Google Named GEO, Then Debunked Most of It

On 5 June 2026 Google updated its Search Central documentation with an official guide to optimizing for generative AI features on Google Search. For the first time, Google's own docs name the thing: generative engine optimization, GEO, alongside answer engine optimization, AEO. After two years of practitioners arguing about whether the category was real, Google put it in writing. GEO is a thing now, officially.

Then, in the same update, Google quietly torched most of what is being sold under that name.

That is the part worth sitting with. The vendor whose surface everyone is optimizing for just published the list of GEO tactics that do nothing, and it is a long list. It reads like an itemised GEO invoice with most of the line items crossed out.

The tactics Google just crossed out

Google's mythbusting section is blunt. You do not need any of the following to appear in its generative AI results:

• llms.txt. In Google's words, you do not need to create new machine-readable files, AI text files, markup, or Markdown to appear in generative AI search. The Search team had been comparing llms.txt to the old keywords meta tag for months. Now it is in the docs.
• Special schema for AI. Structured data is not required for generative AI search, and there is no special schema.org markup you bolt on to win citations.
• "AI chunking." Breaking your content into bespoke chunks for the model is not a requirement. The chunking-as-a-service pitch is smoke.
• AI-specific rewrites. You do not write a separate, machine-flavoured version of your page. The same page that serves a human serves the model.
• Inauthentic mentions. Manufacturing brand mentions to game the answer engines is a tactic Google warns against, not one it rewards.

Look at that list against the GEO retainers people post about. A "structured data stack". An "llms.txt implementation". An "AI chunking strategy". A "citation gap audit" whose remedies are mostly the four bullets above. Google just published the docs that say those moves are noise.

What Google says actually works

Strip the debunked tactics out and what Google leaves standing is almost insultingly plain. Write unique, genuinely useful content. Ground it in real sources. Keep the structure clean. Make sure the site is crawlable and indexable. Give people a clear experience. From Google's perspective, optimizing for generative AI search is optimizing for the search experience, so it is, in their phrase, still SEO.

That is the advice nobody can build a retainer around. There is no proprietary file to sell, no markup stack to install, no chunking pipeline to license. It is the work, done properly, the way it was always supposed to be done. Which is exactly why it is uncomfortable for a chunk of the GEO market, and exactly why it is correct.

None of what Google torched was ever the MX claim

Here is where I have to be honest, because I build in this space and I have a dog in this fight.

You might expect Machine Experience to be on Google's debunked list. We talk about machine-readable content. We serve an llms.txt. We care about structured data. Surely Google just torched us too?

No. Read what we actually wrote.

On llms.txt, our position has been the unpopular one for a while: llms.txt probably isn't working. It was never a Google ranking trick, and we said so. It is a convenience for the agents that choose to read it, not a signal Google Search promised to honour. Google saying "this does nothing for our generative results" is not a correction of MX. It is the same thing we said, from the other side of the fence.

On schema, our line has been that structured data is necessary and nowhere near sufficient. Schema.org keeps growing, and the provenance layer still does not exist. Google deprecating rich-result types that publishers gamed, and now telling you not to obsess over markup, lands on the same point: schema describes a claim, it does not evidence one. We have never told anyone to stack schema and call it a strategy.

On "AI chunking" and AI-specific rewrites, MX is the opposite of that. The whole argument is that one well-built artefact serves every reader, human and machine, with no special machine-flavoured fork. Google's "the same page serves both" is our Convergence Principle in Google's own words: a page an agent can read accurately is a page a screen reader handles cleanly and a person understands without a handover.

So when Google torched the GEO surface tactics, it did not torch MX. It cleared the floor that MX has been building on the whole time.

The one thing Google's guide structurally cannot give you

"Still SEO" is true, and it is also the tell. Google's guide is about the page, on Google's surface, found through Google's index. It is silent on the file once it leaves that surface, because Google Search is not in the business of telling an arriving agent who published a document, when, and whether it has been altered since.

That is the gap. The four bullets Google left standing get you found and get you read. They say nothing about whether the thing reading you can prove where you came from. When your contract is lifted into a procurement portal, your policy file into a regulatory submission, your article into a training corpus, the page is gone and the Google index is not in the loop. What travels with the file is whatever the file carries on its own.

That carried layer is provenance: who published this, when, under what version, and a way to check it has not changed. MX is the discipline that puts it in the file. Reginald, currently in beta, is the registry that makes it verifiable. MX makes content machine-readable. Reginald makes it machine-trustworthy. Google's guide, by design, does neither, and never claimed to.

This is also where the regulatory weather is heading. The EU AI Act, the European Accessibility Act, and the digital-records rules arriving alongside them expect organisations to produce documentation that is structured, logged, and verifiable on request. MX and Reginald do not grant compliance with any of them; that stays a legal duty of the organisation. They make the documentation that organisation has to produce tamper-evident and checkable, which is precisely the part a search optimization guide is never going to cover.

Note: This page describes regulatory frameworks in general terms only. Nothing here is legal advice. Requirements vary by jurisdiction, organisation type, and use case. Consult qualified legal specialists for guidance specific to your situation.

So what changed for the people who actually write?

Less than the invoices imply, and that is the good news.

If you were already writing unique content, citing real sources, and keeping your structure clean and crawlable, Google just told you that you have been doing GEO correctly the entire time. Keep going. Do not buy the chunking pipeline.

If your visibility plan was a stack of the tactics Google just crossed out, you were renting it, on a lease Google can cancel in a docs update, which it just did. The same lesson runs under the platform churn we wrote about in Game the Signals, Lose the Engine: the surface advice contradicts itself every quarter because it tracks one product team's release notes, while the machine layer underneath does not move.

And if you want the part Google's guide cannot reach, the part that survives the next system-prompt rewrite and travels with the file when the page is nowhere in sight, that is the layer to build now. Not "how do I rank in this AI system this quarter", but "what does my content carry so that any machine, anywhere, can prove what it is".

Google named GEO and debunked most of it in the same breath. The unglamorous floor it left standing is real, and it is the floor MX builds on. The provenance layer it left untouched is the part worth owning. Found, understood, used: Google just confirmed the first two and stayed quiet on the third.

If you want the longer argument, it is in GEO is a tactic, MX is the specification, and the standard itself is debated in the open at tg.community.