Google’s AI Optimization Guide landed last week and every agency listicle quoted it straight. The most interesting thing about the document is not what it says but the genre it belongs to: public guidance that minimizes a ranking input later confirmed by leak or trial. The sections below line up the four verses of that song and end with what to do about the fourth.
The guide is a defensive document, not a strategy document
Google’s AI Optimization Guide for Search reads as a posture, not a playbook. Published on the developers site, it tells SEOs that ranking in AI Overviews and AI Mode is the same work as ranking in Search, with no AI-specific tactics worth naming. The document lands in the middle of an antitrust environment where AI Overviews are the central exhibit.
Chegg sued Google in February 2025, alleging AI Overviews hurt its traffic and revenue by extracting answers from indexed content. Press Gazette, citing Chartbeat, reported in early 2026 that the average news site saw Google referral traffic fall 33 percent year over year. Against that backdrop, a guide that frames AI search as continuous with regular Search hedges against a narrative that AI Overviews are a new and value-extracting product.
The argument here is from incentives, not from documented intent. A company facing publisher lawsuits and regulator scrutiny has a reason to publish documents that minimize the gap between Search and AI search, and the gap is precisely what the lawsuits and the Chartbeat data are measuring.
The pattern is older than AI search
Google has published public guidance that minimized a ranking input three times before, and each time the leak or the trial later confirmed the input mattered. The fourth-verse framing only lands if the first three verses are on the page, so the next paragraphs walk through them.
Verse one is links. Google spent a decade telling SEOs that links matter less than they used to and that link building was a lower-leverage activity than content quality. The 2024 Search API leak, analyzed by Mike King at iPullRank and first surfaced by Rand Fishkin at SparkToro, showed link-related features tied to source authority and click-driven trust scores active in the ranking pipeline. The public statement was “links matter less”, the receipt was thousands of leaked feature names.
The shape of verse one is worth pausing on, because every later verse copies it. The public guidance never quite says links do not matter, it says they matter less, that the audience is over-indexing on them, and that effort is better spent elsewhere. The leak then shows a pipeline that still grades sources by link-derived authority. The gap is not between “matters” and “does not matter”, it is between “matters less than you think” and “matters enough to ship a feature for”.
Verse two is domain authority. Google has said for years that domain authority “isn’t a thing they use”, framing it as a third-party metric invented by Moz. The same 2024 leak surfaced a feature literally named siteAuthority, with Search Engine Land’s leak coverage cataloguing it alongside site-level quality signals. The public statement was “there is no site-wide authority score”, the receipt was a field with that name.
Verse three is click data. Google has repeatedly said clicks are not a ranking factor and that user behavior signals are too noisy to use directly. The leak surfaced NavBoost, a system that re-ranks results using long-click data. Sworn DOJ testimony from Pandu Nayak and Eric Lehman, summarized in Search Engine Land’s antitrust trial coverage, confirmed NavBoost uses click data from Chrome and from logged-in search sessions. The public statement was “clicks are not a ranking factor”, the receipt was a system named in court.
”Just write for humans” is the fourth verse
The AI Optimization Guide’s core message is that there are no AI-specific tactics worth naming, that the same SEO foundations apply, and that writing for humans is the work. That sentence is the same rhetorical move applied to AI search: name an input the audience suspects matters (AI-specific optimization), and reframe it as a thing not worth optimizing for. The verb pattern is identical to the prior three verses.
Google’s AI Optimization Guide does not describe how passages are selected for AI Overviews, how source diversity is weighted, how citation eligibility is decided, or how AI Mode chooses which retrieved chunk to surface. It tells SEOs that the same foundations carry over and that quality content is what wins. The omission is the point: the document is structurally a guidance document about Search, with AI named in the title and absent from the mechanics.
The pattern is not “Google lies”, it is “Google publicly minimizes what later turns out to matter”.
A document that would falsify the pattern is easy to describe. It would name a specific ranking or retrieval input used by AI Overviews or AI Mode, describe the signal it depends on, and explain how a publisher could optimize for that signal. No Google document has ever done that for any ranking input, and the AI Optimization Guide does not break the streak.
The “AEO doesn’t exist” pushback is the wrong read of the same guide
The strongest opposing view is that the AI Optimization Guide proves AEO and GEO are marketing-team fiction. The argument runs: Google itself says there are no AI-specific tactics, so the cottage industry of AI optimization consultants is selling air. Concede the shared ground first. Foundational SEO does carry into AI retrieval, and the SEO basics chapter covers most of what the guide names as table stakes.
The pushback is the wrong read because the guide is evidence about what Google chooses to publish, not about whether AEO is a real surface. The same document would have been published whether AI Overviews use a distinct retrieval stack or not, because the incentives to publish it sit on the antitrust and publisher-relations side, not on the technical-disclosure side. Reading the guide as proof that AEO is fiction requires assuming Google’s public guidance is a faithful description of the system, and the prior three verses show that assumption has a poor track record.
The clearest evidence the playbook is already running is where SEO budgets are moving. Teams that ran organic-only programs for years are now investing in Reddit, YouTube, and social presence, because those off-site footprints are the signals AI systems sample and the citations brand pages can point back to. That spend exists because operators have already concluded the surface is different, regardless of what the AI Optimization Guide calls it.
The guide is a floor, the real work starts above it
The right frame for the AI Optimization Guide is compliance floor. Crawlable, parseable, indexable, mirroring the foundations that earn a page eligibility for any retrieval system. That floor is necessary and it is also the entirety of what the document describes.
The work above the floor is where the guide deliberately stops. DevClass reported in January 2026 that Stack Overflow questions dropped 78 percent year over year as developers moved their queries to AI assistants, a shift in query patterns the AI Optimization Guide does not address. Press Gazette’s 33 percent referral drop from Chartbeat is the cost publishers pay when they treat the floor as the ceiling, optimizing the page for Search and assuming the AI surface will follow. The prior post on FAQ rich results going away is a related case of reading Google guidance against its incentives, rather than at face value.
Above-the-floor work has at least three named threads. Answer-capsule shape for passage retrieval, so the model can pull a clean 40 to 60 word answer without inference. Source diversification beyond Google properties, because AI systems sample from a wider corpus than the SERP. Citation tracking as a first-class metric, treated with the same dashboard weight as keyword rankings used to get.
Stop quoting the guide as strategy
The action for SEO leads is concrete: stop quoting the AI Optimization Guide in client decks as if it were the strategy. Cite it as the compliance baseline and then write the strategy doc the guide is structurally incapable of writing, because that doc is the one that names specific retrieval inputs and how to optimize for them.
The cleanest receipt for the pattern is the click-data verse. Google said for years that clicks are not a ranking factor, and Pandu Nayak and Eric Lehman confirmed NavBoost uses click data under oath in the DOJ trial. That sequence is the template: a public statement that minimizes an input, a later confirmation that the input is active. What would change the pattern is a Google document that names a specific AI retrieval input and explains how a publisher can optimize for it. Until that document exists, the AI Optimization Guide is a floor.
The four verses, links, domain authority, clicks, and now AI, share one move: a public document that frames the audience’s suspicion as a thing not worth optimizing for. The reader’s job is to build the strategy above the floor, because the floor is the only thing Google’s document was ever going to describe.