Is SEO dead now that people ask AI instead of Google?

The mechanism is documented, not speculative. When an assistant answers a question about current businesses, it runs a retrieval step and reads results. OpenAI publishes its crawlers separately and they do different jobs: GPTBot gathers training data, OAI-SearchBot builds search results, and ChatGPT-User fetches a page because a user's question sent it there.

That separation is the practical finding. A site that blocked GPTBot to stay out of training may be entirely available for search. A site that blocked everything is invisible to the surface it wanted to appear on. And because robots.txt controls crawling rather than indexing, the file is the wrong instrument for several of the things people point it at.

On Google's side, the requirement is explicitly the same one as before. Its AI features documentation states that there is no additional technical requirement to be eligible for AI experiences in Search beyond the ordinary ones — be crawlable, be indexable, be useful. There is no AI schema to add and no submission form.

So the retrieval target changed shape; the requirement to be retrievable did not. What died is the assumption that a position on a results page is the only way to be found.

What we will not tell you is by how much your organic traffic will fall. The figures circulating for that come from vendor studies, not from primary measurement, and this page does not launder them into facts.

It raises the cost of not having a good one. An assistant that recommends a business has to read something to decide; the site is the document you control in that set.

Our own measurement, stated with its limits. On 2026-10-01 we ran 68 questions in this market across four assistant surfaces and recorded which domains were cited. Our site was cited zero times — including on the twelve questions we had already published answers to. That is a real number from our own instrument, and it cuts against us as much as anyone: publishing a page is not the same as being cited, and we will not pretend otherwise while the figure reads zero. The two questions with the fewest cited domains anywhere in the bank — seven distinct domains each — are both already published and both still read zero. A thin field is not the lever either.

What follows from that, honestly. A website is still the artifact: assistants cite documents, and a business with no document has nothing to be cited. Retrievability is a precondition, not a strategy — crawlable, indexable, and readable without JavaScript execution is the floor. Specificity is what gets quoted: a page with a price, a date and a named standard gives an assistant something to lift, and a page of adjectives does not. And nobody can sell you a recommendation, because there is no submission form and no placement to buy.

The claim we are not making: that doing all four earns citations. We are measuring that hypothesis on our own site in public, and at the last reading it has not paid off yet.