SEO and AI search in August 2026

Google confirmed one update in August, the spam update that started on the 18th, and the trackers had already been lit up for two weeks before it arrived (August 1 to 3, then the 5th and 6th, then the 12th and 13th) with no explanation from Google for any of that earlier movement. A fair number of people are currently attributing a loss to the wrong thing.

Eight stories below. The one about Europe got maybe a day of coverage and I think it is the most consequential item in the list. I’m going to run one of these every month.

The August 2026 spam update

It ran from the 18th to the 21st, two days and sixteen hours, global and across all languages.¹ Third one this year.

SE Ranking’s numbers afterwards are worse than a spam update normally produces: across 100,000 US keywords in 20 industries, comparing August 17 to 22 against a quieter stretch of July (the 26th to the 31st), 16.71 percent of URLs that had been sitting in the top 10 dropped past position 100, where the July baseline for the same measurement was 9.2 percent.² That makes a top-10 URL about 1.8 times more likely to fall off the first hundred results entirely during the update than outside it. Movement ran upward too. There was a 12 percent rise in URLs turning up in the top 3 that hadn’t been inside the top 20 the week before, and volatility cleared 74.64 percent in all 20 industries tracked, topping out at 85.55 in fashion and beauty.² SE Ranking didn’t look at the domains, so nobody can tell you yet what kind of site got hit or what took its place.

If your traffic broke in the first half of the month this update is not the cause, and rolling back whatever you shipped around the 18th won’t bring it back.

Worth rereading the spam policy documentation either way, because the framing changed earlier this year and plenty of people haven’t looked since. Google now defines spam as techniques used to deceive users or manipulate Search into featuring content prominently, “such as attempting to manipulate Search systems into ranking content highly or attempting to manipulate generative AI responses in Google Search.”³ Manipulating AI Overviews and AI Mode is a policy violation under the same rulebook that covers the ten blue links. Most of the aggressive GEO tactics on sale right now (the mass-generated answer pages, the citation bait, the reworded competitor FAQ) sit inside scaled content abuse, which Google defines as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users.”³ The definition finishes with “no matter how it’s created,” which forecloses the argument that a human edited the output. Whether the August update enforced any of that, Google didn’t say.

Google stopped enforcing its site reputation policy in Europe

On August 30 Google split the site reputation policy in two. Search from outside the European Economic Area and a manual action for site reputation abuse still demotes the affected section of the site; search from inside the EEA and the same manual action does nothing at all. Google’s wording: “The impact of the manual action will not apply. The affected section of the site may be separated in our systems so that, over time, it ranks independently from the rest of the site.”⁴

This wasn’t Google’s idea. The European Commission investigated the policy and required the change. The announcement is about as close to visibly annoyed as Search Central gets: “Our European users are no less frustrated by parasite SEO and other deceptive, pay-to-play tactics that degrade search results, and we stand by our Site Reputation Policy.”⁴ EEA site owners still get the notice in Search Console and can still file a reconsideration request, which now costs and buys them nothing either way.

My read is that this gets abused fast. Renting a subfolder on a newspaper domain (a coupon directory, a sponsored review hub, a casino section) was already good economics, and the only real deterrent was that Google would eventually notice and drop a manual action on the section, which is the deterrent that went away on Sunday for anyone whose traffic is European. What’s left is the algorithmic separation, so a rented section stops borrowing the host domain’s authority but also stops being punished for the arrangement, and it competes on its own merits, which for a coupon directory sitting on twenty years of accumulated links is a considerably better outcome than “separated in our systems” makes it sound.

Two things follow for the day job. Auditing EEA results you’ll see pages ranking that would have been suppressed in a US SERP, and that is the policy working as designed rather than a bug. And reporting a competitor’s parasite play through the spam form is now worth close to nothing in European markets, which someone should hear before they spend a week building the case.

The Search Console AI report is now on every property

Google announced the generative AI performance reports on June 3 and gave them to a subset of properties.⁵ On August 11 they turned up on all of them, with a pop-up in the interface pointing people at the new report, and no announcement of any kind.⁶

You get impressions inside AI Overviews and AI Mode, split by page, country, device and date.⁷ No clicks, no click-through rate, no position, no queries. Clicks out of AI Overviews and AI Mode do get counted, but they land in your Web search type totals alongside ordinary organic clicks with no dimension anywhere to pull them apart again, and the report isn’t exposed in the Search Console API either, so any export is a button and a spreadsheet. Which means you can see that Google put your URL in front of someone inside an AI answer forty thousand times last month and still not learn whether that produced a single visit.

It’s worth opening anyway. Impressions tell you which of your pages the AI systems find quotable, and that is a real input into what you write next, but anyone building a traffic estimate on top of impressions is guessing, including the tools that will sell you one. I wrote up the metrics I think survive this a few weeks back.

Shopify’s Q2 numbers

Shopify reported on August 5. Traffic from AI tools into merchant storefronts tripled year over year and orders that originated in AI tripled with it, which is the pair everyone quoted, though the number I keep coming back to is that half of AI-referred sessions land straight on a product detail page, 2.5 times the rate for traditional search.⁸ Those people arrive already decided, having done the comparison somewhere you have no visibility into.

Traditional search grew through all of it, still around a third of storefront sessions and up 1.3x across two years.⁸

Then 75 percent of AI-attributed purchases happened outside the top 100 categories.⁸ Harley Finkelstein described the mechanism as agents making repeated calls into the catalog, “working with richer structured data to match products with the buyer’s specific intent, rather than just keywords,”⁸ That is product schema being used as a query surface instead of a display format. Thin product data (no materials, no dimensions, no compatibility fields, one generic paragraph) gives an agent nothing to match a shopper against, on a request the shopper never phrased as a keyword you could have targeted anyway.

All of it is Shopify’s attribution of Shopify’s own traffic, announced by Shopify on an earnings call, so I’d take the direction and leave the multiples alone.

AI Mode started booking hotels

On August 27 AI Mode learned to book things. Flight price tracking with email alerts went live in more than 180 countries against pricing from over 300 airlines and travel sites, hotel booking arrived through conversation (US and English first, paid for inside Google with Google Pay), and prices can now be shown in points or miles anywhere in the world.⁹ Google calls the result “an AI travel agent of sorts.”⁹

The hotel partners at launch are Booking.com, Choice Hotels, Expedia, Hilton, Hotels.com, IHG, Marriott, Priceline, Trip.com and Wyndham.⁹ Ten names. If you run an independent property or a mid-size OTA there is no content strategy that puts you on that list, and getting onto it is a business development problem rather than a search one, which is worth saying out loud before somebody proposes a content sprint at it. Google ran the same play in retail with agentic checkout, and travel is usually where it tests the shape first.

Cloudflare’s crawler defaults change on September 15

Not August news, but it lands in two weeks and I keep meeting people who haven’t heard about it.

Cloudflare announced on July 1 that from September 15 its defaults block mixed-use crawlers on any page carrying ads. Mixed-use means a crawler that blends search, agent use and AI training and won’t declare which of those it is doing on a given request. Declared search crawlers still get through; training and agent crawlers get blocked on ad-bearing pages automatically.¹⁰ The defaults change for new Cloudflare customers, for new sites created under existing accounts, and for every existing free-plan customer, which is the part that catches people. Paying customers keep what they have and can override from the dashboard.¹⁰

Cloudflare also swapped Pay Per Crawl for Pay Per Use, which pays when content shows up in an answer rather than when a bot fetches the page. Two partners so far (Ceramic.ai on a pay-per-query model, You.com paying on demand for premium content).¹¹

Go and look at your bot settings before the 15th. When I ran the crawler access study the most common finding was sites blocking AI crawlers without anyone having decided to, and this will add to that.

Publishers went to court in Washington and to the regulator in Paris

Judge Amit Mehta spent part of the last week of August hearing Google’s motion to dismiss Penske Media’s antitrust suit over AI Overviews. He is the same judge who found in 2024 that Google holds an illegal monopoly in general search, so the market power question is already settled in that courtroom in a way it would not be anywhere else. Reporting out of the room has him unimpressed with Google’s framing of AI Overviews as a routine product improvement, saying the arrangement “all seems really unfair” and that product improvements “are not immune from antitrust scrutiny.”¹² No ruling yet.

Penske’s theory is narrow. OpenAI and Perplexity pay publishers for content; Google takes the same content for free because no publisher can afford to delist itself, and the free input raises rivals’ costs.¹²

Then on August 11, roughly 300 French newspapers under APIG filed with France’s competition authority, alleging Google launched AI Overviews there without publisher consent and in breach of a 2022 compensation agreement, which is the same agreement the authority already fined Google 250 million euros over in 2024 (a different part of it), so this is not a regulator that needs convincing.¹³ APIG’s president Marc Feuillee: “The editors are not seeking to halt innovation. They are asking for this value to be shared, and for compensation of the use of their content.”¹³

Neither resolves in 2026, and I mention them because they are already moving what publishers think they can ask for.

What an AI licensing deal is worth in citations

Press Ranger and OtterlyAI took 129.3 million citations from June 2026 across seven platforms (ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Copilot, Gemini and Claude) and matched them against every confirmed licensing agreement between an AI company and a news publisher.¹⁴

Pages from OpenAI-licensed publishers averaged 10.2 citations each on ChatGPT against 6.9 for unlicensed ones, which is the 48 percent that everybody ran with. Publishers who signed with OpenAI and nobody else came in 112 percent above unlicensed. Across all seven platforms, OpenAI licensees took 46 percent more citations per cited page.¹⁴

Google’s own licensees came out slightly worse: publishers holding a Google deal were cited on AI Overviews less often than comparable unlicensed publishers, and Perplexity’s licensees landed at parity on Perplexity.¹⁴ OpenAI is the only licensor in the data with a home-field advantage on its own product.

The authors are upfront that this is correlation and that they aren’t claiming the deals caused it, and the obvious confound is that OpenAI signed large established brands that get cited more regardless of any contract.¹⁴ Even so, it’s the first dataset of real size on the question, and a publisher sitting across the table from OpenAI now has a figure to start from.

OpenAI also moved its ads business into 31 European markets on August 24, six months after the US pilot, sold through Publicis, Omnicom, WPP, Havas, Dentsu and MediaPlus, with roughly 20 percent of queries showing direct commercial intent.¹⁵ Varun Shetty, who runs media partnerships there, was asked in June whether publishers would see any of that revenue and said “Not at this point.”¹⁶

What I’m watching in September

• Whether the major AI companies actually split their crawlers into declared search, agent and training bots before Cloudflare’s defaults change on the 15th.
• Mehta’s ruling on the motion to dismiss. If it fails, Google goes into discovery on how AI Overviews got built.
• Click data in the Generative AI report. Until that shows up, every AI visibility figure in circulation is somebody’s estimate.
• How fast the EEA carve-out turns into parasite SEO you can see in European results.

Frequently asked questions

What Google updates happened in August 2026?

Google confirmed one update: the August 2026 spam update, which started on August 18 and finished on August 21 after two days and sixteen hours, running globally across all languages. It was the third spam update of 2026. Google did not confirm a core update, despite heavy third-party volatility readings in the first half of the month.

Why did my rankings drop in early August 2026 if the spam update started on the 18th?

The spam update cannot explain movement before August 18. Rank trackers recorded separate spikes around August 1-3, August 5-6 and August 12-13, none of which Google has attributed to an announced update. If your traffic changed in the first half of the month, the cause is something else, and rolling back changes made around the spam update will not recover it.

What changed with Google's site reputation abuse policy in Europe?

From August 30, 2026, manual actions for site reputation abuse no longer affect rankings for searchers in the European Economic Area, following a European Commission requirement. Outside the EEA the manual action still demotes the affected section of the site. EEA site owners still receive the notification in Search Console and can still file a reconsideration request, but the ranking penalty does not apply to them.

Does the Search Console Generative AI report show clicks from AI Overviews?

No. The report shows impressions inside AI Overviews and AI Mode broken out by page, country, device and date, but no clicks, no click-through rate, no position and no query data. Clicks from AI features are counted inside your Web search type totals mixed with ordinary organic clicks, and there is no dimension to separate them. The report is also not available through the Search Console API.

What does Cloudflare's September 15, 2026 change actually do?

From September 15, Cloudflare's default settings block mixed-use crawlers, meaning crawlers that blend search, agent use and AI training without declaring which they are doing, from any page that hosts ads. It applies to new Cloudflare customers, new sites created by existing customers, and all existing free-plan customers. Paying customers keep their current settings and can override the defaults in the dashboard.

Do AI licensing deals actually increase citations?

On ChatGPT, licensed publishers are cited more. A Press Ranger and OtterlyAI study of 129.3 million citations from June 2026 found pages from OpenAI-licensed publishers averaged 10.2 citations each on ChatGPT versus 6.9 for unlicensed publishers. The same pattern did not hold elsewhere: Google-licensed publishers were cited slightly less often on AI Overviews than comparable unlicensed publishers, and Perplexity licensees were at parity on Perplexity. The study is correlational and its authors do not claim the deals caused the difference.

Keep going

More on measuring AI search, what AI Overviews changed, and where Google is heading.

Measurement AI Search Visibility Metrics: What They Are and Why They Matter Now The KPIs that measure whether ChatGPT, AI Overviews, Perplexity and Gemini cite your brand. What each metric tells you, what the data shows, and which ones are worth tracking. AI Search AI Overviews vs Classic SEO: What Actually Changed in 2026 AI Overviews did not kill SEO, they split it into two jobs: ranking on the page and getting cited in the answer. What changed, what stayed the same, and how to earn the citation. AI Search How Google Search Is Trending in 2026: Reading the Signals Google made AI Mode the default at I/O 2026. Here's what that actually changes for SEO: how search behavior, the results page, and measurement shift when search becomes a conversation. Guides SEO Trends 2026: What's Actually Changing (and What's Just Hype) A practitioner's 2026 SEO trends report that grades the loud 2025 predictions against the data: what AI Overviews, AI Mode, and core updates actually changed, what got overhyped, and what to do about it.

Sources

1. Google, Search Status Dashboard: Ranking Updates. August 2026 spam update, started August 18, 2026, duration 2 days 16 hours. ↩

2. Search Engine Land, Google’s August 2026 spam update hit rankings harder than normal (August 27, 2026), reporting SE Ranking’s analysis of 100,000 US keywords across 20 industries, comparing August 17-22 with July 26-31. ↩ ↩²

3. Google Search Central, Spam policies for Google web search. Source for the spam definition covering generative AI responses, and for the scaled content abuse definition. ↩ ↩²

4. Search Engine Land, Google won’t respect manual actions for site reputation abuse in European Economic Area (August 28, 2026), on Google’s Update to the Site Reputation Policy (August 2026). ↩ ↩²

5. Google Search Central Blog, Introducing Search generative AI performance reports in Search Console (June 3, 2026). ↩

6. Search Engine Roundtable, Google Search Console Generative AI Performance Report Live For All (August 11, 2026). ↩

7. Google, Generative AI performance report (Search), Search Console Help. ↩

8. TechCrunch, Shopify says AI search is driving more traffic and sales, not replacing Google (August 5, 2026), on Shopify’s Q2 2026 results. ↩ ↩² ↩³ ↩⁴

9. TechCrunch, Google’s AI Mode can now track flight prices, help book hotels, and more (August 27, 2026). ↩ ↩² ↩³

10. TechCrunch, Cloudflare’s new policy pushes AI companies to pay for publishers’ content (July 1, 2026). Source for the September 15 default change and which customers it applies to. ↩ ↩²

11. Cloudflare, Making AI search smarter (July 1, 2026). Source for Pay Per Use replacing Pay Per Crawl, and for the Ceramic.ai and You.com partnerships. ↩

12. Search Engine Journal, Judge in Penske vs Google says AI situation “seems really unfair” (August 27, 2026). The remarks attributed to Judge Mehta come from reporting posted out of the courtroom, not from a released transcript. ↩ ↩²

13. Search Engine Land, 300 French newspapers file complaint over Google AI Overviews (August 11, 2026). ↩ ↩²

14. Press Ranger and OtterlyAI, via Media Copilot, OpenAI-licensed publishers got 48% more ChatGPT citations in June, study finds (August 20, 2026). 129.3 million citations across seven AI platforms, June 2026 data. ↩ ↩² ↩³ ↩⁴

15. Digiday, OpenAI’s ads business hits Europe at the six-month mark (August 2026). ↩

16. Press Gazette, OpenAI not planning to share advertising revenue with publishers. Varun Shetty speaking at the WAN-IFRA World News Media Congress, June 2026. ↩