ChatGPT keeps your tracking tag.Then Google takes the credit.

I found this while trying to disprove something else. I had a hotel where GA4 reported far more Google organic sessions than Search Console reported Google clicks, and I wanted to know whether AI traffic was hiding in the difference. It was not. But one row in that report (a single session labelled chatgpt.com / organic) turned out not to be the rounding error I took it for.

It is a mechanism, it runs across hundreds of hotel brands, and it points the opposite way to every other AI-attribution problem I have written about. Those are all about AI traffic being invisible. This one is about AI traffic being visibly filed under the wrong name, with a paper trail, because of a tracking tag the hotel added itself for perfectly sensible reasons.

That middle line is an OR. The deliberateness shows in the rest of the table, which joins Direct and every paid channel with AND and reserves OR for Organic Search, Organic Social, Organic Video, Email and SMS. The medium alone is enough. So chatgpt.com / organic satisfies Organic Search without the source mattering at all, which is just as well for Google, since chatgpt.com is not on the list it would have checked.

And it cannot land in the AI Assistant channel instead, because that rule requires the medium to match ai-assistant exactly. With organic sitting in the medium field, the AI Assistant rule fails no matter what order the rules are evaluated in. That matters, because Google documents that ordering for custom channel groups and never for the default one. The outcome here does not depend on it.

It would be easy to write that this makes ChatGPT traffic invisible. It does not, and the distinction is worth getting right.

The Session source dimension still says chatgpt.com. The traffic is misfiled, not missing. What breaks is everything that reads the channel rather than the source: the AI Assistant channel count, every channel-grouped report and comparison, and any attribution model built on the default channel group. Meanwhile the Organic Search number is inflated by clicks Google did not send.

So the hotel that did the most work to measure its listing carefully ends up with two wrong numbers instead of one missing number, and no indication that anything is off. You would only find it by opening the source/medium report and noticing that ChatGPT appears twice, on two different rows, for no obvious reason.

There may be a partial rescue. Google introduced a Source Group dimension in June 2026 with, in its words, “built-in grouping for emerging traffic sources like ChatGPT (OpenAI) and Perplexity”, populated retroactively. Being derived from the source, it ought to bucket these clicks correctly even with the medium hijacked. Google documents neither how it is derived nor how it interacts with manual UTMs, so I am flagging it as a thing to test rather than a thing to rely on.

Open your Google Business Profile, look at the website field, and read the query string. If it contains utm_medium=organic, that is the whole problem.

The instinct behind the tag was right: you wanted listing traffic separated from ordinary search traffic. The error is which field it went in. Medium is the field GA4 uses to assign channels, and organic is a reserved word inside it. A source-only tag gives you the same separation without touching the channel logic:

Two warnings before you go and edit it. Any change to the Business Profile URL takes time to propagate into ChatGPT, since the widget reflects what Google holds rather than what you just typed. And I have seen community reports of UTMs being stripped from that field, and one of a listing suspension after adding them. Those are user-generated forum posts rather than anything Google has stated, so treat them as an unverified risk rather than guidance.

The corpus. ChatGPT responses to hotel prompts captured at scale and flattened to one row per map or place entity, retaining the website_url field verbatim so the query string survives. 230,863 entities. utm_source and utm_medium are extracted by regular expression; values are compared lower-cased, and URL-encoded variants such as seo%20maps and seo+maps are left distinct rather than merged, since they are separate strings as far as GA4 is concerned.

How the overwrite is established. Not by watching ChatGPT rewrite a URL, which we cannot observe directly, but by finding domains that appear with an identical utm_medium under two different utm_source values, one of which is chatgpt.com and the other of which is the business’s own convention. A medium like seo+maps or organic_search is not plausibly ChatGPT’s invention, and it appears on the same domain under both sources.

The GA4 rules are quoted from Google’s default channel group documentation as rendered in September 2026. The source-category file is the artifact linked from that page; its 819 rows were extracted and searched, and the 9 February 2023 date is the HTTP Last-Modified header on the file itself rather than a claim on the page.

Limits. One vertical: these are hotel prompts, so the domain counts describe hotels rather than businesses in general. The corpus records what ChatGPT served rather than what visitors clicked, so this measures exposure and never traffic volume. Whether other assistants with place widgets rewrite utm_source the same way is untested. And the final step of the argument, that a manual medium suppresses Google’s automatic ai-assistant assignment, is composed from two documented facts rather than quoted from one.