How does ChatGPT decide who to recommend?
ChatGPT recommends in two ways. Without web access it answers from what it learned in training, and you can hardly change that. With web search it rewrites the question into queries, reads the pages in the results and turns what it read into an answer. The second is what you can influence: be findable, be readable, and be worth including once read.
OpenAI has not published how it ranks. What it has published is about crawlers: which one handles search, which handles user-triggered visits, and which handles training. We only cover that and do not guess the rest.
OpenAI's three crawlers: which one should you allow?
| Crawler | What it does (OpenAI docs) | Effect on ChatGPT recommendations |
|---|---|---|
| OAI-SearchBot | Lets sites appear in ChatGPT's search features | Opted-out sites do not appear in ChatGPT search answers but may still show as navigational links. OpenAI recommends allowing it |
| ChatGPT-User | Opens pages on a user's behalf when they ask in ChatGPT or a custom GPT | OpenAI says these are user-initiated visits and robots.txt rules may not apply; it does not decide whether content enters search |
| GPTBot | Collects content that may be used to train foundation models | Blocking it signals content should not be used for training; the docs do not say it affects appearing in search answers |
So the key step is allowing OAI-SearchBot. To opt out of training but stay visible in search, block only GPTBot. OpenAI says its search systems take about 24 hours to reflect a robots.txt change. A full comparison of crawlers is in AI crawler user agents: a side-by-side table.
User-agent: OAI-SearchBot Allow: / User-agent: ChatGPT-User Allow: /
What are the steps?
1. Allow OAI-SearchBot first
Open your robots.txt and make sure it is not blocked by a Disallow under `User-agent: *`. Allow it separately with the snippet above, then verify with the AI crawler check.
2. Put your text in the HTML itself
Crawlers read the raw HTML the server returns. The body text and prices on your homepage, pricing page and guide pages must be in it, not wait for a script.
3. Answer one specific buyer question per page
Make the title the question a buyer would ask, give the answer and numbers in the first two or three sentences, then add steps, tables and limits. This is the kind of answer ChatGPT can lift whole.
4. Add the Bing safety net
OpenAI confirms search uses third-party search providers, and the press reports Bing is one of them, but OpenAI gives no proportions. Verifying Bing Webmaster Tools and submitting a sitemap costs little; details are in Does ChatGPT search use Bing.
5. Ask ChatGPT your buyers' questions with web search on
Write 5–10 questions without your brand name and ask each 3 times (new chat each time). Record whether you were mentioned, who was recommended and which URLs were cited. What matters is how many of the three times you appear. The checklist is in Why ChatGPT doesn't recommend my product, and the overall logic of how AI picks brands is in How AI picks which brands to recommend.
One real site: 44 sessions from ChatGPT in 7 days
MochiStamp, a LINE sticker site we run ourselves, has a daily scoreboard. In the scoreboard generated on 2026-11-05, the last 7 days (2026-10-29 to 11-04) had 128 visitors; among session sources, chatgpt.com / ai-assistant had 44, behind Google organic search at 53 and ahead of direct visits at 35.
| Source (last 7 days) | Sessions |
|---|---|
| google / organic | 53 |
| chatgpt.com / ai-assistant | 44 |
| (direct) / (none) | 35 |
| openai / organic | 4 |
This site did three things for AI readers, visible on MochiStamp's LINE sticker size guide (opened and checked 2026-11-05):
- Multilingual guide pages: the same guide exists in Japanese, English, Traditional Chinese and Simplified Chinese, pointing to each other with hreflang.
- Body text in the HTML: the headings and body of the guide are in the returned HTML, not rendered by script.
- FAQ: the page carries FAQPage structured data, and the questions and answers match the visible text.
To be clear: the scoreboard records only the source, not which pages those 44 sessions landed on, so we cannot say which of the above brought them, and certainly not that doing the same will give the same numbers. All it shows is that for a new site doing these things, ChatGPT became the second-largest source.
FAQ
How do I get ChatGPT to recommend my product?
There is no way to guarantee it. What you can do: let OAI-SearchBot reach your pages, put the body text in the HTML, answer one specific buyer question per page, then ask ChatGPT those questions a few times, see which pages it cites and fill in what those pages lack about you.
Will blocking GPTBot stop ChatGPT from recommending me?
OpenAI's documentation says blocking GPTBot signals content should not be used to train foundation models. Whether you appear in search answers corresponds to OAI-SearchBot in the docs. To opt out of training but stay in search, block GPTBot and allow OAI-SearchBot.
ChatGPT-User is blocked in robots.txt, so why do I still see visits?
OpenAI's documentation says ChatGPT-User visits are user-initiated and robots.txt rules may not apply. It also does not decide whether content enters search.
How long does a robots.txt change take to apply?
OpenAI's documentation says search-related changes take about 24 hours to be reflected.
How do I know whether ChatGPT is visiting my site?
Look at sessions with source chatgpt.com in GA4 and at OAI-SearchBot in your server logs. No visits does not mean you were not cited, but a visit shows you appeared at least once.
Sources
- OpenAI · Overview of OpenAI Crawlers （Checked 2026-11-05）
- MochiStamp · LINE sticker size guide （Opened and checked 2026-11-05）
- kagentx · Does ChatGPT use Bing （Updated 2026-11-05）
- kagentx evaluations: the data on the case study pages; each example gives its date.