News & analysis
ChatGPT's Missing Referral Tags Scramble Site Analytics
Graphite CEO Ethan Smith reported ChatGPT-attributed GA4 sessions fell 22% to 57% starting Oct. 9, which he and others tie to a tracking bug, though there isn't yet enough data to rule out some real loss of visits.
By Roomer Editorial · · 4 min read
What happened to ChatGPT traffic in GA4?
Starting October 9, sites tracked by Graphite with meaningful AI Assistant traffic saw ChatGPT-attributed sessions in Google Analytics 4 fall between 22% and 57%, according to Graphite CEO Ethan Smith's LinkedIn post. Smith estimates that's roughly a 45% drop from the prior baseline.
LLM Pulse reported a similar pattern: ChatGPT sessions on its monitored sites were about 46% lower during Oct. 9–11 compared to the same days the two weeks before. Total traffic on those sites held steady.
Adobe's Cedric Huesler looked at roughly 500 pages where ChatGPT drove at least 5% of visits. Comparing Oct. 9–13 to the prior week, he found ChatGPT visits down about 31,000 a day while Direct visits rose about 26,000 a day. Search traffic fell 7%, consistent with broader search trends.
Why did it happen?
Smith says ChatGPT began opening links in answer text without the utm_source=chatgpt.com tag and without a referrer, so those visits land in GA4 as Direct instead of AI Assistant. Huesler ties the change to the GPT-6 rollout, which he says opened links in a new tab with a browser setting that strips the referrer, on top of dropping the UTM tag that's been attached to ChatGPT citations since search launched in 2024.
Not every link type was affected the same way. Smith noted inline links in answer text lost the tag, while Sources-panel links and shopping links kept it. He also found that 95% of mobile ChatGPT visits to Graphite's sites relied on the tag alone, which makes app traffic the most exposed to this kind of change.
Huesler's data backs up the Direct-channel shift: Direct rose on two of every three pages that lost ChatGPT traffic, versus about half the week before. On mobile, the share of Direct visitors shown the consent banner, his marker for a first-time visit, climbed from 36% to 44%, closer to the 55% rate he sees among ChatGPT visitors.
Profound connects the change to a new answer format. Among ChatGPT answers with citations it collected, the share with at least one tagged source link dropped from about 98% on Oct. 8 to 66% by Oct. 14. Over the same window, Intelligent UI, ChatGPT's format for charts, comparisons and small tools, grew from 2% of all collected answers to roughly 35%. Links opened through a product or information panel still carried tags.
Separately, Peec AI's Tomek Rudzki ran one million prompts before and after the change. He found ChatGPT inserted a direct webpage URL in 42% of responses, up from roughly 8-10% previously, and that 54% of those URLs pointed to home pages rather than deeper content. These links had no tags or referrers attached. Rudzki's data reflects links shown in responses, not confirmed clicks.
Has OpenAI fixed it?
Smith called the change a bug and expected a quick fix. Huesler says OpenAI confirmed on Oct. 14 that it was a bug, and in a follow-up post reported a fix began rolling out early Oct. 15. By 9 a.m. PT that day, he saw tagged ChatGPT clicks back to roughly 95-100% of the same hour the week prior, across thousands of tracked domains.
The picture isn't fully settled. SEO consultant Glenn Gabe said he wasn't seeing the change on paid or free accounts yet. Profound said it hadn't seen official confirmation of a fix, and OpenAI's ChatGPT release notes for Oct. 1 through Oct. 15 don't mention link tracking. Huesler also noted that most ChatGPT clicks still arrive without a referrer, so he recommends tracking the UTM tag specifically rather than referrer data alone.
Graphite, LLM Pulse, Profound and Peec AI all sell SEO or AI search tools, and Huesler works at Adobe. Neither the LLM Pulse nor Profound posts disclosed sample sizes.
What should sites do now?
Both Smith and Huesler recommend annotating October 8 in your reporting before treating the dip as a real loss of visitors. Huesler now puts the affected window at Oct. 8–15.
Until the tagging is confirmed stable, compare ChatGPT and Direct traffic side by side for any page that previously got ChatGPT referrals, plus your homepage, where Rudzki's data shows more than half the untagged direct links point.
Google's AI Assistant channel, introduced in May, already classified referrer-less AI traffic as Direct by default. The ChatGPT UTM tag was the workaround that let sites separate that traffic out, especially on mobile. Smith notes that UTM parameters you set yourself for ChatGPT Ads clicks should be unaffected by this bug.
Gemini moved the opposite direction two weeks ago, adding UTM parameters to its links, a change covered separately by Roger Montti.
Roomer Editorial's take
If you're a travel creator whose Roomer site has been pulling inquiry traffic from ChatGPT, don't panic over a GA4 dashboard that suddenly shows a cliff starting Oct. 9. Based on this reporting, a tracking bug is the likeliest explanation, though it's not yet possible to fully rule out some real drop in interest.
What this means for your own site: check your Direct traffic for the same window. If it rose while ChatGPT referrals fell, that's consistent with visitors landing on your bookable pages the same way they always did, just mislabeled. Your 100% commission on bookings made through your own site isn't affected by how GA4 categorizes the click that got someone there.
The bigger lesson is about dependency on any single platform's tagging behavior. A creator relying on an AI assistant's citation format to prove their content drives bookings is at the mercy of that platform's engineering decisions. Owning your site and your booking rails means the traffic shows up in your own numbers regardless of what OpenAI changes next.
One concrete action: before assuming ChatGPT referrals to your site have dropped, pull your Direct traffic for the same pages and date range, and look for a matching rise.
Source: https://www.searchenginejournal.com/chatgpt-traffic-drop-appears-to-be-a-tracking-problem/593157/
