Full disclosure: This post contains affiliate links. If you make a purchase through these links, 99signals may earn a commission at no additional cost to you.
For the past 18 months, an entire cottage industry has grown up around Generative Engine Optimization.
Agencies started selling llms.txt installation. Consultants pitched content chunking refactors. Vendors promised special schema markup that would make AI engines cite you. The pitch was always some version of the same thing: AI search is new, it has secret rules, and we know them.
Two recent developments just pulled the rug out from under most of that.
First, on May 15, 2026, Google published official guidance on optimizing for generative AI features, and it explicitly named several popular GEO tactics as unnecessary.
Second, Ahrefs released a six-month study spanning over 1 billion data points that tested whether these tactics actually move AI citations. The data confirmed what Google’s guidance implied.
The short version: most of the GEO playbook being sold right now is either unnecessary, unsupported by evidence, or both.
Table of Contents
What Google Actually Said
Google’s position is blunt, and it’s worth quoting the framing directly. The company’s stance is that AEO and GEO are, for Google Search, still just SEO.
The guide names specific tactics site owners can ignore.
llms.txt files. Google says you don’t need to create machine-readable files, AI text files, or special markup to appear in generative AI features. Google may crawl and index those files, but that doesn’t mean they get any special treatment.
Content chunking. There’s no requirement to break your content into small pieces for AI systems. Google’s systems can understand the nuance of multiple topics on a page and surface the relevant part. Danny Sullivan had said much the same in January 2026, noting that Google engineers recommended against chunking.
AI-specific rewriting. You don’t need to rewrite your content to sound “machine-friendly.” AI systems understand synonyms and general meaning.
Schema as an AI ranking signal. Structured data still earns rich results in traditional Search, but Google says it isn’t required for AI search visibility.
Inauthentic brand mentions. Google’s existing spam systems already filter the kind of artificial mention-building that some GEO vendors sell, and AI features depend on those same systems.
Then Google did something that made the message even sharper. On June 5, 2026, it updated its “Do you need an SEO?” documentation with a vendor-vetting question: is your SEO’s advice on AI optimization aligned with Google’s official guidance?
That’s Google handing businesses a framework to audit their GEO vendors. If a consultant is selling llms.txt files and content chunking as their core deliverable, Google just gave you permission to ask why.
What the Ahrefs Data Confirmed
Guidance is one thing. Independent data is another. And the Ahrefs study, drawing on over a billion data points across 14 separate analyses, tested several of these tactics against real citation behavior.
The schema finding is the one that should end a few sales pitches. Adding schema markup had almost zero impact on AI citations. Content structure and factual density mattered far more than markup type.
A few other findings reshape how you should think about AI citations entirely.
“Best of” listicles dominate. Pages formatted as “Best X” or “Top X” roundups are the single most-cited content type, accounting for roughly 40% of all pages cited by ChatGPT.
Being retrieved isn’t being cited. ChatGPT pulls dozens of sources per query but only cites about half of them. The rest become invisible background context. Getting into the retrieval set and actually being quoted are two different games.
AI citation and Google ranking are diverging. 28.3% of ChatGPT’s most-cited pages have zero organic visibility in Google. The overlap between AI Overview citations and top-10 organic results has dropped sharply, from around 76% in mid-2025 to roughly 38% by early 2026.
YouTube mentions correlate strongest with AI visibility. Among all the factors studied, brand mentions on YouTube showed the highest correlation with brand visibility in AI answers. Not schema, llms.txt, or any of the technical tactics being sold.
The Stat That Needs a Caveat
There’s one widely shared number from the study that deserves more honesty than it’s getting in most coverage.
The headline version says 67% of ChatGPT’s top citations come from sources you can’t influence: Wikipedia (29.7%), homepages (23.8%), and app stores (6.6%). Only 32.3% are content types brands can actually create.
That number is real, but read it carefully before you internalize it, as Quattr’s analysis points out.
That 67% is a corpus-level average across the entire dataset, including every query where your brand was never going to be the right answer. It’s not a ceiling on what you can achieve in the specific space where you actually compete.
The question that would actually inform your strategy isn’t “what percentage of all citations are off-limits?” It’s “for the queries where my brand is a reasonable answer, what percentage of citations are genuinely contestable?” The study doesn’t answer that, and nobody should let 67% function as a hard limit on what’s possible.
This is the kind of nuance that separates useful analysis from doom-posting. The data is genuinely useful and the interpretation requires care.
So What Actually Works?
Strip away the debunked tactics and a clear picture emerges. The things that drive AI visibility aren’t AI-specific tricks. They’re the same fundamentals that have driven SEO for years, applied with more discipline.
First-hand experience and original perspective. This is the single biggest lever Google’s guidance emphasizes. AI systems can synthesize commodity content trivially. What they can’t synthesize is genuine expertise, proprietary data, and the kind of first-hand insight only you have. Whatever you know that nobody else has written down is now your biggest content advantage.
Strong fundamentals. Technical health, crawlability, quality content, and topical authority. The boring stuff that was always the foundation. Google’s entire message is that AI visibility is downstream of doing SEO well, not a separate discipline with separate rules.
Multi-channel brand presence. The YouTube correlation isn’t a fluke. Brand presence across the platforms AI models pull from (YouTube, Reddit, Wikipedia, authoritative third-party sites) shapes whether you get cited. This is brand-building, not optimization hacking.
Your homepage as an entity hub. Since homepages are the second most-cited source type, treat yours like a definitive entity reference. Clear brand statements, what you do, why it matters, structured in a way both humans and machines can parse.
How to Audit Your GEO Strategy Before You Waste More Budget
The practical takeaway isn’t “stop caring about AI visibility.” It’s “stop guessing whether your GEO activity is working and start measuring it.”
Pressure-test your vendor against Google’s own guidance. If you’re paying for GEO services, run the June 5 test: does their advice align with Google’s published documentation? If their core deliverables are the exact tactics Google named as unnecessary, that’s a conversation worth having.
Stop investing in the debunked tactics. llms.txt for Google, content chunking, AI-specific rewrites, and schema-as-AI-ranking-signal. Redirect that budget toward original content and brand presence. (One caveat: llms.txt may still matter for Anthropic, OpenAI, and Perplexity crawlers, which have different semantics than Google. Just don’t expect it to do anything for AI Overviews or AI Mode.)
Measure where you’re actually being cited. You can’t optimize what you can’t see. You need visibility into which AI platforms are citing you, which competitors are winning citations you’re missing, and how your AI presence tracks against your organic rankings.
This is exactly the gap Semrush One was built to close. Its AI Visibility toolkit tracks brand mentions and citations across ChatGPT, Perplexity, Gemini, and Google AI Overviews, while the Keyword Gap and competitive research tools surface where competitors are visible (in both organic and AI answers) and your brand isn’t. Pairing that with traditional Position Tracking lets you see whether your AI citations and your organic rankings are converging or diverging, which is the single most useful thing to know in a post-GEO-hype landscape.
Semrush One offers a free 7-day trial with access to 55+ tools if you want to benchmark your AI visibility before committing.
The Bottom Line
The GEO hype cycle was always going to hit this moment. New technology arrives, a thousand “secret tactics” get sold, and eventually the evidence catches up and sorts the real levers from the snake oil.
Google’s guidance and the Ahrefs data did that sorting in the same few weeks.
The verdict is clarifying rather than discouraging. AI visibility matters more than ever. But the path to it isn’t a special file, a schema trick, or a content-chunking refactor. It’s original expertise, strong fundamentals, genuine brand presence, and the discipline to measure what’s actually working.
The shortcuts were never the strategy. They were just easier to sell.
Related Articles