Google’s own guidance says optimizing for generative AI search is still SEO. That’s accurate, and it’s also not the whole story. The ranking system underneath AI Overviews and AI Mode is the same one that ranks regular search results, so most technical SEO fundamentals still apply. But the layer built on top of that system, the one that decides what gets quoted, summarized, and cited in an AI answer, behaves differently enough that specific content choices measurably change your odds of showing up there. Both things are true at once.

What Google actually said

In May 2026, Google Search Central published its first official guide to optimizing for generative AI features in Search. The core message: AI Overviews and AI Mode are built on Google’s existing Search ranking and quality systems, not a separate one. The guide goes further than a general statement. It specifically tells site owners they don’t need new markup for AI features, that there’s no dedicated schema.org type for showing up in an AI Overview, and that splitting content into rigid segments for easier AI parsing isn’t something Google’s systems require.

That last point matters because it’s where a lot of paid AEO and GEO advice diverges from what Google itself is asking for. Search Engine Journal’s coverage of the guide put it plainly: this is still SEO.

The skepticism isn’t coming from nowhere

Google isn’t the only one making this argument. Forrester’s research on SEO’s shifting role in marketing describes AEO as significantly, but not fundamentally, different from SEO, a useful framing because it doesn’t dismiss the differences, it just refuses to call them a new discipline. CMSWire has reported on SEO professionals watching organic traffic flatten or drop while being sold AEO retainers with no clear measurement attached, and the piece is blunt about vendors overselling what those retainers actually change.

Then there’s llms.txt, probably the single most oversold tactic in this space. John Mueller has compared it to the old keywords meta tag: something site owners add because it feels like it should matter, not because any evidence says it does. Gary Illyes has said the same at a Search Central event: Google doesn’t use it and has no plans to. An Ahrefs analysis of 137,000 sites found that 97% of published llms.txt files were never fetched by anything in a full month. If a tactic doesn’t move the system it’s supposedly optimizing for, it isn’t optimization. It’s a checkbox.

Worth noting: llms.txt isn’t universally pointless, it’s pointless for Google specifically. Perplexity and Anthropic’s Claude do fetch it. That distinction, that a tactic can be dead weight for one platform and genuinely useful for another, comes up constantly once you stop treating AI search as one single system.

Where the still-SEO framing understates things

Here’s where I think the flat nothing-has-changed reading goes too far. A 2024 study out of Princeton, IIT Delhi, Georgia Tech, and the Allen Institute for AI, formalized as Generative Engine Optimization at KDD 2024, tested which content changes actually moved visibility inside AI-generated answers. Adding citations to sources, direct quotations, and specific statistics each produced meaningful gains on their visibility metric, in the range of 30 to 41 percent. Keyword density, the classic SEO lever, did almost nothing in the same tests.

That’s not a contradiction of Google’s guidance. Ranking and being selected for a generated answer aren’t the same event. A page can rank well under ordinary Search signals and still lose out to a competitor’s page when an AI system is choosing what to quote, because the AI is applying a second layer of judgment on top of ranking: is this passage citable, specific, and attributable. That second layer rewards different things than the ranking layer does, even while running on the ranking layer’s output.

There’s a second data point worth sitting with. Ahrefs has reported that a brand’s web mentions correlate with AI citation frequency at roughly 0.664, compared to 0.218 for backlinks, the metric classic SEO has spent two decades optimizing for. If that holds, it means the thing predicting whether you get cited by an AI system isn’t the thing that’s predicted rankings for years. Same underlying search index, different selection criteria on top of it.

And the two are visibly drifting apart in practice. Ahrefs has also tracked the overlap between a query’s top 10 organic results and the sources an AI system actually cites for that same query, and reported that overlap falling from around three quarters in mid-2025 to somewhere between 17 and 38 percent in early 2026, depending on the topic. Ranking well is clearly still necessary. It’s becoming less sufficient on its own.

So what’s the honest answer

Google is right that there’s no separate ranking system to build for, no special schema, no file format that unlocks AI visibility on its own. Anyone selling AEO or GEO as an entirely new technical discipline, with its own infrastructure requirements, is selling something Google’s own documentation contradicts.

But the selection layer sitting on top of ranking, the part deciding what gets quoted once a page has already earned its place in the index, does appear to reward specific, citable, source-backed writing over the kind of thin, keyword-led content that used to be enough to rank. That’s not a new discipline. It’s closer to old-fashioned editorial rigor mattering again, just measured by a different system than backlinks.

Practically, that means: keep doing real technical SEO, because it’s still the entry ticket. Don’t buy llms.txt or AI-specific schema as a fix for a visibility problem Google has explicitly said neither one solves. Do pay attention to whether your content is specific enough, sourced enough, and quotable enough to survive being the thing an AI system picks to cite, because that appears to be where the actual competition has moved.

None of this is guaranteed to change your citation rate or your rankings. Nobody can promise that, and anyone who does is the exact problem this piece is about. What follows from here is testing it directly rather than taking anyone’s word for it, including the sources cited above. That’s what the Research and Experiments section of this site exists to do: run the crawler-access and llms.txt questions raised here as live tests rather than repeating the debate.