How do you get recommended by ChatGPT?
ChatGPT recommends a business when its search finds a page that answers the question asked, and no one can buy or promise that place. What a site controls is whether ChatGPT can read it: let its search crawler in, put the words in the HTML, say plainly what the business does and where, and answer the question near the top of the page.
GuideBy mttrPublished Updated 4 minute read
- 0.9%Of sites shut out ChatGPT's search crawler by nameOver the 316,711 sites read, as of October 26, 2026
- 11%Of pages show some words only once script runsOver the 5,923,925 pages it applies to
- 37%Of pages carry no structured data that parsesOver the 15,202,701 pages it applies to
- 69%Of article and service pages ask no question in a headingOver the 5,634,535 pages it applies to
Counted from the 316,711 sites mttr has read, each at its latest read, as of October 26, 2026.
What decides whether ChatGPT can show a site?
Its search crawler, OAI-SearchBot. OpenAI says sites that shut it out in robots.txt will not be shown in ChatGPT's search answers, though they can still appear as a link to themselves, and that a change to robots.txt takes about a day to count. GPTBot is a different crawler: it collects pages that may train OpenAI's models, and shutting it out does not keep a site out of search. ChatGPT-User fetches a page when a person asks for it, and OpenAI says it does not decide what appears in search.
How a page reaches a ChatGPT answer, as OpenAI describes its crawlers
- A person asksWho fixes roofs near me, or which tool does this best
- ChatGPT searchesAmong pages its search crawler was let in to read
- It reads the pagesThe HTML each one sends, and what it says it does and where
- It answersNaming the businesses and linking the pages it drew on
- Sites whose robots.txt lets it in
Most sites that shut out one of them shut out training, not search: 0.9% of the 316,711 sites mttr has read shut out OAI-SearchBot by name, against 4.3% for GPTBot. See AI crawlers for each one.
A robots.txt that keeps ChatGPT's search and a person's own lookups open while it keeps training crawlers out names only the training crawlers:
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: *
Allow: /
Sitemap: https://yoursite.com/sitemap.xmlCan ChatGPT read the page it finds?
A crawler that runs no script reads the HTML the server sends, and nothing a script adds after. 11% of the pages mttr has read show some of their words only once script runs. A page whose answer is drawn by script can be found and still say nothing.
Does the site say what it does, and where?
An assistant recommends a business for a question such as "who fixes roofs near me", so the site has to say, in words and in data, what the business does and where it works. A title and opening that name the service and the place, and structured data that says the same, do it. 37% of the pages mttr has read carry no structured data that parses.
Does the page answer the question?
A passage an assistant can lift on its own is a question as a heading and the answer right under it, in a few sentences, before anything else. 69% of the article and service pages mttr has read have no heading phrased as a question, and 90% of article pages do not answer the first one right under it. See what is GEO for a passage before and after.
Pages failing each fix
| Fix | Pages failing, lower is better |
|---|---|
| Put the text in the HTML | 11% |
| Add structured data that parses | 37% |
| Phrase a heading as the question | 69% |
| Answer it right under the heading | 90% |
| Show when the page was updated | 19% |
What does not work?
- Paying for it. ChatGPT's answers are not sold as places, and anyone selling a guaranteed recommendation is selling a guess.
- Words written for the model, not the reader. Hidden text, keyword lists and instructions addressed to an assistant are what search engines treat as spam, and they read as spam to the person who clicks through.
- Many near-copies of one page. A page for every town with only the name swapped is what search engines call scaled content.
- Claims the site cannot back. Reviews, ratings and awards in structured data that the page does not show are claims a search engine can penalise.
How do you make the fixes?
Read the site the way a crawler does, fix what is worth the most first, and read it again once the change is live. mttr does the reading and the checking; your agent, or you, make the change in the site's own code. See agentic SEO for the loop.
Make the site readable to ChatGPT
Use mttr to read yoursite.com the way ChatGPT's search crawler does. Check that robots.txt lets OAI-SearchBot in, then fix the three things worth the most, one at a time, in the templates that make the pages, using only facts the pages already state. After I deploy, have mttr read the site again and tell me what changed.
Hand this to Claude Code, Codex, Cursor or any agent connected to mttr.
A score or a rule passed says pages are well formed. It does not say an assistant cites the site. A site ChatGPT can read is one it can recommend; whether it does is ChatGPT's to decide.
Sources
- Overview of OpenAI crawlersOpenAI, on OAI-SearchBot, GPTBot and ChatGPT-User
developers.openai.com - Spam policies for Google web searchGoogle Search Central, including hidden text and scaled content abuse
developers.google.com - Creating helpful, reliable, people-first contentGoogle Search Central
developers.google.com
How the figures were counted. Every figure on this page is counted from the 316,711 sites mttr has read, each at its latest read, with drafts left out, as of October 26, 2026, and worked out again as sites are read. A share of pages counts only pages that were read: a site read only in part speaks for the pages read and no others. Each rule is the one in mttr's rulebook, decided by code. A score or a rule passed says pages are well formed. It does not say an assistant cites the site. The methodology explains each count.