Guide · AI search measurement
ChatGPT referral traffic: what broke in October and how to track it
Short answer
From about 9 October, analysts reported that links in ChatGPT answers stopped carrying utm_source=chatgpt.com and a referrer, so visits could appear as Direct traffic. A fix was reported from 15 October, but OpenAI had not published a statement we could find. If you report ChatGPT traffic, annotate 8 to 15 October, compare ChatGPT and Direct side by side, count tagged requests in your server logs as a floor, and say that the numbers are a minimum.
This page summarizes reports by Search Engine Journal and Search Engine Roundtable and adds a measurement routine. CheaperStack did not run its own test of ChatGPT's links. Several of the analysts quoted sell SEO or AI search services, and some of the samples are not described, so read the numbers as early signals.
What was reported
- The drop. Graphite's CEO reported that GA4 sessions attributed to ChatGPT fell 22% to 57% from 9 October on sites with meaningful AI assistant traffic, about 45% from baseline. LLM Pulse measured about 46% fewer ChatGPT sessions on 9 to 11 October than on the same days in the prior two weeks, while total traffic stayed roughly flat.
- Where visits went. An Adobe analyst compared 9 to 13 October with the prior week on about 500 pages where ChatGPT drove at least 5% of visits: ChatGPT visits fell by about 31,000 a day and Direct rose by about 26,000 a day, while Search fell 7%. He called it early, from four days of sampled data. LLM Pulse did not observe a Direct increase.
- The likely cause. Inline links in ChatGPT's answer text stopped carrying the tag and a referrer. Profound reported that the share of cited answers with at least one tagged source link fell from about 98% on 8 October to 66% on 14 October, as a new answer format grew from 2% to about 35%. Peec AI's analyst found direct URLs in 42% of responses, up from 8% to 10%; those are displayed links, not measured clicks.
- The fix. The Adobe analyst reported that OpenAI confirmed a bug on 14 October and that tagged clicks were back to roughly 95% to 100% of the same hour a week earlier by the morning of 15 October. Search Engine Roundtable's report included no OpenAI statement, and it was updated on 15 October to say UTMs were returning. Profound saw no official confirmation, and OpenAI's release notes for 1 to 15 October do not mention link tracking.
What is not settled: whether it was a bug or a change, whether the fix applies to every account, and how many visits were lost rather than relabelled. The aggregated data cannot separate the two.
A measurement routine that survives this
- Annotate the window. Mark 8 to 15 October in GA4 or your analytics tool. One analyst now uses that range. Do not compare periods that straddle it.
- Group AI sources. In GA4, a custom channel group can match the source values you actually see in your reports (for example chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com). Check your own source list first; do not assume the values.
- Compare ChatGPT and Direct side by side for the pages ChatGPT used to send visitors to, and for the home page. A rise in Direct on deep URLs, not only the home page, is a hint, not proof.
- Count tagged requests in your server logs. They show requests whether or not the visitor accepted analytics. The commands are on our crawler log data page.
- Report a floor. Write "at least N tagged visits" with the date range and the method, and say that untagged visits are not counted.
- Leave ads out of it. One analyst said UTM parameters set on links from ChatGPT Ads should be unaffected; check your own setup.
What our own data shows
In our server log for 30 September to 18 October, 26 requests carried utm_source=chatgpt.com, from 16 addresses, on two pages. That window includes the days in question, so the figure is a floor. We have no matching number for untagged visits. Our analytics loads only after a visitor accepts it, so it undercounts further; the log is the more complete source, with its own limits (requests, not people). The full numbers and method are in AI crawlers in server logs.
What not to conclude
- That ChatGPT traffic fell by 45%. That is how much tagged traffic fell in some analysts' samples.
- That every Direct increase is ChatGPT. Direct has many causes.
- That the fix restored the old picture. It was reported for tagged clicks; most ChatGPT clicks still arrive without a referrer, according to the Adobe analyst.
The verdict, by situation
| If you… | Do this |
|---|---|
| Report AI traffic to clients | State a floor, the method and the 8 to 15 October window |
| Saw a ChatGPT dip in GA4 after 9 October | Check Direct on the same pages before calling it a loss |
| Have access to server logs | Count tagged requests as a cross-check |
| Run a small site | Track the trend monthly; do not act on a week of data |
| Want to know if AI engines can reach you at all | Start with the crawler access check |
Frequently asked questions
Did ChatGPT really stop sending traffic in October 2026?
Not necessarily. Analysts reported that from about 9 October, ChatGPT links stopped carrying utm_source=chatgpt.com and a referrer, so visits could land in Direct instead. One analysis of about 500 pages found ChatGPT visits down and Direct visits up, while another saw no rise in Direct. The data cannot separate lost visits from visits that lost their label.
Is the problem fixed?
One analyst reported that tagged clicks returned from 15 October to roughly 95% to 100% of the same hour a week earlier, and that OpenAI had confirmed a bug. Other reports found no official statement, and release notes for 1 to 15 October did not mention link tracking. Treat the fix as reported, not confirmed.
How do I find ChatGPT traffic that shows as Direct?
You cannot identify individual visits, but you can compare. Annotate 8 to 15 October, put ChatGPT and Direct side by side for the pages ChatGPT used to send visitors to, and look for Direct landing on deep URLs rather than the home page. Count requests with utm_source=chatgpt.com in your server logs as a floor.
Should I report ChatGPT traffic share to a client?
Only with the caveat. Tagged visits are a floor, and part of the dip in October was a labelling problem. State the date range, the method and the gap, and avoid before-and-after comparisons that straddle 9 to 15 October.
Does Google Analytics see all of this?
Only for visitors who accept analytics. On this site Google Analytics loads after consent, so server logs show more requests than analytics does. Use both and say which one a number comes from.
Sources
Read in October 2026: Search Engine Journal on the ChatGPT traffic drop and Search Engine Roundtable on the missing UTM parameters. Search Engine Land also covered it; we could not open that page, so none of the figures above come from it.
Related
AI crawlers in server logs
18 days of real log data from a new site, with the limits.
Search Console and Bing AI reports
What the free reports show, and what they leave out.
Semrush vs Ahrefs for AI visibility
What each tracks in AI answers, and what it costs.
Free ways to check your AI visibility
No-signup tools and manual methods.
Last reviewed . Reports on this change were still developing when this was written, so it is re-checked as they update. Spotted one out of date? Tell me.