How to get your business recommended by ChatGPT

Updated 21 September 2026

Ask ChatGPT for the best option in your category and it names a handful of businesses. Whether yours is among them comes down to three things you can check and mostly fix: whether the assistant can reach your site, whether what it finds there answers the question, and whether the other sources it trusts mention you.

Where the answer comes from

An assistant draws on two sources. One is what the model learned in training: a large snapshot of the web, frozen at a cutoff date. The other is what it looks up while it answers. With search switched on, ChatGPT, Perplexity and Claude run searches, read a few pages and cite them.

The two can disagree about you. A business that launched last year may be unknown to the model but easy to find with search. An older one may be well known to the model and still missing from the pages a live search turns up. That is why Noksaro reports both: what the assistants say with web search on, and what they say with it off.

They also respond to different work. Live search rewards pages that exist, can be fetched, and answer the question today. Training rewards being mentioned widely, over time, by sources other than yourself.

1. Let the right crawlers in

Each AI company runs separate crawlers for separate jobs, and you allow or block each one in robots.txt. OpenAI's crawler documentation is the clearest example:

• OAI-SearchBot finds pages for ChatGPT's search answers. OpenAI says sites that opt out of it will not be shown in ChatGPT search answers, apart from navigational links.
• GPTBot collects content that may be used to train future models. Blocking it does not remove you from search answers.
• ChatGPT-User fetches a page when someone using ChatGPT asks it to. Because a person started the request, OpenAI notes that robots.txt rules may not apply.

Anthropic runs the same three: ClaudeBot for training, Claude-SearchBot for search and Claude-User for requests from people using Claude. Perplexity runs two: PerplexityBot for its index and Perplexity-User for user requests.

The full list, with what each one does and what blocking it costs, is in which AI crawlers to allow.

A common mistake is a blanket rule that blocks every AI crawler, and so shuts out the search crawlers along with the training ones. If you want to be recommended, the search crawlers have to be allowed:

User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

Whether to allow the training crawlers is a separate decision. Allowing them lets future models learn that you exist; blocking them keeps your content out of training. Neither choice decides whether you can appear in search answers.

robots.txt is not the only gate. Bot protection at a CDN or firewall can turn these crawlers away while robots.txt says they are welcome, so look for them in your server logs, not just in the file. OpenAI notes that a robots.txt change can take about a day to reach its search systems.

2. Serve pages an AI can read

The major AI crawlers read the HTML your server sends and do not run JavaScript: when Vercel and MERJ measured them, none executed it. A site that builds its content in the browser can look empty to them, even while it ranks well on Google, which does render JavaScript.

The quick test is to open your important pages with curl, or with "view source" rather than the browser's inspector, and check that the text you care about is in what the server returns.

Then make sure a page exists that answers the question being asked. Assistants recommend what they can quote. For each question your buyers really ask, such as "best X for Y", "X near Z" or "X or Y", you want a page that answers it early and specifically: what you offer, for whom, where, at what price, and how you compare. Vague marketing copy gives an assistant nothing to cite.

Structured data helps too. Organization and Product markup with your real prices, locations and contact details states plainly what the page already says, so a machine does not have to guess.

3. Be mentioned where your buyers already look

When an assistant searches for "best X", it reads what ranks for that: comparison sites, review platforms, forums, roundups and directories. If those pages name three competitors and not you, the answer usually will too, and no work on your own site changes what other sites say.

Start by finding out which domains the assistants cite for your questions; Noksaro's report lists them under "who gets cited instead". Then work through them: get listed where listing is possible, earn a place in the roundups that matter, and take part honestly in the communities where your buyers ask for advice. Those mentions also become part of what future models learn about your category.

4. Measure, change one thing, measure again

The same question asked twice can get two different answers, so a single screenshot proves little either way. What tells you something is the share of answers that name you, across a fixed set of questions, tracked over weeks. Change one thing, give the crawlers time to catch up, and compare.

What does not move it

• llms.txt. No major AI provider has committed to reading it, and large studies have found no link between having one and being cited. What the evidence shows.
• Repeating the question on your pages. Assistants are looking for an answer they can quote, not a phrase repeated five times.
• Fixing only your own site. It is necessary, but on its own it is rarely enough, because the pages doing the recommending in step 3 belong to someone else.