How to rank in ChatGPT: get your brand named in its answers

Most advice on ranking in ChatGPT is a list of twenty tactics. The mechanism is simpler than that, and once you see it the order of the work follows from it.

How a ChatGPT answer gets built

When a question needs current information, ChatGPT searches the web before it writes. It sends a handful of searches of its own, reads some of the pages that come back, and writes an answer from them. The brands it names are, by and large, the brands those pages name. Its searches are called fan-out searches, and they are often longer and more specific than what the person typed.

You can see this in the data. We asked ChatGPT eight questions a buyer of an AEO tool would ask and kept every answer (the sample, as a CSV). Asked for the tools that track whether a brand shows up in ChatGPT, it ran one search, "best AEO tools track brand visibility mentions in ChatGPT answers 2026 Profound Otterly Peec AI Ahrefs Brand Radar", and named exactly the five products its sources listed: Peec AI, Profound, Otterly.AI, Ahrefs Brand Radar and Semrush's AI Visibility Toolkit. The pages it cited were those vendors' own sites and help centres.

Two kinds of question get answered from two kinds of page:

• Buying questions ("which tool should I use", "cheaper alternatives to X") are answered from roundups, review sites and the vendors' own product pages. Across the four buying questions in our sample, ChatGPT cited roundups like Zapier's and the vendors' help docs.
• Fact and how-to questions ("what is AEO", "how do I get mentioned") are answered from explainers. Asked how to get a brand mentioned in ChatGPT, it named no tool at all and cited OpenAI's own documentation, Search Engine Land and a Semrush article.

One condition comes before all of it: ChatGPT can only use your pages if its search crawler may read them. OpenAI says that sites which opt out of OAI-SearchBot "will not be shown in ChatGPT search answers", apart from plain navigational links. Check your robots.txt for OAI-SearchBot before anything else (the full list of AI crawlers covers the other assistants). Blocking GPTBot, the training crawler, is a separate choice and doesn't affect search answers.

The four fixes, in order

Work through these in order for each question where you're left out. Each one is cheaper and faster than the next, and a later fix rarely works while an earlier one is still broken.

1. Fix what it gets wrong

Read the answers that mention you before you try to be mentioned more. A wrong price, an old product name or a feature you dropped two years ago will be repeated every time you're named. Find where the wrong fact comes from (the answer's citations usually show it) and correct it there, as our guide to fixing wrong AI information sets out: your own page first, then the third-party page, with a polite note to its author.

2. Update a page it already reads

If ChatGPT cites one of your pages but doesn't name you for the question, the page is being read and isn't saying the right thing. Put the direct answer to the question in its first two or three sentences, in the buyer's words. Add the specifics an answer needs: what you sell, who it's for, the price, how you differ. This is the cheapest change with the fastest effect, because nothing has to be discovered.

3. Write the page it's missing

If rivals' pages answer the exact question and none of yours does, write one. Answer first, then the evidence (a table, your own data, a dated source), then the detail. One page per question. A page that tries to answer ten questions answers none of them as directly as a page written for one.

4. Get onto the pages it cites

For buying questions, the pages that matter most are usually not yours. If ChatGPT names three rivals from a roundup that doesn't list you, the work is getting into that roundup: a factual one-page sheet for the author, a free account so they can try the product, a correction where they have your facts wrong. Review sites and community threads count too; our guide to getting on the lists AI cites takes each kind in turn. This is the slowest fix, so it comes last, but on category questions it's often the one that decides the answer.

How to tell if it worked

Answers vary. Ask ChatGPT the same question three times and you can get three different lists of brands, so one answer before and one after proves nothing. To know whether a change worked:

• Ask each question several times a week, on each assistant, and count the share of answers that name you, rather than reading one answer.
• Compare the questions your change targeted with control questions you didn't touch. If both rise, the market moved, not your page.
• Wait. Answers change as pages are re-crawled. We read results at day 21 and again at day 35 after a change goes live; proving AEO results has the whole method, and measuring AI visibility the weekly numbers under it.

Here is what that looks like in Keats, on Prepco, the made-up brand in our demo workspace. A comparison page aimed at a set of buying questions moved them +35 points by day 35, while the questions it didn't touch moved +2 points, against a noise band of 12 points.

Expand

The difference between targeted and untouched questions is the result. Without the control, you'd credit your page with a rise that the whole market shared. Our guide to AEO, SEO and GEO covers what to measure and why it differs from rankings.

What doesn't work

• Writing for prompts nobody asks. Some keyword tools list phrasings like "most reliable AI search optimization tool for data accuracy" with hundreds of searches a month. Many look like AI trackers' own prompts run through Google. Write for the questions buyers ask: your sales calls, your Search Console queries and the forums your buyers read are better sources, as our guide to choosing the questions to track explains.
• A file that promises to fix it. An llms.txt file helps agents find their way around your site. It doesn't make ChatGPT name you.
• Judging by one answer. See above: one answer is a sample, not a rate.

Doing this every week

The loop is: measure the questions, find the gaps, make the fix the order above calls for, report it live, and read the result against the controls. That's what Keats does each week, and it's built so your AI agent can run the loop through Keats's MCP server: it reads the gaps, drafts the work from a brief, and you approve what gets published.