Is ChatGPT visiting your site? We checked OpenAI's three bots in our own logs (October 2026)
The short answer: OpenAI sends not one visitor but three: one for search (OAI-SearchBot), one for training (GPTBot), and one that opens a page because a user asked (ChatGPT-User). Check each separately. On our site pw.migurei.ai, visitors calling themselves OAI-SearchBot came 36 times in three weeks, every time for /robots.txt only, and fetched no page at all. The training bot GPTBot, by contrast, fetched 15 pages once, on 16 October, by following links from the home page.
Recorded 21 October 2026 (each item gives its own counting date). Every number comes from the logs of one site of ours and from our records of questions put to AI apps.
Words used on this page
- We / our: migurei (a subsidiary of HBLAB), which runs Parallel World and the test site pw.migurei.ai ("pw" below).
- Bot: a program, not a person, fetching pages. The server log keeps the name it gives (UA, user agent) and the IP address it came from.
- IP check: anyone can claim a UA, so we count a visit as genuine only when its IP address is in the list OpenAI publishes.
- Run: one trial in which we put a fixed question to an AI app and record the answer and the server log.
OpenAI's three visitors
OpenAI describes three separate visitors on its official page (https://developers.openai.com/api/docs/bots).
| Name (UA) | Job | What our log shows |
|---|---|---|
| OAI-SearchBot | Collects pages for ChatGPT search | /robots.txt only; 0 pages |
| GPTBot | Collects pages for training AI models | Once, on 16 October: 15 pages |
| ChatGPT-User | Opens a page when a user asks ChatGPT to | Came when we pasted a URL in our Runs |
- OAI-SearchBot comes in two forms:
OAI-SearchBot/1.4, which fetches pages, andOAI-SearchBot/1.4; robots.txt;andOAI-SearchBot/1.0, which only check permission inrobots.txt(same OpenAI page, read 18 October 2026). "OAI-SearchBot came" in a log may mean a permission check and nothing more. - OpenAI says sites that disallow OAI-SearchBot in
robots.txtwill not be shown in ChatGPT search answers, though they can still appear as navigational links. - OpenAI also says: "If your site has allowed both bots, we may use the results from just one crawl for both use cases to avoid duplicative crawling." It says "may", and not when or how often.
- The official page as we read it (18 October 2026) describes no way to submit a sitemap or URLs.
Our log: OAI-SearchBot, the search bot
- Cloudflare (the service that relays traffic in front of our site) recorded 36 requests to pw from visitors calling themselves OAI-SearchBot between 1 and 21 October 2026. All 36 were for
/robots.txt, and all 36 were answered normally (HTTP 200). - We were not blocking it at the edge. All of Cloudflare's AI-bot settings were off, and there were 0 block events on pw (checked 21 October 2026).
- In our server log, counting only IP-checked visits (30 days to 19 October 2026), the permission-check forms came 30 times and the page-fetching
OAI-SearchBot/1.4came 0 times. Of the 19 URLs in our sitemap, OAI-SearchBot fetched 0.
Our log: GPTBot, the training bot
- On 16 October 2026, in the 18 seconds from 03:11:45 to 03:12:03 UTC (12:11:45 to 12:12:03 Japan time), GPTBot (IP-checked) followed links from the home page and fetched 18 paths: 15 of the 17 URLs then in our sitemap. The 2 it skipped were pages nothing on the site linked to.
- 0.3 seconds earlier, OAI-SearchBot had read
/robots.txtfrom the same IP address range. - Since then GPTBot has read the sitemap (
/sitemap.xml) every day (4 times, 16 to 19 October). But it fetched none of the 4 new pages that were only in the sitemap, with no link to them. Since 16 October GPTBot has fetched 0 pages (as of 19 October).
So OAI-SearchBot's 0 does not mean the pages are missing from OpenAI's search: OpenAI itself says one crawl may serve both purposes, and GPTBot did fetch 15 pages. Whether those pages were in fact used for search, our log cannot tell.
Our Runs: what showed up in ChatGPT's answers
- ChatGPT-User: when a question in our Runs contained a pw URL, ChatGPT-User came to fetch the page: 173 requests in 143 Runs, 1 to 17 October 2026.
- Pages cited: across 1,571 Runs from 30 September to 15 October 2026, ChatGPT cited pages of our company site migurei.ai (
/enin 25 Runs,/jain 17, and others) and never a pw page. Most of those Runs, though, had a pw URL pasted into the question or named the service. - Questions that name nothing: we asked 9 questions that give neither the service name nor a URL (one of them asks how to see how a site looks to AI) 99 times across 4 apps. Of the 433 links captured from the answers, 0 pointed to our sites (counted 13 October 2026). Sources ChatGPT folds into a "+N" button, and Gemini's sources, were not captured.
- People arriving from AI apps: pw has no record of a person arriving from ChatGPT or another AI app, 1 to 16 October. But pw is a little-known test site that gets almost no human visitors at all, from AI apps or anywhere else.
An outside case (one, not checked by us)
An article on Zenn (by hirajo, 11 October 2026, https://zenn.dev/hirajo/articles/crawler-google-vs-chatgpt) reports that on a newly built site, OAI-SearchBot read only robots.txt for the first days and read an article page for the first time two days after the article went up. The author says the IPs were checked against OpenAI's list.
That is one author's report about their own site, and we have not checked it. The site is run by the author themselves. That site had articles answering questions, while pw, when we started counting (14 October), had only pages describing the service; whether that difference explains anything, we do not know.
This page is part of the test too
We have published this page on pw as an article answering a question. We will record whether OpenAI's search bot (the page-fetching OAI-SearchBot/1.4) comes to fetch it within 72 hours of publication, and add the result here, including if it does not come. Our expectation before publishing is "it will not": on pw, OAI-SearchBot has read nothing but robots.txt, and OpenAI has not yet fetched any page that links here either.
How to check your own site
- Count the three separately in your access log: lines whose UA contains
OAI-SearchBot,GPTBotorChatGPT-User. For OAI-SearchBot, separate the page-fetchingOAI-SearchBot/1.4from the permission checks markedrobots.txt;. - Check the IP: count only visits whose IP is in the lists on OpenAI's official page (https://developers.openai.com/api/docs/bots).
- Check `robots.txt` and what sits in front: make sure
robots.txtdoes not disallow OAI-SearchBot and that your CDN (Cloudflare or another) does not block it through its AI-bot settings. OpenAI says arobots.txtchange takes about 24 hours to apply to search. - Look for people: check whether the referrer is
chatgpt.comor the URL carriesutm_source=chatgpt.com. ChatGPT often adds this, but not every app leaves a referrer.
When each bot first came to pw is in When do AI search bots visit a new subdomain?. How JavaScript, structured data and llms.txt were read is in Is AI reading the numbers and structured data on your pages correctly?.
What this record does not show
- One site, a few weeks. The numbers are counts, not rates or trends. Your site may differ.
- Why OAI-SearchBot fetches no pages, we do not know. Reading
robots.txtregularly may be a permission check only; it does not show the bot "intends" to fetch pages. Zero so far does not mean never. - Whether the pages GPTBot fetched are used for ChatGPT search, we do not know. It only does not contradict OpenAI's "may use one crawl for both".
- Why ChatGPT cited migurei.ai, we do not know either. Whether ChatGPT search uses Bing's results is a question we have not been able to settle. And most of the citing Runs had a URL or the name in the question.
- A log only sees requests that reach the server. If content reaches an app through another company's search index or cache, our log does not show it. Cloudflare's 36 is counted by the name the visitor gave, without an IP check.
- We wrote this and we measured it. No independent party has checked it.
Records this page uses
knowledge/claims/oai-searchbot-polls-only-robots-on-pw.md(our observation, updated 19 and 21 October)knowledge/claims/gptbot-claudebot-first-crawled-pw-after-worker-cutover.md,gptbot-reads-pw-sitemap-daily-fetches-no-new-url.md(our observation)knowledge/claims/openai-one-crawl-may-serve-search-and-training.md(outside source: OpenAI's official page)knowledge/claims/ai-apps-fetch-pasted-urls-with-distinct-uas.md,chatgpt-cites-migurei-ai-pages-not-pw.md,no-ai-referred-human-visit-to-pw.md,pw-human-page-views-mostly-own-tools-and-team.md(our observation)docs/evidence/2026-10-13-n3-cited-domains-v2.md(our count)knowledge/claims/new-site-oai-searchbot-reads-pages-after-articles.md(outside source: one Zenn article)