How to get your website recommended by ChatGPT: what we measured on ours

ChatGPT usually finds tools by searching as it answers. What decides whether it names yours, from our own site’s numbers, and what no file or setting can do.

ChatGPT names websites in two ways. Some it knows from training. For a question like “is there a tool that checks X”, it usually searches the web as it answers, then reads and cites what it finds: most of the answers we recorded cited their sources. So being recommended is mostly about being found by that search, and being a clear answer to the exact question asked.

We measured this on our own site. In the week to 14 November 2026, ChatGPT’s links brought us 96 visits and 124 scans, more than every other source together. Yet when we asked it the 25 questions our tools answer, in the words people type, it named us in 4. Here is what we learned from both numbers.

1. Let ChatGPT’s search crawler in

OpenAI runs separate crawlers for separate jobs, and documents them in its overview of OpenAI crawlers. OAI-SearchBot is the one that decides whether a site can appear in ChatGPT’s search answers, and OpenAI recommends allowing it. GPTBot collects pages for training. ChatGPT-User fetches a page when someone asks about it, and OpenAI says it doesn’t decide what appears in search.

Blocking GPTBot keeps your pages out of training, not out of answers. Sites often block more than they mean to: in October 2026, of 1,691 popular sites whose robots.txt we read, 23% blocked GPTBot and 9% blocked OAI-SearchBot (the full study). Check your robots.txt for each name, and for a rule that blocks every bot.

2. Be in the index it searches

ChatGPT’s search draws on outside search indexes. Bing’s is the one most often named, though whether Bing or Google counts for more is disputed. A page no index knows can’t be found, however good it is.

Ours is a young domain, and that was our biggest gap. On 15 November 2026, Bing knew 173 of the 5,200 addresses our sitemap lists and had indexed 113. Google indexed 313 of our pages and left out 2,579. What helped us:

  • Bing Webmaster Tools and IndexNow. IndexNow tells Bing and others the moment a page is new or changed; Bing’s URL Submission takes a daily quota of addresses (100 for us) for the pages that matter most.
  • Google Search Console, to see which pages are left out and why, and to ask for the important ones again.

What decides it in the end is slower: links from other sites. A new domain with few links gets crawled little, and no tool changes that.

3. Answer the exact question, on one page

The four questions ChatGPT names us for (“how can I tell if a website was built with AI”, a tool to check it, Bolt, and an API for it) are ones a page of ours answers in those words. For Lovable, v0, Base44 and Replit, it explained the checks itself and cited other guides, though we have guides for two of them. A page that answers is needed, but it isn’t enough: it also has to be found (above) and named in places ChatGPT trusts (below).

So: one page per question people ask, with the answer in its first lines, in the words they use. Google’s autocomplete shows those words: type the start of the question and read the suggestions.

4. Get into the lists it cites

When ChatGPT recommends software, studies by marketing agencies find that the tools’ own sites make up roughly half of the pages it cites, and “best X tools” lists and roundups about a fifth (Onely). Their figures differ and are indicative, but they agree on the shape: a tool named in the lists ChatGPT reads is a tool it can name. Across our 25 questions it named Wappalyzer and BuiltWith 7 times each, and us in none of the tech stack questions.

5. Make the card beside your name yours

When ChatGPT first recommended our AI Website Detector, on 28 October 2026, it showed another detector’s image beside our name. The page it linked named no share image of its own, its title didn’t carry our name, and our structured data named no logo: nothing told search engines which image was ours.

Every page of ours now names its own image, and our structured data names our logo and the site’s icons. It’s worth checking yours: a page’s og:image, a logo in its Organization structured data, and an icon that search engines can fetch. We can’t say yet how quickly assistants pick the change up; we check it in our monthly run.

6. Measure it, with the same questions every month

Answers vary from run to run and from place to place, so one question asked once tells you little. We keep 25 fixed questions, asked in fresh chats while signed out on the first Monday of each month, and note whether we’re named, linked, and in what position. ChatGPT also tags its links with utm_source=chatgpt.com, so your analytics can count the visits it sends; ours is how we know it sends more than any other source.

What no file or setting does

  • No file makes ChatGPT recommend a site. An llms.txt helps AI tools read a site, but the big AI companies haven’t said they use it, and it doesn’t put a site in any index. If you have one, our llms.txt checker checks its format.
  • No one controls its answers from outside. An agency that promises a ranking in ChatGPT is promising something it can’t deliver.
  • Being recommended once isn’t being recommended always. Answers vary; what you can move is how often, over many questions and months.

Check your own site

The AI Crawler Checker reads your robots.txt and shows which AI crawlers you let in, for training and for answers apart, beside an AI readiness score of what AI tools can find, read and use on your pages, with the evidence for each check.

Put it to work

Something here out of date or wrong? Tell us and we’ll fix it, with the date of the change.