ChatGPT traffic dropped in GA4 from 9 October 2026: what is known
From 9 October 2026, several analysts saw ChatGPT sessions in GA4 fall (22 % to 57 % on sites with meaningful AI traffic, per Graphite) while Direct rose. The reported cause is that ChatGPT links lost their utm_source tag and referrer. OpenAI has made no public statement that we have seen; one analyst says OpenAI confirmed a bug to him on 14 October and that a fix began on 15 October.
Last checked · How we check
On this page
Status as of 19 October 2026
Open. A fix is reported by one person (Cedric Huesler, Adobe), relayed by Search Engine Journal. Search Engine Journal quotes no public OpenAI statement and notes that OpenAI's release notes for 1 to 15 October do not mention link tracking. Relevant Audience writes that OpenAI has not confirmed it. Huesler adds that most ChatGPT clicks still arrive without a referrer. We have not independently measured any of this.
Timeline
| Date (2026) | What was reported | Who |
|---|---|---|
| 7 to 9 Oct | Drop in ChatGPT human and ChatGPT-User traffic and a rise in OAI-SearchBot traffic on his own sites, not quantified | Sébastien Billard, per Relevant Audience |
| 8 Oct | About 98 % of ChatGPT answers with citations in Profound's collected set had at least one tagged source link | Profound, per SEJ |
| 9 Oct | Start of the drop in GA4-attributed ChatGPT sessions | Ethan Smith (Graphite), per SEJ |
| 14 Oct | Search Engine Roundtable reports the issue (we read only the summary by Relevant Audience); Huesler says OpenAI confirmed to him it was a bug; the tagged share in Profound's set is 66 % | SER, Huesler, Profound |
| 15 Oct | Huesler: a fix began rolling out early in the day; by 9 a.m. PT tagged clicks were back to roughly 95 to 100 % of the same hour a week earlier across thousands of domains he tracks. SEJ publishes the report | Huesler, per SEJ |
The numbers and what they rest on
| Source | Reported figure | Basis and limits |
|---|---|---|
| Graphite (Ethan Smith) | ChatGPT-attributed GA4 sessions down 22 % to 57 % from 9 Oct; he estimates about 45 % | Sites with meaningful AI Assistant traffic. His explanation: ChatGPT started opening links in the answer text without the tag or a referrer |
| LLM Pulse | ChatGPT sessions about 46 % lower on 9 to 11 Oct than on the same days in the prior two weeks; total traffic roughly constant | Sites it monitors. The authors say aggregated data cannot separate lost traffic from traffic no longer labelled ChatGPT |
| Adobe (Cedric Huesler) | ChatGPT visits down about 31,000 a day, Direct up about 26,000 a day, 9 to 13 Oct vs the prior week; Direct up on two of three pages that lost ChatGPT traffic; search down 7 % | About 500 pages where ChatGPT drove at least 5 % of visits; four days of sampled data. The two daily figures are not a reconciled net change |
| Profound | Share of cited ChatGPT answers with at least one tagged link: about 98 % (8 Oct) to 66 % (14 Oct) | Profound's own collected answers, not visits. It had not seen official word on a fix |
| Peec AI (Tomek Rudzki) | A direct URL appeared in 42 % of responses, up from about 8 to 10 %; about 54 % of those pointed to home pages | 1 million prompts, 2 to 7 Oct vs 9 to 13 Oct. Links displayed, not clicks |
Glenn Gabe, quoted by SEJ, said he was not yet seeing the change on paid or free accounts.
What we do not know
- Whether the change was intended. Search Engine Roundtable is reported as unsure; Huesler says it was a bug.
- Whether the fix applies to every surface (web, mobile, apps). Huesler says most ChatGPT clicks still arrive without a referrer.
- How large the effect is on a typical site. All figures above come from samples selected by the people who published them.
- What OpenAI says. We found no first-hand statement, and we could not read OpenAI's help pages (HTTP 403).
Is your site affected?
- In GA4, compare ChatGPT sessions for 1 to 8 October, 9 to 14 October and 15 October onwards.
- Put Direct next to it on the same pages. Deep pages that ChatGPT used to link to are more telling than the home page.
- Check new-user share and engagement of the extra Direct sessions against your September ChatGPT sessions.
- Annotate the break around 8 to 10 October in GA4 or your report, and report ChatGPT and Direct together for affected pages rather than ChatGPT alone.
- Server logs will not recover the label: Relevant Audience points out they face the same missing referrer. Bot lines (OAI-SearchBot, ChatGPT-User) are a separate signal.
Advice in the last two steps follows Relevant Audience's analysis. For the background see why clicks end up as Direct and what the tag is; use the referrer checker on a visit you have captured.
Updates
2026-10-19: page published; sources read on this date. This page will be edited when a primary statement appears or when the reports change; the date at the top moves only then.
Follow-up questions
Did ChatGPT traffic really fall?
The reports say fewer sessions were labelled ChatGPT. Adobe's sample shows Direct rising by a similar order, so some of the fall may be relabelled visits rather than lost ones. The sources themselves say the data cannot separate the two.
Will my GA4 history be corrected?
Nothing we read says so. Clicks that arrived without a referrer or tag were not labelled when they happened, and analytics tools do not relabel them later.
Where can I see this kind of change again?
On the referrer behaviour changelog, which lists dated changes with their sources and status.
Sources
Read on . Links lead to the original pages.