utm_source=chatgpt.com: what it means, and how to track ChatGPT and AI traffic in GA4 · Citable

Description

written for teams no score out of 100 no ten-step checklist written for teams no score out of 100 no ten-step checklist written for teams no score out of 100 no ten-step checklist

What does utm_source=chatgpt.com mean?

It means the visitor clicked a link inside a ChatGPT answer. ChatGPT adds ?utm_source=chatgpt.com to the links it shows, so the site on the other end can tell where the visit came from. A URL like yoursite.com/pricing?utm_source=chatgpt.com is your normal pricing page, opened by somebody who asked ChatGPT a question and was pointed at you.

It is not spam, it is not a hack, and nobody edited your site. It is the same kind of tag a newsletter or an ad puts on a link — added by ChatGPT instead of by you.

Why does ChatGPT add it?

Because a browser does not always say where a click came from. Apps and privacy settings strip the referrer, and a visit with no referrer is filed under “Direct”, as if the person had typed your address. The tag travels in the URL itself, so the visit is counted correctly even when the referrer is gone. It is why arrivals from ChatGPT are the easiest AI traffic to see.

Should I remove utm_source=chatgpt.com from my URLs?

No, and you cannot stop ChatGPT adding it. The tag does not change what the page shows. If you are worried about the tagged address showing up in search results, a canonical link on the page already tells search engines which address is the real one. Stripping the tag with a redirect only throws away the one clear signal you were being sent.

How to see ChatGPT traffic in Google Analytics 4

Open Reports → Acquisition → Traffic acquisition .

Change the table’s first column to Session source / medium .

Type chatgpt in the search box above the table. Visits from ChatGPT appear as chatgpt.com / referral , or with the medium not set.

Add Landing page as a second column to see which pages ChatGPT is sending people to.

How to track all AI traffic in GA4

ChatGPT is the only engine that tags its links. The others are recognised by referrer, so they need a rule of their own. Make one channel for all of them:

Open Admin → Data display → Channel groups and create a new channel group.

Add a channel called AI , with the condition Source matches regex :

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|meta\.ai|duck\.ai|chat\.mistral\.ai Move the AI channel above Referral in the list, so it is matched first, and save.

In any acquisition report, switch the channel group to yours. AI visits now have a row of their own.

Those are the same sources Citable matches on: the full list, with what each engine is and what people ask it, is under AI engines .

What analytics cannot tell you

It undercounts. A link opened inside WhatsApp, Instagram or a phone app often carries no referrer and no tag. Those visits are in your “Direct” number, and nothing in analytics can pull them back out.

It only sees people. Before anybody clicked, a crawler read your page — sometimes many times, sometimes badly. Analytics runs in a browser, and a crawler is not one, so that half is invisible to it.

It cannot say why. A page with fifty reads and no visits is a different problem from a page nobody read. You need both numbers side by side to tell them apart.

The tag tells you ChatGPT sent somebody. It does not tell you what ChatGPT was given when it read the page first — and that is the part you can change.

Seeing both halves

Citable counts the same arrivals — the tag for ChatGPT, the referrer for the rest — at your own edge, and puts them beside the reads that came before: which crawler, which page, how often, whether a person was waiting. A page that is read and never visited, and a page that is visited from an answer, sit in the same table. Your visitors’ details never reach us: what is counted is the arrival, not the person.

To see what ChatGPT is given when it reads one of your pages today, run the free scan . For what to do about it, see how to get cited by ChatGPT , and for the crawlers that do the reading, the ChatGPT page .

The chain, drawn Access, then comprehension, then the choice.

Access — an engine you refuse never reads anything.

Comprehension — the facts have to survive the markup.

Choice — which stays the model’s, whatever anyone sells you.

Read next The rest of the set.

6 min read AI search optimization

How an assistant decides what to say about your site, and which of those steps a team can actually influence.

7 min read What GEO can and cannot do

What generative engine optimisation means, what it borrowed from SEO, and which parts of it are sold as certainty but aren't.

5 min read How AI crawlers read a website

Rendering, fetch budgets, robots.txt, and why a JavaScript-drawn price is invisible to almost every engine reading you.

8 min read Generative engine optimization (GEO)

What it means, how an AI engine chooses its sources, the strategies with evidence behind them, and what GEO tools actually do.

7 min read Answer engine optimization (AEO)

What it means, three before-and-after examples, the practices that matter — and the step that comes before any of them.

6 min read How to get cited by ChatGPT

How ChatGPT picks the pages it cites, and the six things a site can actually do to show up in its answers.

5 min read Can ChatGPT read my website?

How it reads a site, why it does not run JavaScript, the five things that stop it, and how to check in a minute.

5 min read llms.txt: what it is, and what it does not do

What goes in the file, a working example, best practices — and whether any AI search engine actually reads it.

6 min read Should you block AI crawlers?

Training bots, search bots and live fetch are three different decisions — and the one blocked by accident costs the most.

5 min read GEO vs AEO vs AIO vs LLMO

The full form and meaning of each, where the terms came from, and the four steps underneath all of them.

Or just look at it One page, served both ways.

What survives each trip, with the token count for both.

About Citable Who is behind this page, and what we stand on.

When an AI agent — ChatGPT, Claude, Perplexity or any of the 81 that Citable recognises — opens a page, Citable removes the theme, scripts, navigation and layout and serves only the content: the same facts, at the same URL, on the site's own domain. People and Google still get the full designed page.

Who opens the page What they are given

A person in a browser The full designed page, untouched.

Googlebot, Bingbot and other search engines The full designed page, untouched.

An AI agent — GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot and the rest The content only: no UI, no scripts, the same facts at the same URL.

Measured, not estimated. A real product page of 386.8 KB is served to AI agents as 4.6 KB — 98.8% smaller, with the same facts. Every figure on this site comes from a fetch anybody can repeat.

One list. The 81 AI agents named on these pages are the list the product itself runs on. A page here cannot describe a crawler Citable does not serve.

Search is never touched. Googlebot and Bingbot always get your real page. The clean copy is for AI agents only — at the same address, with the same facts.

Checked, not trusted. A user agent is a claim, and anyone can send one. Where an operator publishes its addresses, every read is checked against them and recorded as verified or not.

Nothing about your visitors. Citable sees which crawler read which page. It does not see who your customers are, and their details never reach us.

Honest about limits. Nobody can promise a citation or a ranking, and we do not. What we show is what an AI agent was given, and whether it sent somebody back.

How it works · The measurements · What it touches and stores · Why Citable · Pricing · Questions, answered plainly · Talk to us

Free · about 20 seconds · no signup Find out whether AI has ever read your site.

Type your address. We fetch a page exactly as GPTBot does, count the tokens, and score fifteen things an answer engine needs. If the log then stays empty for a month, that is worth knowing too.

No contract · cancel any time · Googlebot untouched