 47%, large down 22%.
Data: Chartbeat 2026 Publisher Playbook (reported 24 September 2026).

What to actually do about it

For most small businesses (not publishers), the specific figures matter less than the direction. Any strategy that assumes Google organic will grow next year needs a plan B.

  • Build a mailing list, even if it feels boring. Direct traffic and email are the categories that held up.
  • Focus content on things that can’t be summarised in two sentences. First-hand data, local expertise, genuinely proprietary knowledge.
  • Treat branded search as a KPI. When someone Googles your business by name, AI Overviews rarely intercept the click, because there’s nothing to summarise.

5. Google’s R4T-Diffusion: The Research Paper Worth a Ten-Minute Read

On 15 September Google Research published a blog post explaining a new query fan-out framework called R4T-Diffusion, or Retrieve-for-Train. The details are technical. The implication is straightforward.

Current AI search generates sub-queries one at a time from the model. The paper describes generating them in parallel via a diffusion retriever, 12x to 20x faster, and cheaper. Google describes it as “production-ready” for scale.

Why it matters for SEO

If Google can generate more sub-queries per user prompt without the cost blowing up, AI Overviews and AI Mode will pull from a wider set of pages per answer. That rewards content that covers a cluster of related questions rather than a single narrow keyword. Which is what Google has been saying for two years, but now the infrastructure catches up with the advice.

Nothing to do today. But when you’re planning your next content refresh, ask: does this page answer the obvious question, the follow-up question, the objection, and the comparison? If yes, it’s more likely to survive.

Quick-Fire Round-Up

Update What Happened Practical Takeaway
Google testing new sitelinks across organic and ads (22 Sept) Sitelink design tests expanding beyond sponsored-only PPC + SEO: Watch your CTR data. Nothing to change yet
Google testing AI Overviews linking to AI Mode Links inside AIOs now sometimes point to Google’s own AI Mode SEO: Another route for clicks to stay on Google. Focus on the answer, not the click
Google Business Profile spam review alerts Google now emails business owners when a spike in fake reviews is detected Local SEO: Check your GBP notification settings are on
Google Business Profile “Permanently closed” option missing UI change; unclear if intentional Local SEO: If a business genuinely closed and you can’t mark it, contact Google support
Google Ads: gun accessory ads allowed in US for 6 months from October Policy relaxation, US only PPC: Not relevant unless you sell in that category
Search Console crawl stats report missing 15 Sept data Second reporting glitch this month SEO: Cross-check with GA4 and server logs; the raw data is fine

Your Weekend Checklist

  • ☐ Open Search Console and pull a baseline of Web: multimodal traffic
  • ☐ Note the September 2026 spam update start date (24 Sept) alongside your traffic charts
  • ☐ Don’t touch spam-affected pages until the rollout finishes around 8 October
  • ☐ Check GBP notification settings for spam review alerts
  • ☐ If running ChatGPT Ads: switch to oCPM and check platform targeting
  • ☐ If on Shopify: install the ChatGPT Ads app if you’re testing
  • ☐ Sanity-check your 12-month traffic forecast against the 40% publisher decline data

The Bigger Picture

Two themes cut through late September. First, Google is quietly making it easier to see AI-adjacent data (the multimodal filter, the generative AI performance report, better spam update transparency) even as it makes it harder to earn a click. Visibility is going up. Traffic is going down. Both trends will continue.

Second, the platforms have stopped pretending that “test and see” is a strategy small businesses can run indefinitely. ChatGPT Ads now has proper targeting and billing. Google’s spam updates now come with core-update-length rollout windows. The kit is professional; the effort required is professional too.

None of which means the sky is falling. It just means the businesses that plan around AI-era search, with clear pages, real expertise, a direct audience relationship, and honest measurement, keep pulling ahead of the ones waiting for things to go back to normal.

Sources for this round-up include Search Engine Land, Search Engine Roundtable, Search Engine Journal, Search Engine Watch, Google Search Central Blog, Google Research, OpenAI, Chartbeat’s 2026 Publisher Playbook, and reporting from Digital Applied, PPC Land, Qwestyon, Common Thread Co. and Coalition Technologies. Snapshot as of Sunday 27 September 2026.

SEO & Digital Marketing News Round-Up: What Actually Happened in the Last Two Weeks (Mid-September 2026)

On By DaveIn Uncategorized2 Comments

The September deadlines have started to pass. Some quietly, some noisily. Here’s what actually matters for UK small businesses since the last round-up – with the least amount of panic possible.

TL;DR: The Five Things Worth Knowing

#What ChangedWhenWho It AffectsUrgency
1Google’s /goto redirect confirmed across search resultsConfirmed 26 Aug, near-full rolloutAnyone using rank tracking tools or GA4 attribution🟠 High
2AI Overviews got significantly bigger (dynamic expansion + generative UI)19 Aug and 27–28 AugEvery business with informational or how-to content🔴 Critical
3ChatGPT changed how it searches the web overnight8 Aug (still playing out)Anyone chasing AI citations🟠 High
4Parasite SEO enforcement paused inside the EEA (but not the UK)30 AugUK sites competing against EU-hosted parasite content🟡 Medium
5Google Content API for Shopping starts failingProgressive from 1 SeptEcommerce sites with old Merchant Center integrations🟠 High

1. Google’s /goto Redirect: Your Rank Tracker Might Be Lying

Hover over any Google search result now. The link no longer points to the site you’re about to visit. It goes to google.com/goto?url=... first, then bounces to the destination.

Google confirmed this on 26 August. By late August, one SERP tracking firm (Nozzle) reported the redirect on nearly 100% of results across several residential IP ranges. The change is quiet, the impact is not.

Why this matters for you (even if you don’t scrape SERPs)

There are two knock-on effects worth watching.

The first is rank tracking. Tools that used cheap HEAD requests to resolve destination URLs can’t any more; Google blocks them on /goto links, forcing a full GET request for every single result. One SEO estimated a standard five-page ranking report now takes 500 to 1,000 requests instead of the handful it used to. Which means your tracker either eats the cost or delays your data.

The second, and more important for small business owners, is GA4 attribution. If the redirect strips referrer data on a click, that organic session can quietly reclassify as (direct) / (none). Your organic channel shrinks. Direct traffic swells. Neither number is real.

Who’s affected -and who says they aren’t

ToolImpact reported
Google Search ConsoleNo impact – still reliable
SE RankingConfirmed no impact on position tracking
SemrushConfirmed no impact on position tracking
AhrefsTemporary data inconsistency, fix in progress
SerpAPIMultiple products affected; resolved 5 Sept

What to do

  • Cross-check any recent traffic dip against Search Console before blaming your site. GSC data still comes from the source.
  • If your GA4 organic traffic has fallen since late August but Search Console impressions and clicks are stable, that’s the redirect at work – not a ranking loss.
  • Ask your rank tracking provider directly whether they’re impacted. They will have an answer, or they’ll be lying to you.

2. AI Overviews Just Ate More of the Page

Two changes landed within about ten days of each other, and together they change how much space Google’s own answer occupies before a searcher ever reaches a blue link.

Change one: Generative UI arrived in AI Overviews (19 August)

Custom visual layouts, interactive tools, calculators, comparison widgets. All built by AI on the fly. Google demonstrated it with a pH scale visualisation. In practice this hits any business built around explainer tools, calculators, converters, comparison tables or interactive guides. If Google can build a working mortgage calculator inside its own answer, a searcher has one less reason to click.

Change two: Dynamic expansion (27–28 August)

AI Overviews can now grow into a longer, AI-Mode-style answer automatically. The manual “Show More” button , the one that used to give you a fighting chance below the fold -has been removed for some queries. Google decides when to expand. You don’t.

Independent tracking put AI Overview presence at around 48% of Google searches in early 2026. Health, education and B2B tech queries can hit 80%. AI Mode sessions themselves end in a click to an external website only about 4.5% of the time, compared with roughly 24% for traditional search. That is not a small delta.

What to actually do

  • Assume the top of your service pages will be summarised. Write the first 60 to 80 words as if that’s the only thing Google will show. Get to the answer.
  • Check the new Generative AI performance report in Search Console (rolled out worldwide on 31 August). This is the only place you’ll see AI Overview impressions attributed properly.
  • If your business runs on interactive tools, treat the tool page like a lead magnet. The tool result itself is now in competition with Google’s own answer, so the value has to sit around it – data, context, next steps.

3. ChatGPT Rewrote How It Searches the Web

On 8 August, ChatGPT started scoping most of its background searches to specific trusted domains instead of the open web. Two independent tracking firms caught it in real time. The chart below shows the scale of the change:

Chart showing the ChatGPT retrieval shift on 8 August 2026 — site: operator use jumped 46x, Reddit's citation share fell 86%. Data: Promptwatch citation tracking; independently verified by Suganthan Mohanadasan (16–20 Aug 2026).

Between 16 and 20 August, ChatGPT also swapped its search tool from a JSON structure to a compact query language with freshness windows, domain targeting and separate verticals for products, places, images and widgets. Which sounds dry until you notice what it means: ChatGPT is now behaving less like a search engine that reads whatever comes back, and more like a research assistant that already knows which sites to trust for each topic.

What actually changed for citations

Domains that gained ground: official brand sites, documentation, government sources, primary research, standards bodies. Domains that lost: forums, aggregators, Reddit, community discussion sites.

What to do

If getting cited by ChatGPT matters to your business:

  • Make sure OAI-SearchBot and ChatGPT-User can crawl your site. Cloudflare’s crawler changes (covered last time) matter here.
  • Own your own answers. If someone else’s page is the primary source for a topic in your niche, ChatGPT will scope its search to them and skip you.
  • First-party data, original research, and clear factual pages beat blog posts stitched together from other sources. The shift rewards the source, not the summariser.

4. Parasite SEO Enforcement Paused in the EEA (But Not the UK)

From 30 August, Google stopped applying manual actions under its site reputation abuse policy to searchers inside the European Economic Area. That’s the 27 EU states plus Iceland, Norway and Liechtenstein. The change followed a European Commission investigation under the Digital Markets Act.

Here’s the awkward part for UK businesses: the UK is not in the EEA. Post-Brexit, the UK sits outside the change. Google’s enforcement continues here exactly as before.

Which creates a strange competitive dynamic worth understanding.

RegionSite reputation abuse enforcement (from 30 Aug)
UK searchersFull enforcement third-party content on trusted sites still gets manual actions
EU + EEA searchersManual actions no longer applied; system may separate section and let it rank on its own
Rest of world (US, Canada, Australia, etc.)Full enforcement continues

For a UK small business, the practical read is that competitors relying on parasite SEO tactics against your keywords will still be penalised in UK search results. But if your competitors are getting European traffic through EU-hosted parasite pages, that content is now less exposed to enforcement across the channel. Interesting, more than alarming.

Google also clarified four factors it uses to identify site reputation abuse: extent of first-party involvement, integration with the host site, whether the section brings any value to the host site’s users, and whether the content is presented in a way that misleads.

5. Google’s Old Shopping API Starts Failing Today (Well, 1 September)

If you’re on Shopify, WooCommerce, BigCommerce or any modern platform, this probably isn’t your problem most have already migrated. If you’re running a custom-built store, or an older integration, read on.

Google’s Content API for Shopping officially reached its sunset on 18 August. From 1 September, requests from any client without an active extension start failing intermittently with an HTTP 410 error. Full shutdown is scheduled for early 2027.

What to do

  • Ask your developer or agency whether your Merchant Center feed uses the Merchant API (new, still working) or Content API for Shopping (old, now failing).
  • If it’s the old one, you can request an extension via Google’s form. Don’t wait for products to start disappearing from Shopping listings.
  • Symptoms to watch for: sudden drops in Merchant Center feed processing, products vanishing from Shopping ads, sync errors in your ecommerce backend.

Quick-Fire Round-Up: The Rest of What Happened

UpdateWhat HappenedWhat To Do
Google Search Console: AI reports go globalThe Generative AI performance report is now available in every market as of 31 AugustLog in and check impressions from AI Overviews and AI Mode – new data source you haven’t seen before
Google Preferred Sources buttonAn embeddable button (20 Aug) that lets publishers ask readers to prefer them in AI Overviews, AI Mode and Top StoriesOnly relevant if you publish news or industry content; if so, drop the button somewhere prominent
Meta Ads placement exclusions removedMeta is removing the option to exclude specific placements from ad setsAudit brand safety exclusions now -Audience Network placements will be forced back into your delivery
Gemini 3.8 Flash added to Google AI ModeUsers can pick 3.8 Flash from the “+” menu in AI Mode’s ask barNothing to action – but useful to know when testing how your business appears in AI Mode responses
ChatGPT desktop app updatesSite URL changes without redeploying; Codex chat sharing; Box/Dropbox/SharePoint in LibraryHandy if you use ChatGPT Work; low-priority for most small businesses

The Bigger Picture

Two themes emerged over the last fortnight, and both are worth internalising.

AI is now the source of truth for citations, not the middleman. ChatGPT’s shift towards scoping searches by domain, combined with Google’s expanding AI Overviews, means the sites AI systems already trust for a topic keep getting more visibility. Everyone else has to compete on being the primary source, not a good summariser of primary sources. Original data, first-hand expertise and clear factual pages compound. Rewritten explainers don’t.

Measurement is getting harder before it gets easier. /goto obscures organic attribution. AI Overviews absorb the click. Rank tracking changes accuracy. If you make decisions based only on what a single tool tells you, the odds of making one bad call go up this month. Cross-check Search Console against GA4 against your ecommerce or CRM data before changing anything.

Your Mid-September Action Checklist

  • ☐ Cross-check organic traffic in GA4 against Search Console – the /goto redirect can misclassify sessions
  • ☐ Open the Generative AI performance report in Search Console (now global) and note baseline impressions
  • ☐ Ask your ecommerce developer whether your Shopping feed uses Merchant API or Content API
  • ☐ Audit your top service pages for a clear, extractable answer in the first 60–80 words
  • ☐ Check that OAI-SearchBot and ChatGPT-User aren’t blocked at your CDN or firewall
  • ☐ Review Meta ad set placement exclusions before they get removed
  • ☐ Add fresh content or a photo to your Google Business Profile if it’s been more than 30 days

Final Thought

None of this is an emergency for a well-run small business. Most of it can be checked in a Sunday evening pass through the tools you already use. The real value of a fortnight like this is spotting the pattern – which is that AI systems are getting better at deciding what to show and worse at explaining why. Original content, clear pages, real expertise and clean measurement still win. They just win a bit quieter now.

Sources for this round-up include Search Engine Land, Search Engine Roundtable, Search Engine Journal, Google Search Central Blog, Google Developers documentation, SUSO Digital, Promptwatch, Suganthan Mohanadasan, Nozzle, SerpAPI, and reporting from PPC.land, Digital Applied and Adido Digital. Snapshot as of 15 September 2026.

SEO & Google Ads/PPC: News – Sept 2026

On By DaveIn Business, seoLeave a comment

SEO & Digital Marketing News Round-Up: What UK Small Businesses Need to Action Right Now (Late August / Early September 2026)

If you’ve blinked in the last three weeks, you’ve missed a lot. Here’s the plain-English rundown of what actually changed, what it means for your business, and more importantly: what you should actually do about it.

Laptop screen showing digital marketing analytics dashboard with charts and data
Photo by Unsplash. Swap for your own or upload to your Media Library.

TL;DR The Five Things That Actually Matter This Month

If you only read one thing, read this table. Everything else is context.

# The Change Deadline Who It Affects Urgency
1 Google Ads AI Max auto-migration 1 Sept 2026 Anyone running Google Search Ads with broad match or auto-created assets 🔴 Critical
2 Google Performance Max AI video auto-resize 4 Sept 2026 (opt-out) Anyone running PMax with video assets 🔴 Critical
3 Cloudflare AI crawler default changes 15 Sept 2026 Anyone hosting a site on Cloudflare (especially free-tier) 🟠 High
4 Google August 2026 Spam Update (finished rolling out) Complete – 21 Aug 2026 Sites relying heavily on AI/programmatic content 🟠 High
5 Bing Webmaster Tools old API retired 31 Aug 2026 Anyone using dashboards/tools that pull Bing data 🟡 Medium

1. Google Ads AI Max: The Silent Migration You Might Have Missed

This is the single biggest paid-media change of the year, and Google has been remarkably quiet about it.

From 1 September 2026, Google is automatically migrating Search campaigns using either “Automatically Created Assets” (ACA) or the campaign-level broad match setting over to its new AI Max system. There is no advance decline – you either configured it yourself before the deadline or Google configures it for you.

What actually changes

When AI Max flips on, Google’s AI can:

  • Rewrite your ad text – new headlines and descriptions generated based on your website and existing ads (words you didn’t write and didn’t approve)
  • Match queries beyond your keyword list – using “search term matching” similar to how Performance Max already works
  • Swap landing pages – potentially pointing traffic to different pages on your site

Your brand inclusions/exclusions, budget and bid strategy do carry over.

The uncomfortable data

Google says AI Max delivers a +7% conversion lift. Independent testing tells a different story:

Source Finding
Google’s own aggregate data +7% conversions at similar CPA (excludes retail)
Smarter Ecommerce (250+ retail campaigns) Median 13% increase in conversion value, 0% median ROAS change
Independent test of 30,000 search terms 99% of AI Max impressions produced zero conversions
Lunio (nine-month study) Invalid-traffic rates rose from 2.46% to 5.28%

Not exactly a slam dunk.

Your action list

  • ☐ Before 1 September: Log into every Google Ads account and audit whether campaigns use ACA or campaign-level broad match
  • ☐ Decide per campaign whether to opt in deliberately (Campaigns → Settings → AI Max) or disable the legacy feature first
  • ☐ Warning: Once you disable the legacy ACA feature, you can’t turn it back on. AI Max itself is reversible; ACA disablement is a one-way door.
  • ☐ Export current search term reports and conversion data now so you’ve got a clean before/after baseline
  • ☐ Audit “Limited by budget” campaigns – Google explicitly said it won’t adjust bid targets for you

Regulated industries (health, finance, legal): AI-rewritten ad copy could breach your compliance rules. Take this seriously.

2. Performance Max: Google’s AI Will Reframe Your Videos

If you didn’t already have enough to worry about, there’s a separate PMax change with an even earlier deadline.

From 4 September 2026, Google will use generative AI to auto-create missing aspect ratios (vertical, square) from your existing horizontal video assets in Performance Max campaigns. This applies to video you’ve already uploaded and approved, not just new uploads.

Why this matters

A 16:9 original cut down to 9:16 vertical keeps roughly a third of the width. That means:

  • Burnt-in text overlays likely disappear
  • Logos at the edges get cropped out
  • Legally required disclaimers can vanish
  • Generative AI invents the rest

How to opt out

Contact your Google account team or submit Google’s opt-out form before 4 September. Video settings can still be changed inside your account after that date, but the default from that point is “on”.

If your PMax video has your premises, staff, or actual product on screen – look at what the AI produces before letting it run in your name.

3. Google’s August 2026 Spam Update: What Got Hit

Google’s third spam update of the year ran from 18–21 August 2026. It rolled out globally in every language over about two days and 16 hours.

What Google said

  • No new spam policies were added
  • It re-scores sites against existing rules
  • Link spam and site reputation abuse were explicitly excluded from scope
  • Recovery is measured in months, not weeks

What actually got hit

According to SE Ranking’s study of 100,000 keywords across 20 industries:

  • 16.71% of URLs that had ranked in Google’s top 10 fell beyond position 100 during the update
  • That’s roughly 1.8 times more than a comparable baseline period
  • Common patterns among impacted sites: programmatic content at scale, AI-generated content published without meaningful human review, thin affiliate pages, near-duplicate content hubs

What to do if you were hit

Priority Action
1 Wait until rollout completed (it has – 21 Aug) before making changes
2 Check Search Console dates. Line up traffic drops with 18–21 August
3 Audit pages with low engagement, near-duplicate content, or high AI-probability scores
4 Consolidate or remove thin content rather than mass-rewriting good pages
5 Don’t file a reconsideration request, those are for manual actions, which this isn’t

One thing NOT to do: Don’t panic-rewrite perfectly good pages because an AI detector scored them highly. Detection tools are noisy and rewriting well-performing content in a panic is how you make things worse.

4. Cloudflare AI Crawler Changes: 15 September Deadline

This is the one everyone’s got slightly wrong. Let’s clear it up.

From 15 September 2026, Cloudflare’s new defaults kick in for:

  • New domains being onboarded
  • New sites added by existing customers
  • Existing free-tier accounts that haven’t touched their settings

On pages that show ads, Training and Agent crawlers will be blocked by default. Search crawlers stay allowed.

The actual risk (this is important)

Multi-purpose crawlers like Googlebot, Applebot and Bingbot crawl for both Search AND Training under a single user agent. Cloudflare’s new rule is: the most restrictive setting wins.

Translation: If someone on your team ticked the old “Block AI Bots” toggle in 2025, from 15 September you could accidentally be blocking Googlebot too, which would quietly disappear you from Google Search.

Your ten-minute check

  1. Log into Cloudflare → Security → Settings → AI bot policies
  2. If you see the old blanket “Block AI Bots” toggle, that’s your risk. Change it to the granular Search / Agent / Training controls
  3. Allow Search crawlers unless you have a very specific reason not to
  4. Make a deliberate choice on Training (are you happy for your content to train AI models?) and Agent (do you want ChatGPT, Perplexity etc to fetch your pages when users ask?)
  5. Save. Even saving your current choice explicitly protects you from the new defaults being applied

If your growth loop includes getting cited in AI answers, blocking Training and Agent bots is an SEO own-goal. Most small businesses should probably allow all three.

5. ChatGPT Ads: Now Live in the UK (and 30 Other European Markets)

OpenAI has quietly turned ChatGPT into a self-serve ad platform, and as of 31 August 2026, self-service Ads Manager access is now live across 31 European markets, with the UK already included in the earlier launch cohort.

The state of play

Fact Detail
Where ads appear Free and Go tier users, logged in, adult accounts
UK availability Live (self-serve Ads Manager)
Bidding oCPC (conversion-optimised) now available
Formats Sponsored product cards, multi-product carousel (new in August)
Partners Etsy, Shopify integrations for in-chat checkout
Early UK/US CPM Around $30 (down from initial $60 estimate)

Should you actually spend on this?

Honestly? Not yet, for most small UK businesses. Here’s why:

  • Ads only reach Free and Go tier users (Plus and Pro users see none)
  • Independent tests have been mixed at best – one agency test recorded 8,940 impressions, 58 clicks and zero leads
  • The channel behaves more like brand awareness than direct response
  • Measurement stack is still immature

But: get your OAIQ pixel installed and Conversions API wired up now, so you’re ready to test cheaply in Q1 2027 when the data will be clearer.

6. Quick-Fire Round-Up: The Other Stuff That Happened

Update What Happened What To Do
Bing Webmaster Tools API SOAP/POX interfaces retired 31 Aug 2026 Migrate any dashboards to REST/JSON API
Google Search Console Third-party views for IG, YouTube, TikTok now showing Check your GSC – new data source for social attribution
GA4 & Google Ads AI Overviews and “Ask Advisor” rolling out for benchmarking Explore the new prompts for quick insights
Google Assistant Being replaced by Gemini on Android from 4 Sept Test how your business appears in Gemini responses
Reddit citations in ChatGPT 86% drop reported by Promptwatch in mid-August Diversify – Reddit was a huge citation source for many niches
EU AI Act Watermarking rules for AI content now in force If you use AI-generated marketing content in EU markets, comply with disclosure rules
DaVinci Resolve, CapCut, Premiere Pro All shipped major AI-powered updates Genuine time-savers for video content teams

The Bigger Picture: Three Themes Worth Internalising

1. “Doing nothing” is now an active choice

Google, OpenAI and Cloudflare are all shipping changes with defaults that apply if you don’t act. Small businesses without dedicated marketing ops are the most exposed. Put September deadlines in your diary this week.

2. AI search rewards the same things good SEO always has

Generative Engine Optimisation (GEO) – getting cited by ChatGPT, Claude, Google’s AI Overviews and Perplexity – rewards clear answers, structured content, real expertise and trustworthy brand entities. The tactics haven’t changed as much as the terminology.

Data point worth knowing: SE Ranking found AI-referred visitors spend around 68% more time on site, and Semrush has reported they convert roughly 4.4 times higher than standard organic. AI referrals are still a trickle – but they’re a trickle of pre-qualified people.

3. Your Google Business Profile decays faster than you think

Recent local SEO data suggests GBP impressions drop noticeably for businesses that haven’t posted an update or added a photo in over 30 days. If you haven’t touched yours in a while, that’s a 10-minute Sunday-evening job.

Your September Action Checklist

Screenshot this, save it, tick it off.

Before 1 September:

  • ☐ Audit Google Ads accounts for ACA and campaign-level broad match usage
  • ☐ Export current search term reports as your baseline

Before 4 September:

  • ☐ Decide on Performance Max video auto-resize opt-out
  • ☐ Contact Google account team or submit opt-out form if needed

Before 15 September:

  • ☐ Log into Cloudflare and set explicit AI bot preferences (don’t leave on defaults)
  • ☐ Confirm Googlebot is not accidentally blocked

By end of September:

  • ☐ Review Search Console for August spam update impact
  • ☐ Post fresh content to your Google Business Profile
  • ☐ Test how your business appears in ChatGPT, Claude and Google AI Overviews
  • ☐ Audit your service pages for clear, extractable answers to buyer questions

Final Thought

Most of the damage in a month like this comes from the reaction rather than the news itself. Panic-rewriting content, changing settings without measuring, buying tools that “fix” problems you don’t have. Check your data against Search Console before you change anything. Put the September dates in your diary. Then get back to running your business.

The businesses that win in AI-era search aren’t the ones with the biggest content machines. They’re the ones with clear money pages, real expertise, strong local signals, and settings they’ve actively chosen rather than defaults they’ve inherited.

Any questions on any of this or want help auditing your setup ahead of the September deadlines? Drop a comment below.

Sources for this round-up include Search Engine Roundtable, Search Engine Land, Search Engine Journal, Google’s own Ads and Commerce Blog, OpenAI, Cloudflare, SE Ranking, Yoast, and reporting from Two Octobers, Seafoam, ROI Revolution, Ignite Visibility and Digital Loop. Snapshot as of early September 2026 – this stuff moves fast, always sense-check dates before acting.

Using Command Prompt & CURL for SEO Testing [2026]

On By DaveIn Business, seoLeave a comment

Why Every SEO Should Learn Curl (And How to Start in 5 Minutes)

Most SEOs rely on Screaming Frog, Sitebulb, or Chrome DevTools to check how a website is performing technically. All brilliant tools, but there’s something faster, free, and already installed on your computer: a tiny program called curl.

First, what’s a “terminal”?

Before computers had pretty desktops and clickable icons, you controlled them by typing instructions. That text-based interface is called a terminal, or on Windows, Command Prompt. Think of it as texting your computer directly: you type a command, it does the thing, and prints text back at you.

It’s still around today because for certain technical jobs, typing one line is faster than clicking through ten menus.

What is curl?

curl stands for “Client URL.” It’s a built-in tool on Windows, Mac, and Linux that fetches webpages without rendering them. Instead of seeing the pretty version your browser shows you, you see the raw response the server sends back. Exactly what Googlebot sees when it crawls your site.

That’s the gold for SEO. Your browser hides things; curl shows everything.

Your first curl check in 4 steps

  1. Open Command Prompt. Press the Windows key, type cmd, hit Enter.
  2. Type the command: curl -I https://yourdomain.com (the capital I means “headers only”).
  3. Hit Enter. The server’s response headers print out instantly.
  4. Read the results. Status code is on line one — 200 means OK, 301 means redirect, 404 means missing.

That’s it. You’ve just made the same request Google makes to your site thousands of times a day.

Why this matters for SEO

A single curl check reveals things a browser hides: status codes (is the page actually returning 200?), caching rules (is the CDN configured properly?), security headers (HSTS, X-Frame-Options), server fingerprints (what platform is the site built on?), and redirect chains (are there unnecessary hops slowing down crawling?).

It’s also the fastest way to spot staging environments accidentally leaking into Google’s index, broken canonicals, or compression that’s silently switched off.

Save the full response for deeper analysis

Want the entire HTML page too? Use this:

curl -i --compressed https://yourdomain.com > response.txt

That saves the complete response, headers and HTML- to a text file you can open in VS Code or Notepad++. Brilliant for auditing canonicals, hreflang, structured data, and meta tags in seconds.

The takeaway

Curl is the closest thing you’ll get to seeing your site through Google’s eyes. Five minutes to learn, a lifetime of faster audits. Open Command Prompt today and try it on your homepage, you’ll be amazed what it tells you.

It’s a way of pulling info from websites, without a browser. I think

How Google Bot Crawls JavaScript [2025]

On By DaveIn seoLeave a comment

Ever wondered how Google turns your lovingly handcrafted website into a ranking somewhere below a Reddit thread from 2013? It’s not magic, it’s just a long queue of tiny robot librarians fetching HTML, executing JavaScript, and occasionally having nervous breakdowns when they hit your React app.

This is the life cycle of a webpage inside Google’s digestive system: crawl, render, index, panic. Let’s go step by step before your sitemap starts crying.

1. Crawling: getting the raw HTML

1.1 URL discovery & crawl queue

Googlebot first has to discover your URLs. That can happen via:

  • Links from other pages
  • XML sitemaps
  • Manual submit / “Inspect URL → Request indexing” in Search Console
  • Other Google systems (e.g. feeds, previous crawls)

Discovered URLs go into a crawl queue with priority based on things like page importance and your site’s crawl budget.

1.2 robots.txt and basic checks

Before requesting the URL, Googlebot:

  1. Fetches robots.txt
  2. Checks if the URL (and key resources like JS/CSS) are allowed
  3. Applies host load limits and crawl budget rules

If the page or important JS/CSS files are blocked in robots.txt, Google:

  • Won’t crawl them
  • Won’t be able to fully render your JS content later

Practical implication: Never block /js/, /static/, /assets/, etc. in robots.txt.

1.3 Fetching the HTML (“first wave”)

Googlebot makes a normal HTTP request (like a browser without UI):

  • Gets the initial HTML (without having run JS yet)
  • Parses head tags (title, meta description, canonical, meta robots, hreflang, etc.)
  • Extracts links from the HTML and adds them to the crawl queue
  • Notes references to resources (JS, CSS, images)

At this stage, only what’s in the raw HTML is visible. If your content is 100% client-side rendered (React, Vue, etc.), Google might see almost nothing yet.

Google can sometimes do basic indexing directly from the HTML (e.g. if content is already there), but JS-heavy pages need the next phase.

2. Rendering: running your JavaScript

Google describes the JS pipeline as: Crawling → Rendering → Indexing. Rendering happens in a separate system using an evergreen version of Chromium (a headless Chrome kept relatively up-to-date) called the Web Rendering Service.

2.1 The render queue (“second wave”)

After the initial crawl:

  1. Google adds the page to a render queue.
  2. When resources allow, that queue feeds URLs into the rendering system.
  3. Until rendering happens, Google only “knows” what was in the raw HTML.

This is why people talk about “two waves of indexing” for JavaScript:

  • Wave 1: Index from HTML (if possible)
  • Wave 2: Index updated content after JS has run

Modern research suggests the process is smoother and faster than years ago, but there is still a render queue and potential delay for JS content.

2.2 How Google’s renderer behaves

When a page reaches the renderer:

  1. Google loads it in an evergreen Chromium environment (no UI).
  2. It fetches JS, CSS, and other resources (subject to robots.txt, CORS, etc.).
  3. It executes JavaScript for a limited amount of time (shorter than a user session).
  4. JS can:
    • Modify the DOM
    • Inject content
    • Fetch JSON/XHR/data and add it to the page
    • Add structured data (application/ld+json) dynamically

Important constraints (from Google’s docs & tests):

  • No user interactions: Google doesn’t click, type, or scroll like a user.
  • Time limits: Long chains of async calls may never complete before the renderer stops.
  • Resource limits: Heavily blocking scripts or endless network calls can break rendering.
  • “Noindex = no render” effect: If a page is noindex, Google generally won’t bother rendering it.

2.3 Post-render snapshot

Once JS finishes (or time runs out), Google:

  1. Takes the final DOM snapshot (what a user would see after JS).
  2. Extracts:
    • Visible text content
    • Links added by JS (e.g. SPA navigation)
    • Structured data present in the DOM
    • Meta tags if they are changed or added by JS

This rendered snapshot is what feeds into the real indexing stage.

3. Indexing: storing & scoring the rendered content

With the rendered HTML/DOM in hand, Google moves to indexing.

3.1 Understanding the content

From the rendered DOM, Google:

  • Tokenizes and stores text (words, phrases, headings).
  • Maps entities, topics, and relationships.
  • Reads links (anchor text + target URL) added by your JS navigation.
  • Parses structured data (schema.org, etc.) that JS may have injected.

This is the version of the page that can now rank for queries matching that content.

3.2 Canonicals, duplicates & signals

Indexing also handles:

  • Canonical selection (HTML tags, redirects, link signals).
  • Duplicate / near-duplicate detection, especially if JS rewrites similar pages.
  • Applying meta robots and HTTP headers from the final state after JS (for most cases).

If Google decides another URL is the canonical, your rendered JS content might be stored but not shown as the main result.

3.3 Final result: searchable document

After indexing, the document is:

  • Stored in Google’s index with:
    • Content (from rendered DOM)
    • Links
    • Structured data
    • Various quality & relevance signals
  • Ready to be retrieved and ranked when a user searches for related queries.

4. Where JavaScript sites usually break this flow

Because JS adds extra moving parts, a bunch of things can go wrong between crawl → render → index:

  1. Blocked JS/CSS in robots.txt
    Google can’t render layout or content if the files are disallowed.
  2. Content only after interaction
    If key text appears only after a click/scroll or in a modal that never opens, Google won’t see it.
  3. Too slow or broken rendering
    Heavy JS, long waterfalls, or failing XHR calls mean the DOM never contains the content when Google takes the snapshot.
  4. Infinite scroll / SPA routing without proper URLs
    If content is loaded endlessly on one URL without crawlable links or pagination (e.g. no ?page=2, no anchor links), Googlebot may only see the initial “page”.
  5. Client-side only structured data that doesn’t materialise in time
    If JS injects JSON-LD but too slowly (or fails), rich results won’t trigger.

5. How to see what Google sees (JS-specific)

To understand how your JS is being processed:

  • Use URL Inspection → “View crawled page” & “Screenshot” in Search Console to see the rendered DOM.
  • Compare “HTML” vs “Rendered HTML” to spot what content only appears post-JS.
  • Use “Test live URL” if you suspect render-queue delay.
  • Check Coverage / Pages report for “Crawled – currently not indexed” patterns that often show render/index issues.

So there you have it — from lazy bots fetching half your HTML to a headless Chrome pretending to be a real user for 0.3 seconds. Somewhere in that chaos, your content might actually get indexed.

If your JavaScript site isn’t showing up, don’t blame Google just yet — try unblocking your own files and giving the crawler half a chance. Think of it as SEO mindfulness: eliminate obstacles, breathe deeply, and let the bots eat your content in peace.

Explained in simpler terms – How Googlebot Crawls Javascript –

Stage 1 – Discovery & Crawling: “Finding your page and grabbing the first copy”

1. Google finds your URL

Google finds pages from things you already know:

  • Links on other sites
  • Your internal links
  • Your XML sitemap
  • Stuff you submit in Search Console

It puts those URLs in a big to-do list (crawl queue).

2. robots.txt check

Before visiting a URL, Google checks your robots.txt file:

  • If the page or important files (JS/CSS) are blocked, Google is basically told: “Don’t look here.”
  • If they’re allowed, it moves on.

Simple rule for you:
Never block your JS/CSS folders in robots.txt.

3. Google downloads the HTML (Wave 1)

Google now requests the page, just like a browser:

  • It gets the basic HTML (before any JavaScript runs).
  • From that HTML it grabs:
    • Title, meta description, canonical, meta robots, etc.
    • Any plain text that’s already there
    • Links to other pages
    • Links to JS/CSS/images

At this point, Google has not run your JavaScript yet.

If your important content is already in the HTML (e.g. server-side rendered), Google can often index it right away from this “first wave”.

Stage 2 – Rendering: “Actually running your JavaScript”

Now Google needs to know what your page looks like after JS runs – like a real user would see it.

Because this is heavy work, Google doesn’t do it instantly for every URL.

4. Render queue (waiting line)

After the first crawl, JavaScript pages go into a render queue:

  • Think of it like: “We’ve saved the HTML. When we have time, we’ll come back and run the JS.”

So for a while, Google might only know the bare HTML version of your page.

5. Headless Chrome renders the page

When your page reaches the front of the queue, Google loads it in something like Chrome without a screen (headless browser).

This browser:

  • Downloads JS/CSS (if not blocked)
  • Executes the JS for a short amount of time
  • Lets JS:
    • Change the DOM (the page structure)
    • Insert more text
    • Add more links
    • Inject structured data (JSON-LD)

Then it takes a snapshot of the final page – the “after JS” version.

This is basically:

“What a user would see if they opened your page and waited a bit.”

6. Things that can go wrong here

This is where JS sites often break:

  • Blocked JS/CSS → Google can’t see the layout or content properly.
  • Very slow JS → content appears after Google stops waiting.
  • Content only after a click/scroll → Google doesn’t usually click buttons or scroll like a human.
  • Broken scripts / errors → content never appears at all.

Result: Google’s snapshot may miss your main content.

Stage 3 – Indexing: “Filing your page in the library”

Now Google has:

  • Version 1: HTML-only (first wave)
  • Version 2: Rendered DOM (after JS runs)

7. Understanding the rendered page

From the rendered snapshot Google:

  • Reads all the visible text
  • Sees headings and structure
  • Follows any extra links added by JS
  • Reads structured data (schema)
  • Applies canonical tags / meta robots, etc.

This updated information is used to update your page in the index (second wave).

8. Search results

When someone searches:

  1. Google looks through its index of pages (which contains that rendered version).
  2. It decides which pages are most relevant.
  3. It shows your URL in the results.
  4. When the user clicks it, they go to your live site, not to Google’s stored copy.

Quick “JS SEO” checklist for you

If you remember nothing else, remember these:

  1. Can I see my main content in the raw HTML?
    • If yes → you’re mostly safe (e.g. SSR or hybrid).
    • If no → you’re relying heavily on rendering; be extra careful.
  2. Are JS & CSS allowed in robots.txt?
    • They should be.
  3. Does important content require a click/scroll?
    • Try to have key text and links visible without interaction, or use proper URLs for loaded content.
  4. Is the page reasonably fast?
    • If it takes ages for content to appear, Google may bail before seeing it.
  5. Use Search Console’s URL Inspection → “View crawled page”
    • Compare:
      • HTML Google saw
      • Rendered HTML
    • If you don’t see your text in the rendered version → that’s a problem.

Checking product Schema On a Raspberry Pi

On By DaveIn seoLeave a comment

Goal of the mini-project

The aim here is to –

Verify that product schema (JSON-LD) is implemented correctly on example.co.uk after the migration to Adobe Commerce (Magento).
The script crawls your chosen product URLs and reports if required fields like price, brand, sku, and availability are present.

Step 1 – Open a terminal

Click the black terminal icon on the Pi desktop.

Step 2 – Check Python 3

python3 --version

You should see something like Python 3.9.2 (any 3.7+ is fine).

Step 3 – Install libraries

sudo apt update
pip3 install requests beautifulsoup4

Step 4 – Create a working folder

mkdir ~/schema_check
cd ~/schema_check

Step 5 – Create the script file

nano check_schema.py

Then paste this entire script:



import requests, json, csv, time
from bs4 import BeautifulSoup

# ---------- configuration ----------
# Put your product URLs here (you can add as many as you like)
urls = [
    "https://www.example.co.uk/example-product-1",
    "https://www.example.co.uk/example-product-2"
]

# Fields you want to confirm exist in the Product schema
required_fields = ["name", "brand", "sku", "price", "priceCurrency", "availability"]

# Optional delay between requests (seconds)
delay = 2

# ---------- functions ----------
def extract_product_schema(url):
    try:
        r = requests.get(url, timeout=15)
        soup = BeautifulSoup(r.text, "html.parser")
        for tag in soup.find_all("script", type="application/ld+json"):
            try:
                data = json.loads(tag.string)
                if isinstance(data, list):
                    for item in data:
                        if item.get("@type") == "Product":
                            return item
                elif data.get("@type") == "Product":
                    return data
            except Exception:
                continue
    except Exception as e:
        print(f"Error fetching {url}: {e}")
    return None

def check_fields(product_json):
    found = json.dumps(product_json)
    return [f for f in required_fields if f not in found]

# ---------- main ----------
results = []
for u in urls:
    print(f"Checking {u} ...")
    product = extract_product_schema(u)
    if not product:
        print(f"❌ No Product schema found: {u}")
        results.append([u, "No Product schema", ""])
    else:
        missing = check_fields(product)
        if missing:
            print(f"⚠️ Missing: {', '.join(missing)}")
            results.append([u, "Missing fields", ", ".join(missing)])
        else:
            print(f"✅ All key fields present")
            results.append([u, "All fields present", ""])
    time.sleep(delay)

# ---------- save to CSV ----------
with open("schema_results.csv", "w", newline="") as f:
    writer = csv.writer(f)
    writer.writerow(["URL", "Status", "Missing Fields"])
    writer.writerows(results)

print("\nDone! Results saved to schema_results.csv")

Save and exit:

  • Ctrl + O, Enter → save
  • Ctrl + X → exit

Step 6 – Edit your URLs

Later, open the script again (nano check_schema.py) and replace the two example links with your 10–50 product URLs.
Each URL must be inside quotes and separated by commas.

Step 7 – Run the script

python3 check_schema.py

It will:

  • Fetch each page
  • Extract the Product JSON-LD
  • Report any missing fields
  • Save a summary to schema_results.csv in the same folder

Step 8 – View the results

cat schema_results.csv

or open the file in LibreOffice Calc / Excel.

Example output:

URL,Status,Missing Fields
https://www.example.co.uk/football-goal.html,All fields present,
https://www.example.co.uk/tennis-net.html,Missing fields,priceCurrency availability
https://www.example.co.uk/baseball-bat.html,No Product schema,

Optional tweaks

  • Increase delay = 2 to 5 if you test hundreds of URLs (avoids rate limits).
  • You can import hundreds of URLs from a CSV by editing the script — ask later if you’d like that version.
  • Re-run anytime to confirm schema fixes.

Quick recap

StepActionCommand
1Open terminal(click icon)
2Check Pythonpython3 --version
3Install depspip3 install requests beautifulsoup4
4Make foldermkdir ~/schema_check && cd ~/schema_check
5Create scriptnano check_schema.py
6Edit URLsinside script
7Run itpython3 check_schema.py
8View resultscat schema_results.csv

That’s it. Job done.

You’ve now got a simple tool that checks your product schema in seconds. No fancy platforms. No monthly fees. Just a Raspberry Pi doing proper work.

Run it whenever you push changes. Catch broken schema before Google does. Keep your rich results intact.

The script sits there, ready to go. Update your URLs. Hit run. Get answers.

This is what proper validation looks like – fast, local, and under your control.

Next steps?

  • Bookmark this guide for when you migrate sites
  • Test 50 products now, then spot-check monthly
  • If you need the CSV import version, you know where I am

Your structured data matters. Now you can actually prove it’s working.

Go check your products. Then sleep better knowing your schema’s solid.

Questions? Issues? The comments are open. MOFOs