utm_source=chatgpt.com: What It Means in Your URL and Analytics

utm_source=chatgpt.com is a tag ChatGPT adds to the links it shows, so the website you land on can tell its analytics tool that the visit came from ChatGPT. The tag contains only the text chatgpt.com: nothing about you, your account or what you asked. Deleting it from a link changes nothing for you. For a site owner it is the most reliable AI-referral signal there is, but it only counts clicks on tagged links, so treat it as a floor, not a count.

What the tag means

UTM parameters are labels that analytics tools read from the end of a URL. Google Analytics defines utm_source as the “Referrer” of a visit, with examples such as google or newsletter4. When ChatGPT links to a page, the address you land on looks like this:

https://example.com/pricing?utm_source=chatgpt.com

OpenAI’s Publishers and Developers FAQ says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs so publishers can measure that traffic. OpenAI’s help centre refuses automated requests, so we could not load the page ourselves on 13 October 2026; the wording comes from its indexed copy and matches what site owners see in their reports. OpenAI documents only the source; it mentions no utm_medium or utm_campaign, and Google notes that “Missing UTM parameters will result in (not set) values being present in reporting”, which matters below.

Is it safe, and can I remove it?

Yes and yes. If you are copying a link to share, delete utm_source=chatgpt.com together with the ? or & in front of it. Leave any other parameters alone: they may be needed to load the right product, search or page.

• https://example.com/guide?utm_source=chatgpt.com becomes https://example.com/guide
• https://example.com/shoes?size=9&utm_source=chatgpt.com becomes https://example.com/shoes?size=9

Removing it does not hide you from the site. Your browser makes the same request either way; the only difference is that the site’s analytics can no longer credit ChatGPT for the visit.

Should a site owner strip it?

No. A server-side redirect that drops the query string before your analytics tag runs throws the attribution away on the first page view. Deal with the duplicate URL the normal way instead:

• Put a rel=canonical on every page pointing at the clean URL, so search engines and AI crawlers treat the tagged and untagged addresses as one page. The details are in canonical URLs and Open Graph for AI citations.
• If your CDN includes query strings in its cache key, every tagged URL is a separate cache entry. Most CDNs can ignore utm_ parameters in the cache key; change that setting rather than stripping the tag.

Where ChatGPT traffic shows up, and where it can’t

The tag travels inside the URL, so it survives even when a browser or app drops the referrer. What it cannot survive is the link not being clicked.

The last two rows are the big ones. Practitioners report again and again that people act on an assistant’s recommendation later, by typing the name, and the tag never sees that visit. The tag measures clicks, not influence.

What we observed on 13 October 2026

We requested the public entry page of each major assistant and read the referrer policy it sets. That policy decides whether a browser tells your site where a click came from. Under the default policy, strict-origin-when-cross-origin, MDN explains that cross-origin requests send only the origin, and nothing at all from HTTPS to HTTP.

Two caveats. These were logged-out pages, and a single link inside an answer can override the page’s policy with rel="noreferrer", which we could not inspect without an account. And a referrer only ever names the site: no assistant tells you what the person asked. One practical finding: if your site still serves plain HTTP, ChatGPT’s policy sends no referrer at all, so untagged ChatGPT clicks become Direct.

Other assistants send referrers, not tags

We have not found vendor documentation of a UTM tag for Perplexity, Claude, Gemini or Copilot, so their visits are identified by the referrer alone. Cloudflare’s bot reference lists the referrer domains per company: openai.com and chatgpt.com for OpenAI, anthropic.com and claude.ai for Anthropic, perplexity.ai for Perplexity. Google’s own AI answers are the exception that trips people up: clicks from AI Overviews and AI Mode come from google.com and stay in Organic Search, and Google’s definition of the GA4 AI Assistant channel says it “excludes Google’s AI Overviews and AI Mode”. Copilot answers inside Bing search arrive as bing, which GA4 lists as a search engine.

How to find ChatGPT visits in GA4

1. Open Reports > Acquisition > Traffic acquisition.
2. Change the table’s primary dimension to Session source / medium.
3. Type chatgpt into the search box above the table.

To report those rows as one number, build a custom channel group that matches on Source rather than Medium. Our guide to the GA4 AI Assistant channel has a regex covering every major assistant, tested on 13 October 2026, and custom channel groups apply to past data too. To see which pages ChatGPT sends people to, open Explore > Free form, add Landing page + query string as a row, and filter Session source to chatgpt.com.

Treat it as a floor, not a count

The tag cannot see the larger effect: people who read your name in an answer and come back another way. Two cheap additions close part of the gap. Add an “AI assistant (ChatGPT, Gemini, Copilot…)” option to your “How did you hear about us?” field, and use the Search Console and Bing AI reports, which count exposure rather than clicks. Our guide on how to measure AI visibility puts all of them into one monthly routine.

And ChatGPT can only link to pages its search crawler is allowed to reach. OpenAI says sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers”. Run a free audit to check your robots.txt and firewall for OAI-SearchBot, and read GPTBot vs OAI-SearchBot before you block AI training.