Foundations

Google says nothing special is needed for AI search. Read the rest of the sentence.

By Arnav Mukherjee, founder of TofuBofu · August 25, 2026

Someone asked Google's John Mueller on Bluesky whether there are industries where GEO simply does not matter yet. He answered in one sentence, and by the time the trade press picked it up four days later that sentence was doing duty as a verdict on a whole category.

I read the thread at source rather than the write-up, which is how I know the exchange happened on 20 August 2026 and not on the 24th, and how I know exactly what he was asked. Both details change what his answer covers.

Here is my position before the evidence, so you can hold me to it. Mueller is right, Google's documentation backs him, and none of it settles the question, because the sentence is about Google Search and Google Search is one of the six engines your buyers actually ask.

What he was asked, and what he said

The question, posted at 07:00 UTC on 20 August 2026: "@johnmu.com , are there industries where GEO simply doesn’t matter yet? I’m thinking about sectors like adult, gambling or other areas where Google still seems to drive most discovery."

Mueller's reply, two hours and forty-eight minutes later: "I'm not quite sure what you're asking; from our POV there's nothing really special you need to do for generative AI responses in search."

Read the qualifiers, because they are the whole content. From our POV. He is speaking for Google, which is the only company he can speak for. In search. Not in AI assistants, not in chat products, in search. And the questioner had already framed the field as one where Google still seems to drive most discovery, so the answer lands inside a premise that was never tested.

He also said he was not sure what he was being asked, which the follow-ups confirm: the questioner came back twice trying to get at whether Google processes those queries differently underneath. That is a man giving a careful, narrow answer to an unclear question, not a company issuing a ruling on an industry.

Google's documentation says the same thing, and explains why it has to

This is not an off-the-cuff remark that contradicts the written record. Search Central's AI optimisation guide, last updated 10 July 2026, states it as policy: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

And on the files half the category is selling: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." If you are paying anyone to generate a file so that Google's AI can read you, that sentence is your refund request.

The reason sits one paragraph away, and it is the sentence that bounds the whole claim: "Our generative AI features on Google Search are rooted in our core Search ranking and quality systems. These features rely on AI techniques to highlight content from our Search index." The input is the index. Of course there is nothing extra to do. Google is describing a pipeline whose front door you already know how to walk through. That is a fact about Google's plumbing, and people are reading it as a fact about the market.

Two Google surfaces, one corpus, and they do not agree

If everything Google runs came out of the same index, the two Google surfaces we query would answer buying questions in roughly the same way. Across 46 reports, read on 12 August 2026, they do not. Google AI Mode named the scanned brand on 4.0 percent of buying questions. Gemini named it on 0.4 percent.

Say the caveat before anyone else does, because this argument rests on the gap. Our Gemini number is a floor, not a clean rate. Two faults of our own pushed it down: the model spent most of its token budget reasoning and truncated mid-answer at the budget we ran, and until revision 00203 our brand matcher could rule a mention out but never rule one back in. The true figure is higher than 0.4 percent and we have not measured how much higher, which is why you will not see a multiple in this paragraph.

What survives that caveat is the shape. AI Mode sits inside Google Search and is covered by Mueller's sentence. The Gemini app is a separate product with a separate retrieval path, and it is not. Same company, same week, same questions, and the answer to who should I hire comes back differently. Anyone telling you that ranking on Google is the whole job has to explain that gap, and cannot do it from the index.

Find out which of the six actually names you

One URL, real buying questions, all six engines including the two Google surfaces argued about above. Free, and it takes about a minute.

Run the free scan

Four of the six are not Google at all

We put buying questions to six engines on every plan, including the free one. Exactly one of them, AI Mode, is Google Search. Across the same 46 reports, on buying questions only, the rates run: Perplexity 7.4 percent, Google AI Mode 4.0 percent, Bing Copilot 3.1 percent, ChatGPT 1.7 percent, Claude 1.2 percent, Gemini 0.4 percent.

Those are low numbers and they are supposed to be. They count buying questions only, the ones where a buyer asks who to hire, and they exclude the probes that contain the company's own name, which every engine answers with that company most of the time. We published inflated versions of these figures for weeks because our denominator included the second kind, and correcting that cut them by roughly two and a half times. If a competitor's dashboard shows you a number several times higher than these, ask what is in its denominator before you celebrate.

Two of those figures describe our instrument as much as the engine, and we would rather say so than let you over-read them. The ChatGPT number predates 18 August 2026, the day we forced that probe to search the web; before that it answered from training weights, so 1.7 percent measures what a model remembered, not what ChatGPT finds. And our Claude and Gemini probes still carry no search tool at all, which I confirmed in our own source this week. A brand that published something last month cannot possibly be found by a probe that never looks.

That admission is the point rather than a hedge. The honest version of this category is that everyone's numbers are partly a measurement of their own rig, and almost nobody will tell you which parts. Mueller's sentence is a statement about one engine's pipeline. Our numbers are a statement about six engines and one measuring instrument. Neither is a market rate, and anyone selling you one is not measuring, they are rounding.

What to actually do with it

Take Google literally, in both directions. Stop building Google-specific AI files. Google says it does not read them, in writing, on its own domain. Keep doing the index work, because for Google that genuinely is the whole job, and a site that cannot be crawled and cannot be understood is not eligible for anything downstream.

Then measure the five surfaces the sentence does not cover. Ranking on Google buys you nothing on Perplexity by itself, and Perplexity is the engine that named brands most often in our corpus. A single-engine check is evidence about one engine, and a blended score across engines that disagree this much is arithmetic performed on incompatible things.

Our position has not moved and this exchange does not move it. SEO is the floor. AEO is a distinct layer you can win even when your Google rank does not move, and the reason is sitting in Google's own documentation: the generative features are rooted in Search, and five of the six answers your buyer might read are not.

Frequently asked questions

What did John Mueller actually say about GEO?

On Bluesky on 20 August 2026, answering whether there are industries where GEO does not matter yet, Google's John Mueller wrote, verbatim: I'm not quite sure what you're asking; from our POV there's nothing really special you need to do for generative AI responses in search. Two details matter and both are visible in the thread. The question was about industries, naming adult and gambling as sectors where Google still drives most discovery. And the answer says in search, which scopes it to Google Search. The trade coverage appeared on 24 August, which is the date the exchange usually gets attached to.

Is Google saying AEO and GEO are pointless?

No. Google is saying that its own generative features draw on its own index, so there is no separate Google pipeline to optimise for. Its documentation puts it plainly: our generative AI features on Google Search are rooted in our core Search ranking and quality systems. That is a statement about how one company's plumbing works. It is not a statement about ChatGPT, Claude, Perplexity or Copilot, none of which read Google's index, and it is not a claim that every engine will name you once Google ranks you.

Do I need an llms.txt file to appear in Google Search?

No, and Google states it directly rather than leaving it to inference. Its AI optimisation guide says you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them. Take that at face value for Google. It says nothing about whether other engines or the tools around them read such files, and it is not a reason to skip structured data, which does real work elsewhere.

If Google's AI features use the Search index, why do results differ across Google's own surfaces?

Because not every Google product is Google Search. Across 46 reports read on 12 August 2026, Google AI Mode named the scanned brand on 4.0 percent of buying questions while Gemini named it on 0.4 percent. AI Mode sits inside Search and is covered by Mueller's sentence. The Gemini app is a different product with a different retrieval path and is not. Our Gemini figure is a floor rather than a clean rate, because two faults of our own depressed it, so the real gap is smaller than the arithmetic suggests. The direction is not in doubt: one company, two surfaces, and they do not behave alike.

How much of AI search is Google?

Less than the Google-shaped conversation implies. We query six engines on every plan, including the free one: ChatGPT, Claude, Perplexity, Gemini, Google AI Mode and Bing Copilot. Exactly one of those, AI Mode, is Google Search. On buying questions across 46 reports, Perplexity named the brand most often at 7.4 percent, then AI Mode at 4.0 percent, Copilot at 3.1 percent, ChatGPT at 1.7 percent, Claude at 1.2 percent and Gemini at 0.4 percent. Our ChatGPT figure predates 18 August 2026, when we forced that probe to search the web, so it measures a probe we no longer run.

So is AEO just SEO after all?

For Google Search, Google's own position is that optimising for generative AI search is optimising for the search experience, and thus still SEO. We do not argue with that. Our position is unchanged by it: SEO is the floor, and AEO is a distinct layer you can win even when your Google rank does not move. Both can hold at once because they describe different scopes. The work that gets you into Google's index is the same work. The question of which of six engines names you when a buyer asks who to hire is a separate measurement, and Google does not answer it.

What should I actually do with Mueller's answer?

Take it literally and act on it in both directions. Stop building Google-specific AI files, because Google says it does not read them. Keep doing the index work, because for Google that is the whole job. Then measure the surfaces his sentence does not cover, which is five of the six engines we run, and treat any single-engine check as evidence about one engine only. A tool that reports one blended number across engines that disagree this much is averaging away the thing you needed to know.

Sources and further reading

John Mueller, Bluesky, 20 August 2026. The reply quoted above, read in the thread rather than in coverage of it. The question it answers is two hours earlier in the same thread.
Google Search Central, Optimizing your website for generative AI features on Google Search, last updated 10 July 2026. Source of all three Google quotations here, including the one on machine readable files.
Search Engine Journal, 24 August 2026. The coverage that carried the exchange, four days after it happened. Listed for completeness, not as the source for any quotation above.
Our own scans. Buying-question mention rates across 46 reports, read 12 August 2026, excluding brand-name probes from the denominator. The engine-by-engine figures and their caveats are published on the engine guides.