Google published the AI optimisation rules and they kill three popular tactics

Google shipped a guide in May 2026 called Optimising for generative AI features. It is short, it is free, and almost nobody in marketing has read it. It also contradicts three tactics that are currently being sold as essential.

If you are reviewing an SEO proposal this quarter, read the guide first. It will save you an argument.

What it says, and what that removes

Structured data is not required for generative AI search. The guide states this outright. Schema markup did not stop mattering, but it stopped being the AI citation lever it is often presented as. If a scope of work justifies six weeks of schema implementation primarily on AI visibility grounds, the justification is wrong. There are good reasons to implement schema, covered further down, and this is not one of them.

There is no requirement to break your content into tiny pieces. An entire cottage industry grew up around chunking content for language models, restructuring pages into short, self-contained fragments with headings every 80 words. Google says it is not needed. Retrieval works on your normal page structure.

Manufacturing brand mentions across the web is not as helpful as claimed. The tactic of paying for scattered unlinked mentions to boost AI visibility does not do what the pitch says.

llms.txt files are not used by Google Search. Named directly.

Four checklist items, gone, from the platform itself.

What the guide says to do instead

The advice that remains is unglamorous, and that is the point.

Be indexed and eligible for a snippet. This is the eligibility gate, and it is stated explicitly. If a page cannot show a snippet in normal search, it cannot appear in AI features. That makes snippet directives, nosnippet, data-nosnippet, max-snippet and noindex, the only real controls you have over AI inclusion. Most large sites have never audited them.

Write for people, with unique value. Standard advice, except the guide sharpens it with a worked example of what not to publish. It names commodity content, using something like 7 Tips for First-Time Homebuyers as the illustration. That is not a random example. It is the exact shape of content most content calendars are full of.

Understand that retrieval is grounded and queries fan out. The guide names retrieval-augmented generation and describes concurrent, related queries. Your page is not being matched to the search someone typed. It is being matched to a set of sub-questions the system generated from it.

Why the tactics were popular anyway

None of the four dead tactics came from Google. They came from the marketing industry and were then attributed to Google after the fact.

They share a profile. Each is quick to implement, produces a visible artefact, and has no measurable outcome. That combination is commercially irresistible for a supplier who needs to show progress in month one, and it is very hard for an in-house marketing lead to challenge, because challenging it requires knowing the primary source exists.

There is a second reason. Much of the AI SEO statistics content ranking right now is itself AI-generated recycling of other AI-generated posts. When a claim appears in fifteen articles, it reads as consensus. Trace fifteen articles back and they often collapse to one unsourced original.

Where schema still earns its budget

Dropping schema entirely would be an overcorrection, so be precise about why you keep it.

Structured data still drives the rich results that survive: product listings with price and availability, review snippets, breadcrumbs, video, events, recipes, job postings. Those render, they take up space on the results page, and they influence click-through on the searches that still produce clicks.

Organization, LocalBusiness and sameAs do a different job. They state who you are in a machine-readable way and connect your entity to the profiles that represent it. That is entity disambiguation, and it matters most for companies whose name collides with something else. It is not a snippet play and it is not an AI citation play. It is identity.

Keep schema for those reasons. Drop it from the AI visibility line of your reporting.

What actually moves AI visibility

The guide implies the work rather than listing it, so here is the practical translation.

1. 1.Fix eligibility first. Crawl for nosnippet and max-snippet directives across every commercial template. Fix indexation on pages that matter. This is a one-week technical job and it is the only documented on-off switch in the whole discipline.
2. 2.Map the sub-questions. For each commercial topic, write the ten questions a buyer actually asks. Then check whether any page you own answers each one directly. Fan-out means you are competing on those, not on the head term.
3. 3.Answer in passages that stand alone. Not fragments. Paragraphs that make sense when lifted out of context, because that is how they get retrieved. State the answer, then qualify it, then give detail.
4. 4.Publish the thing only you can publish. Your own data, your own project outcomes, your own methodology, your own pricing logic. Commodity explainers are named as the failure case. Original material is the only durable defence against a system that can already write the generic version itself.
5. 5.Measure citation, not just rank. Fix a prompt set of 50 to 100 real buying questions, run it monthly, record whether you appear. Averages across a rank tracker will not show you this.

The honest limitation

Following every word of Google's guide does not guarantee citation. The guide describes eligibility and quality signals, not a mechanism you can force. Retrieval is probabilistic, answers vary run to run for the same prompt, and a rival with a weaker site can beat you on one sub-question because they happened to answer it more directly.

What the guide does give you is a clean line between work that has a documented basis and work that does not. In a discipline currently full of confident invention, that line is worth more than another tactic.

Read the primary source before you fund the interpretation of it.