Google says the new SEO is just SEO.

A guide that asks for less, not more

For two years, marketers have been told that generative search demands a new discipline. It has acquired new names — answer engine optimisation, generative engine optimisation — and with them a new list of things to buy: machine-readable files, expanded schema markup, content rewritten into fragments an AI can digest.

In May 2026, Google published its own guidance on the subject. The document is notable less for what it asks site owners to do than for what it tells them to stop paying for.

Its position is stated plainly. Optimising for generative AI in Search is optimising for Search. It is, in Google's framing, still SEO — and the company reinforced this by folding the advice into its existing documentation rather than creating a separate discipline alongside it.

The things Google says you do not need

The guide contains a section on what site owners do not need to do. It is unusually direct for a company that normally leaves room for interpretation.

New machine-readable files, including llms.txt, are named specifically. Google says such files will neither harm nor help a site's visibility. Structured data is addressed next: it is not required for generative AI search, and there is no special schema.org markup to add for it.

Content chopping is dismissed too. Site owners do not need to break pages into small pieces so that AI can understand them better, and they do not need to write in a particular register for generative search. Pages, the guidance says, should be made for the audience rather than for the machine.

The final item is the most pointed. Pursuing mentions for their own sake is described as unproductive, because the systems weigh high-quality content above manufactured references.

The independent evidence agrees

Google's word alone would be worth treating carefully, since the company has an interest in site owners not optimising for its competitors. On the most contested item, however, the independent data has arrived and points the same way.

Ahrefs analysed 137,000 sites carrying an llms.txt file and found that 97% of those files received no traffic at all during May 2026. SE Ranking examined roughly 300,000 domains and found no statistically significant relationship between having the file and how often a domain was cited in AI answers.

More tellingly, AI crawlers do not request the file on domains where it is absent. They are not looking for it. As of the first quarter of 2026, no major AI company — OpenAI, Google, Anthropic, Meta or Mistral — had publicly committed to reading it in production.

Google's John Mueller addressed it directly on 2 June 2026, calling the practice "purely speculative" and comparing it to the keywords meta tag: a field site owners diligently filled in that no search system ever read. His suggested test was pragmatic. When an AI platform that actually sends you clients complains it needs the file, make one.

What the guide does ask for

The affirmative half is shorter and considerably harder to sell. Produce unique, useful material with a point of view. Go beyond commodity information that already exists in a hundred places. Use good images and video alongside the text. Ensure pages are crawlable, indexable and technically sound. Use semantic HTML. Keep business information accurate in Google's own products, including Business Profile and Merchant Center.

None of that can be delivered as a one-off technical retrofit, which may explain why it features less prominently in the sales material of firms offering AI-search packages.

Google's strongest claim sits in the same passage. Content that people find genuinely useful will influence a site's presence in generative search more, over time, than any other suggestion in the guide. That is either a platitude or the entire strategy, depending on whether a company is prepared to act on it.

Why the myths persist

The gap between the guidance and the market is not hard to explain. A checklist can be scoped, priced and invoiced monthly. Editorial judgement cannot, and it produces results on a timescale that makes clients nervous.

There is also a legitimate objection to taking Google at its word. Google speaks only for Google, and buyers increasingly ask their questions of ChatGPT, Perplexity and Gemini, each of which sources differently. Advice tuned to one system is not automatically advice for all.

But that objection cuts against the checklists as well, and the SE Ranking data covers citations across AI systems generally rather than Google alone. There is no published evidence that an llms.txt file changes what any system says about a business. There is considerable evidence that what other people publish about a company does.

How to spend the budget instead

For a business deciding where to put a limited budget, the guidance narrows the field usefully.

Money spent making pages faster, clearer and genuinely informative is defensible under every version of the advice, from Google's guide to the most aggressive GEO consultancy. Money spent on files that Google has publicly described as inert, and that studies show are not being read, is not.

The order of operations that follows is unglamorous. Fix crawling and indexing first, because nothing else functions without it. Make sure the business is described accurately and consistently everywhere it appears. Write the two or three pages that answer the questions buyers actually ask, in detail, with evidence. Then, if budget remains, experiment with the speculative work — knowing it is speculative.

That was the work before the AI features arrived. Google's guide, and now the independent data, are a statement that it still is.