A Billion Users, a 38% Traffic Drop, and a Hidden Sources Button: The Week in SEO, GEO, and AI Visibility August 10 to 14, 2026

Three things happened this week that, read together, tell you almost everything about where search is heading. Google announced that its Gemini app crossed a billion monthly users. Nearly 300 French newspapers filed a competition complaint saying AI Overviews have cut their referral traffic by up to 38%. And OpenAI quietly moved the "Sources" button in ChatGPT into a three-dot overflow menu, where almost nobody will find it.

Assistants are getting bigger. The road back to the websites that feed them is getting narrower. That is the story of the week, and it is not subtle.

The short version

Nearly 300 French newspapers take AI Overviews to the competition regulator

On August 11, APIG, the trade body representing France's general news publishers, filed a complaint with the French competition authority accusing Google of breaching a 2022 compensation agreement by launching AI-generated search summaries without publisher consent, as reported by Search Engine Land and France 24 with AFP. Google launched AI Overviews and AI Mode in France on July 22. APIG says publishers were offered an update to their existing licence, with the only alternative being an opt-out that would cost them search visibility.

APIG president Marc Feuillee framed it plainly: "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." Google's response is the one it uses everywhere: AI Overviews help people ask more complex questions and come with tools for publishers to manage how their content appears. This is not a first offence. France's competition authority fined Google 250 million euros in 2024 over parts of that same 2022 agreement, and the EU opened an investigation in December into whether Google used publisher content to train and serve AI features without appropriate compensation.

Why it matters: the 33% to 38% figure is a publisher estimate, not an audited number, and you should treat it that way. The regulatory pattern is what matters for planning. Europe is steadily converting "AI summarises your content" from a product decision into a licensing obligation. If that lands, paid content partnerships become a visibility channel and brands with no licensing leverage compete on structure and citability instead.

Gemini crosses a billion monthly users

Google announced on August 11 that more than a billion people now use the Gemini app every month, making it the fastest-growing product in the company's history. The usage detail is more interesting than the headline: 63% of users now talk directly to Gemini rather than typing, one in five Gemini Live interactions involve a camera feed or screen share, the app generates more than 150 million images a day, and there are over 100 million active users on iOS. TechCrunch covered the milestone the same day.

Why it matters: nearly two thirds of that usage is voice, and voice answers do not have ten blue links. They have one answer. When you model where your brand can appear in a Gemini interaction, a link in a citation panel is the optimistic case and a spoken mention with no link is the realistic one. That pushes up the value of unlinked brand mentions, consistent entity data, and simply being the name a model reaches for, and it makes last-click attribution even less honest than it already was. The image figure matters too: a product generating 150 million images a day is being used to make marketing collateral at enormous scale.

OpenAI buries the ChatGPT sources button

Glenn Gabe spotted that ChatGPT has started hiding the "Sources" control. It used to sit in the footer of a response. Now it sometimes appears only inside the three-dot "More actions" menu, so a user has to click twice to see which sites the answer came from. Barry Schwartz replicated it and covered it on August 13. Gabe's read: "That can definitely decrease visibility and clicks for citations that show up in the right-side panel." OpenAI has not announced this, and it may be a test. Gabe wondered aloud whether the goal is to keep attention on the ads that now appear in ChatGPT responses.

Why it matters: this is the most consequential item of the week for anyone whose GEO programme is justified on referral clicks. Every extra click between a user and your citation cuts click-through, and a two-click path to a source panel is close to a dead channel for most users. Citations do not stop mattering, but their job changes: they become a trust and consideration signal inside the answer, not a traffic source. Plan measurement accordingly, and stop promising clients that AI citations convert like organic listings.

Search Console's generative AI report expands, and still will not show you clicks

On August 11, the generative AI performance report in Google Search Console appeared on many more properties. Barry Schwartz checked around 25 properties and found it on all of them. Google's John Mueller then confirmed on Reddit that it is still not universal: "Correct, it's not yet every domain. (Also, like the discover and news reports, these are only shown if there's sufficient data to display.)" Google's own Search Console documentation notes the report will not appear if a site has not received enough impressions in generative AI features.

The report gives you impressions, pages, countries, devices, and dates. It does not give you clicks or queries, and it does not separate AI Overviews impressions from AI Mode or other generative surfaces.

Why it matters: here is where Google's framing and practitioner reality part company. Google presents this as visibility into generative AI performance. What you get is an impression counter with no query context and no click data, so you can see that a page is being surfaced but not what for, or whether anyone came. Wire it into reporting anyway, because impression trend by page is a genuine leading indicator. Just label it honestly: it is visibility data, not performance data.

Shopify's numbers push back on the death-of-organic story

Shopify released second-quarter commerce data showing AI-referred sessions to merchant storefronts up 197% year over year, covered by Search Engine Land on August 13. The part most people skipped: organic search traffic also grew 12%, and organic still sent more traffic to Shopify merchants than all tracked AI platforms combined.

The behavioural detail is where the value is. In research-heavy categories, where buyers compare specifications, compatibility, and reviews, AI-referred shoppers converted at roughly twice the rate of organic visitors. In broader taste-driven categories organic stayed the bigger discovery channel, though AI brought in about 1.3 times more first-time customers. And AI-referred shoppers converted at twice the rate when the system used structured Shopify Catalog data rather than scraped or third-party feeds.

Shopify CTO Mikhail Parakhin put it bluntly on X: "Everyone (myself included) assumes AI is swallowing Search and killing organic discovery. Surprisingly, Shopify's numbers say otherwise: AI-referred sessions are up 3x YoY, but organic still grew 12% on a much bigger base. The pie itself got bigger." Worth stating the limits: this is first-party data from a company selling commerce infrastructure, and Shopify did not disclose sample size.

Why it matters: the structured-data finding is the actionable one. Doubling conversion by feeding AI systems clean first-party product data rather than letting them scrape is the clearest return on structured data anyone has published this year. If you sell products, feed quality is now an AI visibility lever, not just a Merchant Center chore. And the broader point stands: a channel growing 197% off a small base is not yet a replacement for one growing 12% off a large one.

Microsoft puts a number on what AI crawlers give back

Microsoft Clarity added an AI Scrape-to-Referral Ratio to its Bot Analytics report on August 13. It compares AI scrape activity against referral traffic in a single metric, breaks it down by operator so you can see which AI sources actually send visitors back, and links straight into filtered session recordings so you can judge whether those visitors are worth anything. Microsoft's framing: "you can evaluate the tradeoff between content extraction and traffic return, then investigate how AI-referred visitors behave once they arrive." In the example screenshot Microsoft published, one operator had sent 41 referrals against a crawl volume implying roughly 6,000 scrapes per referral. That is an illustration, not a benchmark, but a vivid one.

Why it matters: until now the "should I block this AI crawler" argument has been almost entirely vibes. Now there is an operator-level ratio you can put in front of a client, alongside recordings showing whether AI-referred visitors scroll, convert, or bounce. Run it before you touch robots.txt. Some operators will look like a bad trade and some will be quietly sending your highest-intent traffic, and you cannot tell which is which without the data.

Rankings moved on August 12 and 13, and nobody confirmed anything

Third-party volatility trackers lit up on August 12 and into August 13 with movement outside the day-to-day norm. Google confirmed nothing. This continues a run of unconfirmed activity, with similar spikes around August 5, August 1 to 3, July 24, and July 18 to 19. The last confirmed update was the June 2026 spam update, June 14 to June 26. Webmaster forum chatter was severe, including Discover drops reported above 70% in Brazil and one operator writing "traffic today is the worst I have ever seen." Forum sentiment always skews toward the wounded, so weight it accordingly.

Why it matters: do not rebuild anything on an unconfirmed spike, and check whether movement is ranking loss or click loss, because those need different responses. Schwartz made the sharper point himself, that these movements matter less each cycle because Google sends less traffic regardless of where you rank. That is the real story behind every volatility post now.

Also worth knowing

After a judge dismissed Google's suit against the scraping service SerpApi last month, Google filed an amended complaint on August 10 built on a different argument: protecting licensed content inside its results. Google alleges Reddit, which licenses content to it, directed Google not to let third parties extract and commercialise that content. If you rely on SERP scraping APIs for rank tracking or GEO monitoring, that is a live supply-chain risk.

AI Overviews in the local pack are citing self-promotional listicles rather than the businesses themselves. Joy Hawkins documented Google quoting a single lawyer's own listicle as the primary source for every firm in a local result. If you do local SEO, getting into the listicles that AI systems trust is now part of the job, uncomfortable as that is.

Google confirmed on August 10 that it is testing "Create images," "Ask about files," and "Brainstorm" buttons in place of the classic Search and I'm Feeling Lucky buttons. Anthropic began watermarking Claude output worldwide on August 11 under the EU AI Act transparency code, something we covered in depth in our piece on the AI content penalty myth. Meta appears to be crawling the web at scale and may be building a search engine. Google cut the Search Profiles subscriber threshold from 100,000 to 35,000 on August 13. And AI-generated images started appearing inside AI Overviews on August 14.

What to actually do with this week