On May 15, 2026, Google published its first official documentation on optimizing for AI Overviews and AI Mode. The document is short, plainly written, and quietly devastating to about $200M of "GEO consulting" services that have been sold over the last 24 months. If you've been pitched any of the following in the last year, you should read what Google actually said.
I'll cover what Google did, what it kills, what it confirms, and what to do about it. Then I'll tell you the part Google didn't say — the part that matters more for SMBs than any of the official guidance.
What Google actually published
The new guide lives at developers.google.com/search/docs/fundamentals/ai-optimization-guide and is titled "Optimizing your website for generative AI features on Google Search." It was announced by John Mueller through the Search Central blog. Three sentences in particular are doing most of the work:
One: "AEO" and "GEO" are not separate disciplines from SEO. Google's exact framing — "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
Two: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search." That is Google saying, on the record, that the llms.txt file your agency or SEO tool has been recommending — possibly charging $79/month to auto-generate — does nothing for Google AI Overviews or AI Mode.
Three: "There's no requirement to break your content into tiny pieces for AI to better understand it." The "content chunking" tactic — pre-breaking your articles into 50-word AI-quotable blocks — is officially not required.
What just died
Four tactics Google explicitly debunked:
- llms.txt files. Google may crawl them like any other text file, but they receive no special indexing treatment. They do not influence AI Overviews or AI Mode.
- Content chunking for AI parsers. Pre-fragmenting articles into AI-friendly bite-sized blocks is unnecessary. Google's systems understand multi-topic pages and extract relevant passages.
- AI-specific rewriting. Rewriting copy to capture every long-tail synonym variation is not required. Google's systems handle synonyms and general meaning natively.
- Pursuing inauthentic mentions. Google's spam systems already block what AI features would otherwise reward. Buying or manufacturing brand mentions to game LLM outputs doesn't work.
What Google confirmed actually works
The other half of the guide is more constructive. Three things Google validated as the actual visibility levers:
1. First-hand, non-commodity content
Google's framing: "Is this something only we could have written?" If your content can be reproduced by a competitor in 20 minutes by feeding the same topic to ChatGPT, it's commodity content, and AI Overviews will not surface it. The differentiator is original observation, original data, original perspective — the stuff a generic content shop can't fake.
2. Multimodal assets
Images and video are being surfaced in AI Overview responses. Articles with original photography, custom diagrams, screenshots from real workflows, or short video explainers get more citation weight than text-only equivalents.
3. Technical hygiene
The floor stuff: indexable, crawlable, semantic HTML, properly structured headings. Google notes that "perfectly valid HTML isn't required" — but the underlying semantic structure helps screen readers and accessibility tools, which increasingly overlap with how AI agents parse content.
The bigger thing nobody is talking about
Here's the wrinkle the analyses are missing. Google's guide is specifically about Google Search's AI features — AI Overviews and AI Mode. It does not govern ChatGPT, Perplexity, Claude, or any other AI search surface.
llms.txt and similar machine-readable files may still serve a purpose for non-Google AI engines. Perplexity has its own discovery patterns. Claude has its own. ChatGPT pulls from its training corpus plus live web fetches that respect different headers. Google explicitly clarified its own behavior. It did not speak for the others.
"There is no llms.txt back door into Google. There never was. But that doesn't mean llms.txt is useless across the entire AI ecosystem — it just means Google specifically isn't using it."
Which means the practical question for marketers in 2026 isn't "AEO or SEO?" — Google answered that one — it's "do I care about visibility on engines other than Google?" If yes, you might still build an llms.txt for Perplexity and Claude. If your traffic mix is 95% Google-originated, skip it.
What this means for your marketing budget
Three concrete moves:
- If you're paying an agency for "GEO optimization" as a separate line item from SEO, get the line item explained. Google has now stated, on the record, that the disciplines are the same. Some agencies are charging twice for one service. Some are doing legitimately different work (e.g., monitoring AI citations as a metric). Get specifics on what's being delivered.
- If you have an llms.txt on your site, you can keep it. It costs you nothing. It probably does nothing for Google. It may help with non-Google AI engines. The cost-benefit makes sense to maintain, not to build a $79/month subscription around.
- Invest in first-hand content. This is the single biggest signal Google validated. Articles that report from inside your business — your data, your case studies, your specific experience — are now meaningfully more citable than generic explainers. The companies that figure out how to systematize "non-commodity content" are going to compound for the next 24 months.
The honest part
I built our own pitch over the last 18 months partly on "we do AEO." That language is now retired. Not because AI search doesn't matter — it matters more than ever — but because Google has clarified that the discipline that makes you visible in AI search is the same discipline that's always made you visible in regular search. Just with more emphasis on what only you could write.
If you've been paying for GEO/AEO services and wondering whether you got value, the right next step isn't to fire anyone — it's to ask your agency to walk you through what changes in their work post-May 15. The good ones will have already adjusted. The ones still selling chunking and llms.txt-generation services six weeks after this guidance dropped are telling you something.
If you want a candid read on whether your current content strategy passes Google's "only-we-could-have-written-this" test, our free Marketing Score covers it. A senior strategist will audit your top 5 indexed pages against the new guidance and tell you which ones are AI-ready vs. commodity-replaceable. Hand-audited, emailed within 1 business day.