How to rank on ChatGPT and AI answer engines

People ask how to rank on ChatGPT because that is the word they have. It is the wrong word, and keeping it costs you the strategy. Search hands you ten blue links and lets you choose. ChatGPT, Perplexity, Gemini and Google’s AI Overviews hand you two or three names in a sentence. There is no position four to climb to. You are named or you are not.

That single difference changes what the work is. Below is what actually decides it, what we can measure, and what does nothing at all.

What decides whether an engine names you?

Two conditions, and they fail independently.

The engine can read your site. A retrieval crawler fetches pages at the moment somebody asks a question. If it is blocked, or the page is empty without JavaScript, there is nothing to quote.
The engine has read about you somewhere else. Models name brands they have encountered on pages that are not yours: directories, comparison lists, review sites, forum threads, press.

Almost every guide on this subject covers the first and stops. The first is necessary and it is not sufficient, and confusing the two is the most expensive mistake in this field. A site can pass every access check ever written and still be invisible, because permission is not discovery.

Condition one: can the engine read you?

This half is mechanical, which is the good news. It is a fixed list, it is finished when it is finished, and nothing about it is a matter of opinion.

Let the retrieval crawlers in. These are the ones that read a page at answer time: OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Claude-SearchBot, Applebot and others. A block here is the one that costs a citation. Training crawlers such as GPTBot and CCBot are a licensing decision, not a visibility one. Google-Extended is not one of those: it also governs whether Gemini may read your pages at answer time.
Do not block the ordinary search crawlers. AI Overviews and Copilot are generated over the indexes Googlebot and Bingbot build. Blocking those removes you from those answers no matter how welcoming your robots file is to everything else.
Serve real text in the HTML. Answer engines do not run your JavaScript. A page that is empty until a script fills it is an empty page to them.
Answer directly under the heading. Retrieval returns passages, not pages. A heading that asks the question, with the answer in the first sentence or list beneath it, survives being chunked. The same answer three paragraphs down does not.
Give every page one subject. When the title, the heading, the URL and the opening line disagree, the engine picks which to believe. That choice should be yours.

None of this is exotic and all of it is checkable. Our own site check runs a hundred of these tests in about half a minute, writes out the fix for each one that fails, and is free for your own site.

Condition two: has anyone else written about you?

This is the half that decides whether you are recommended, and it does not happen on your website.

When an engine answers a question like “best tool for X”, it does not visit each vendor and form a view. It retrieves the pages that already answer that question, which are listicles, directories and review sites, and it names what it finds written there. The pattern is visible in the answers we keep: when the engine fetched a brand’s own site before answering, it recommended the brand, and when it did not, it did not. We wrote that up in read you, recommended you.

So the work is getting written about. In rough order of how often these pages turn up in stored answers:

Where	Why an engine reads it
Product directories	Quoted constantly for “best tool for” questions
Alternatives listings	The page engines read for “alternatives to X”
Review sites	The source of the word “verified” in an answer
Company registries and profiles	How a model decides you exist as an entity
Forum and community threads	Retrieved heavily for honest-opinion questions
Original data somebody cites	The one you cannot buy, and the one that lasts

Being on those pages is slower than editing your own markup and it is the part that moves the answer. There is no version of this where a perfect website skips it.

Why does the same question give a different answer twice?

Because it does, and any measurement that ignores this is noise. Ask the same model the same question six hours apart and the list of names changes, because the search results underneath it changed. We ran that test and not one answer came back identical.

Which means a single check tells you almost nothing. What tells you something is the same prompts, on the same engines, repeated, with the result stored each time. One reading is an anecdote. A series is a measurement.

What does not work

Publishing llms.txt and waiting. It is a convention, no engine has committed to reading it, and it changes nothing on its own. Add it because it is cheap, not because it is a plan.
Keyword stuffing for a machine that reads meaning. These models read sentences. Repetition reads as noise and always did.
Schema that does not match the page. Marking a page up as something it is not is a manual-action risk rather than a shortcut. Invented FAQ blocks are the usual example.
Chasing the abbreviation. Very few buyers search for AEO. They search for the thing they want, in the words they already use.
Optimising once. The engines re-answer every time they are asked. Visibility is a position you hold, not a box you tick.

Where to start this week

Check the access half and fix whatever is broken. It is finite and it is quick.
Write down the ten questions a buyer would actually type before choosing something like yours. Those are your prompts, not your keywords.
Get listed everywhere in the second condition above. This is form filling, and it is the highest-value week of work available to a new brand.
Measure the same prompts again in a fortnight and compare, rather than trusting one reading.

The order matters. The first step is necessary, the third is what actually changes the answer, and the fourth is the only way you will ever know whether it did.

Questions

Can you rank number one on ChatGPT?

No, because there is no numbered list to be first in. An answer engine names two to five brands in a sentence and leaves everything else out. The only outcome worth measuring is whether you are named at all, and for which kind of question.

Does llms.txt make ChatGPT recommend me?

No. It is a convention some sites publish so a crawler can find the important pages quickly. No answer engine has committed to reading it, and publishing one changes nothing about whether you are recommended. It is cheap to add and it is not a strategy.

How long does it take before an AI names my brand?

For a new site, weeks before search engines index you and a month or two before citations elsewhere are old enough for engines to retrieve. Nothing on your own website shortens the second half, because the second half happens on other people’s websites.

Is answer engine optimisation the same as SEO?

They share the plumbing and not the goal. Search ranks pages and gives the reader ten of them. An answer engine names a few brands and gives the reader no list at all. Everything that stops a crawler reading your site hurts both, which is why Oftheard does the SEO first and checks the AI answers alongside it.

How to rank on ChatGPT, and what ranking even means there