SEO

If your Search Console impressions fell off a cliff last week, stop investigating. Google says it was a logging bug. A recipe creator complained on a Thursday about AI Overviews generating her cooking-step photos, and by Monday Google had confirmed the test on the record and stopped it. The bidding change every advertiser braced for went live the same day, alongside two more landing in September that will make any of it impossible to attribute. And new research finds ChatGPT now cites noticeably fewer domains per answer than it did in July. And much more, in today's Search Shift edition.

OpenAI's position is that a disallow line in robots.txt may not bind a fetch a person asked for. Its bot reached blocked URLs on about half the European sites that named it, while Anthropic says all of its crawlers obey the file. Google has now published the dates by which advertisers lose the campaign types it wants retired, and the door on building new ones shut two weeks ago. And a small study suggests ChatGPT writes the brands it will cite into its own search query before it retrieves a single page. And much more, in today's Search Shift edition.

From Search Rankings to AI Recommendations: The New Rules of Brand Discovery Search is changing, and the shift goes far beyond adding AI to the search experience. In our latest Leader Interview, Alessandro Benigni, Co-Founder & CMO at Geogen, shares a provocative perspective on what happens when AI assistants become the new discovery engines. We discuss: → Why traditional SEO may no longer be enough → How Generative Engine Optimization (GEO) is changing digital marketing → Why brands need to measure their Share of Voice in AI responses → What makes a brand trustworthy enough to be recommended by AI → Why proprietary data and original thinking are becoming critical competitive advantages → How marketing leaders should prepare for the next era of search One point particularly stood out: "The future may not belong to the brands that rank highest, but to the brands AI recommends." If you're a founder, marketing leader, CMO, growth leader, SEO professional, or technology leader looking to understand where AI-powered search and brand discovery are heading, this is a conversation worth reading. Read the full interview: [https://lnkd.in/dWFxgMJk] #AI #MarTech #GEO #DigitalMarketing

AI sends 1.1% of publisher visits. That number is being read as reassurance, and it is the wrong reading. A referral log cannot record a visit that never happened. Search Console cannot tell you an edge rule at Cloudflare is the reason Googlebot stopped coming, and two more sites just found that out the hard way. Microsoft now hands you the crawl-to-referral ratio one site owner spent a year computing by hand. And ChatGPT quietly moved its Sources button one click further away. And much more, in today's Search Shift edition.

A third of everything AI quotes is a listicle. 40,968 of the 119,682 citations I looked at this week landed on a URL shaped like "best X", "top 10", "X vs Y", "alternatives to" or "review". 34.2% of all citation weight. 30.7% of the individual pages. I want to be careful about what this does and does not say. It does not say listicles are good. It says they are what the machine has, in the shape the machine needs. Think about what an AI answer has to do. Someone asks which project management tool to buy. The engine needs a set of named options, compared on attributes, with some basis for ordering them. There is exactly one content format that has been optimised for that structure for fifteen years, and it is the format everyone got sniffy about. What surprised me is that this has nothing to do with sector. Retail sites, community forums, publishers and nonprofits all show the same pattern. It is the shape of the page being rewarded, not the kind of organisation that published it. The uncomfortable part for anyone who moved their strategy to thought leadership: essays do not get quoted for buying questions. Comparisons do. The opportunity is that most listicles in most categories are terrible, thin, and written by people who have never used the products. That is a low bar to clear, and clearing it now has a measurable payoff rather than a theoretical one. Last time you published a comparison page, was it because you wanted to, or because a keyword tool told you to? Methodology: 39,335 quoted URLs across 299 domains, Aug 1 to 7, within the 3,163 brands Geogen tracks. A URL counts as comparison-shaped when its path contains best, top, versus, alternatives, review or compare. Weight is citation-weighted, so heavily quoted pages count more than lightly quoted ones. URL-level detail widened from 15 domains to 299 this week, which is why this figure is materially more reliable than any earlier estimate.

A rival allegedly aimed 2,350 backlinks reading "buy steroids online" at a competitor, and a federal judge just let the case past dismissal. Google's scraping suit against a SERP data provider now turns on the licensing contract it signed with Reddit rather than on its own terms of service. Google answered the French publishers who took it to a regulator this week. And on August 24 a number in Merchant Center reports drops sharply without a single visit being lost. And much more, in today's Search Shift edition.

Winning the AI answer is not like winning a ranking. You can hold it one week and be gone the next. I have now measured the same 40 buying searches three weeks running, which finally lets me compare how fast each layer of the page changes. Google's organic top ten held 82.3% of its sources from one week to the next. The AI Overview above it held 48.3%. The week before that: rankings 65.3%, AI sources 38.4%. Two independent comparisons, same 40 searches, same answer both times. AI's source list retains roughly 60% as much as the rankings underneath it. Call it twice the turnover. This inverts the instinct most of us carry. Rankings feel volatile because we have spent careers watching them wobble and building dashboards to track every position change. But relative to the answer box, rankings are the stable layer. Two practical consequences. A single check tells you very little. If you looked at the AI answer for your category a fortnight ago, roughly half of what it cited then is not what it cites now. That is not a trend, that is the baseline noise level, and any conclusion drawn from one look is drawn from a coin flip. And displacement is cheap. Something that took years to earn in organic can be lost from an AI answer in a week, which cuts both ways. The incumbent's position is far less defended than it looks. If your AI visibility report is monthly, it is describing a state that changed twice before you read it. Methodology: fixed basket of 40 commercial buying queries scraped weekly via ScrapingDog, US desktop, on Jul 25, Aug 1 and Aug 8. Stability is the share of a query's sources still present one week later, averaged across queries, comparing like with like on the same query. Geogen tracks 3,163 brands across seven AI engines; this post uses its external SERP basket.

Google shipped the AI performance report to every Search Console property yesterday, minus the two columns that would answer anything. Around 300 French newspapers took a different route to proof and handed a regulator the contract Google already signed. A study of 1,249 answers reverse-engineered what ChatGPT's index actually keeps of a page, and the number is smaller than most people optimize for. Anthropic began stamping Claude's output so it can be recognised later, worldwide. And much more, in today's Search Shift edition.

A Healthline page was published on a Tuesday. By Wednesday, AI had quoted it 143 times. I have wanted to measure this for a month and only just could. It needs two things at once: the real publication date of a page, and the first week we ever saw AI quote it. We now have both. For pages published while we were watching, the median time from publishing to being quoted by an AI engine is seven days. 16.7% were quoted within 24 hours. 61% within a week. The slowest took 25 days. A TechRadar review: 113 citations, two days after publishing. A Business Insider piece: 108 citations, four days. Twenty years of SEO taught everyone to expect a lag measured in months. Write it, publish it, wait for the crawl, wait for the index, wait for the rankings to settle, check back next quarter. That waiting period is largely gone. The machine reads continuously and quotes almost immediately. Which changes what a content calendar is for. If a page can enter the answer layer this week, then publishing in response to what is happening now is a viable strategy again, rather than a bet that pays out in Q4. The catch is symmetrical. Everyone else's pages are being read this fast too, and last week I showed that AI's source list turns over roughly twice as fast as Google's rankings. Fast in means fast out. How long is your current publish-to-live cycle, and is it the machine slowing you down or the approvals? Methodology: 36 pages that were published inside our four observation weeks, drawn from 39,335 quoted URLs across 299 domains, within the 3,163 brands Geogen tracks. Pages published before we began observing are excluded, because their apparent lag would just measure their age. First-seen is resolved to the week, so each figure is an upper bound on the true discovery time. Publication dates read from structured page data.

Three new buttons appeared under Google's search box yesterday, and not one of them returns a list of pages. Google confirmed the test on record. Meta is reportedly building a web index of its own so its AI stops asking Google. A diner can now book a table without ever reaching the restaurant's site. And a study of 50,000 commercial prompts found that buying the ad slot inside ChatGPT almost never earns you a citation in the same answer. And much more, in today's Search Shift edition.