What is GPTBot? We watched it read our JavaScript as text

On day 6 of our new site, GPTBot made 156 verified requests in 105 seconds. 18 were for raw fragments of our JavaScript code, which shows it read the script as text. 79 went to WordPress oEmbed and REST API addresses.

On this page
  1. What is GPTBot?
  2. What GPTBot requested on our site
  3. What GPTBot did with our JavaScript
  4. How to check GPTBot on your own site
  5. What to do
  6. Frequently asked questions
  7. Get the next Census by email
  • 156verified GPTBot requests in 105 secondsCensus bot log, 4 Nov 2026
  • 18requests for 6 raw pieces of our JavaScript code, as if they were linksCensus bot log, 4 Nov 2026
  • 79of the 156 went to WordPress oEmbed and REST API addressesCensus bot log, 4 Nov 2026
  • 23verified GPTBot requests on a second visit, 19 seconds on 10 Nov, none for pieces of JavaScript codeCensus bot log, checked to 13:40 UTC, 11 Nov 2026

Short answerGPTBot is OpenAI’s crawler for AI training data. On day 6 of our new site it made 156 verified requests in 105 seconds, and 18 asked for raw pieces of our JavaScript as if they were links: it read the script as text.

What is GPTBot?

FindingOpenAI lists 3 agents you may see in your logs, and GPTBot is the one it describes as crawling for training data.

AgentWhat OpenAI says it does
GPTBotCrawls content that may be used in training OpenAI’s generative AI foundation models
OAI-SearchBotFinds pages for ChatGPT search; a site that blocks it will not be shown in ChatGPT search answers
ChatGPT-UserFetches a page when a person asks ChatGPT to; OpenAI says it does not crawl the web automatically
Source: OpenAI, bots.

What GPTBot requested on our site

Finding79 of GPTBot’s 156 requests (51%) went to WordPress oEmbed and REST API addresses, more than to web pages and Markdown copies combined.

What GPTBot requestedRequests
WordPress oEmbed addresses (26 JSON, 26 XML)52
Web pages, including 2 category archives30
WordPress REST API addresses (pages, posts and one category)27
Markdown copies of pages (.md)26
Pieces of JavaScript code requested as links18
Category RSS feeds2
llms.txt1
Total156
Source: AI Visibility Census bot log, 08:42:20 to 08:44:05 UTC on 4 November 2026. 155 requests came from 74.7.227.25 and 1 from 172.182.207.102, both on OpenAI’s published GPTBot list. One web page returned 404: an article that was scheduled but not yet published. Free to reuse under CC BY 4.0.

WordPress lists these addresses in the code of every page and post. From launch on 29 October to 13:40 UTC on 4 November, Googlebot and ClaudeBot never requested an oEmbed address or a WordPress page, post or category API address on our site.

What GPTBot did with our JavaScript

FindingGPTBot requested 6 pieces of code from our script file as addresses, 18 requests in all, and no other crawler did.

Our Content Shape Checker loads a script, hsc.js, that builds a printable report after a visitor runs a check. Inside it, the report’s links are written as joined pieces of code:

href="' + escHtml(utm(toolUrl, 'header')) + '"
src="' + escHtml(brand.logo) + '"

What this does not settle: whether GPTBot runs JavaScript. Those links are only built after a visitor uses the tool, so a crawler that ran the script on page load would still not have produced them.

How to check GPTBot on your own site

FindingBefore GPTBot’s first real visit, our log already held 2 requests named GPTBot from outside OpenAI’s list.

  1. Find requests with GPTBot in the user agent and check each IP against openai.com/gptbot.json. Anything outside it is unverified.
  2. Look for requests to your script folders whose paths contain code such as +, quote marks or function names.
  3. Compare the pages it fetched with the pages that need JavaScript to show their content.

Both came from the same IP addresses as our own testing. The free crawler log check skill on our Tools page runs the first check for you. For Anthropic’s crawler, see What is ClaudeBot?

What to do

  1. Serve key content without JavaScript. GPTBot read our script as text; a server-rendered page or a plain HTML or Markdown copy removes the question.
  2. On WordPress, check oEmbed and REST API hits. They were 51% of GPTBot’s crawl of our site.
  3. Block training and search separately. A GPTBot block does not cover OAI-SearchBot, which decides whether you can appear in ChatGPT search.

Next: the CensusIndex™ checks whether leading sites can be crawled and read by AI tools; CensusRank™ and our Editions count which brands and firms AI answers name. Get each result by email.

Cite this

On a new site, GPTBot made 156 verified requests in 105 seconds on 4 November 2026: 79 to WordPress oEmbed and REST API addresses and 18 for raw pieces of JavaScript code requested as links.

Source: AI Visibility Census, “What is GPTBot?”, https://aivisibilitycensus.com/does-gptbot-run-javascript/ (CC BY 4.0)

Frequently asked questions

Can AI crawlers execute JavaScript?

Our crawl does not settle it. GPTBot read our JavaScript file as text and made 18 requests for 6 raw pieces of its code as links, but those links are only built after a click.

How fast does GPTBot crawl a site?

On ours, 156 requests in 105 seconds, about 1.5 a second, from two IP addresses. Its first verified visit came on day 6 after launch. It came back on day 12, at 19:39 UTC on 10 November: 23 verified requests in 19 seconds from one address, mostly our template pages, their Markdown copies and their WordPress oEmbed and REST addresses.

Does GPTBot read llms.txt?

It fetched ours once, seven seconds into its crawl. It also fetched 26 Markdown copies of our pages by their .md addresses. On its second visit, on 10 November, it did not request llms.txt.

How do I block GPTBot?

Add a group for it to robots.txt: User-agent: GPTBot, then Disallow: /. It does not block OAI-SearchBot. Our robots.txt allows both, so we have not tested whether GPTBot obeys a block.

The full JavaScript detail
  • It requested the code itself as a path, for example /wp-content/plugins/human-shape-checker/assets/'%20+%20escHtml(brand.logo)%20+%20'.
  • The first request for each piece named hsc.js as the referring address, so it found these strings in the script file.
  • None of the 468 verified Googlebot or 186 verified ClaudeBot requests from launch to 13:40 UTC on 4 November were for these addresses.

When a visitor prints a report, the code between the quotes becomes a real address. GPTBot took the code itself and requested it as a path in the script’s folder. Each of the 6 pieces was requested once as written in the script (one cut off at its # sign). Our server redirected it twice, first dropping the quote marks and then a trailing space, and GPTBot followed both redirects to an address that returned 404: 3 requests per piece, 18 in all. It took every href and src value written in that part of the script.

robots.txt, llms.txt and the WordPress addresses

GPTBot fetched every published page but one. The REST API addresses return each page’s full text as JSON; the oEmbed addresses return a short embed card for it. If you run WordPress and want GPTBot’s visits spent on your pages, these are the addresses to look at in your own log first.

GPTBot fetched llms.txt once, as its fourth request, after the home page, About and Methodology. Whether llms.txt requests lead crawlers to new pages is what our experiment LL-001 tests, with the result due on 1 December. No GPTBot request for robots.txt reached our log before or during the crawl. OpenAI’s search crawler, OAI-SearchBot, had requested robots.txt twice the day before, at 06:21 and 22:22 UTC on 3 November. Our robots.txt allows every crawler, so this site cannot test whether GPTBot obeys a block.

How we counted, and limits

Counts come from our own bot log, which checks each request against the IP lists that Google, Microsoft, Apple, Anthropic, OpenAI and Perplexity publish. How we count is in the methodology, and every test on this site is listed on our experiments page.

Limits: these are two visits by one crawler on one site with a few dozen pages, and the Finding sections come from the first. They show what GPTBot did on 4 and 10 November 2026, not what every AI crawler does or what GPTBot will do next month. Our log records requests that reach WordPress, so static files such as scripts, styles and images are not in it: we know GPTBot read hsc.js from the referring address on its requests, not from a request for the file. A fetch is also not use: the log cannot show whether any ChatGPT answer draws on these pages. We are timing when each AI tool first cites this site in our article How long does it take a new website to get cited by AI search?

Data on this page is free to reuse under CC BY 4.0; please credit the AI Visibility Census with a link to this page. Spotted an error? Email hello@aivisibilitycensus.com.

Sources: OpenAI: bots; OpenAI: published GPTBot IP addresses (gptbot.json); AI Visibility Census bot log, 29 October to 11 November 2026.

Published by the AI Visibility Census.

Newsletter

Get the next Census by email

Each new edition, list and experiment result, sent when it publishes. About one email a week. One click to leave.

We send a confirmation email first. If you add a website, we may tell you when your company appears in a Census result. Privacy notice