ChatGPT Tracking Failure Exposes AI Traffic Attribution Risks
Summary
The Bigger Challenge of AI Traffic Attribution
Starting around October 8 to 9, as GPT-6 and its new Intelligent UI format rolled out, many ChatGPT links reached websites without the utm_source=chatgpt.com tag or any referrer data. Analytics filed those visits as direct, so ChatGPT referrals appeared to collapse.
Adobe’s Cedric Huesler studied about 500 pages and found ChatGPT visits down roughly 31,000 a day and direct visits up roughly 26,000 a day, while search fell 7%. Most of the drop looks like relabeling, though early data can’t rule out some lost visits. Huesler says OpenAI told him it was a bug, a fix began rolling out on October 15, and Profound measured tags back on about 99% of cited answers by October 16. The October 8 to 15 history isn’t repaired, and the referrer is still missing on most ChatGPT clicks, which leaves the UTM tag as the only signal separating ChatGPT visits from true direct traffic.
Why this matters
The Cost of Mistaking a Tracking Failure for a Performance Decline
Reports are skewed. Anything covering that week understates ChatGPT and overstates direct.
Bad calls are easy to make. A team reading the numbers at face value could decide AI discovery stalled or content stopped working, then cut what was performing.
The ROI case takes a hit. Referral data is the evidence behind AI visibility spend, and conversions follow the visit, so sales from ChatGPT recommendations get credited to direct.
The setup is fragile. One tag now carries the attribution load. Graphite found that 95% of mobile ChatGPT visits to its sites relied on the tag alone, which made app traffic the most exposed.
How ChatGPT Referrals Became Direct Traffic
ChatGPT has tagged its links for some time. Search Engine Journal’s hands-on testing at the October 2024 launch of ChatGPT search found that every citation carried utm_source=chatgpt.com. In June 2025, according to Search Engine Roundtable, OpenAI extended the tag to the “More” sources, which had been plain links. The tag mattered because it worked as a backstop. When a browser withheld the referrer, the tag could still tell analytics where the visit came from.
Barry Schwartz of Search Engine Roundtable reported on October 14 that ChatGPT links appeared to have lost that tag. Huesler, a senior director of product management at Adobe, explained the likely mechanism on LinkedIn. GPT-6 opens links with a browser setting that withholds the referrer, and the tag was gone too, so analytics had nothing to work with and filed the visit under direct.
Not every link was affected. Ethan Smith, CEO of Graphite, reported that inline links in answer text lost the tag while links in the Sources panel and shopping links kept it. Profound found the same for links reached through product or information panels.
Profound tracked the tagged share of cited ChatGPT answers. It started falling on October 8, stood at roughly two thirds by October 14, and dropped to around 5 to 12% that evening, where it stayed for about 27 hours.
Huesler compared October 9 to 13 with the week before across pages where ChatGPT had driven at least 5% of visits. Direct rose on two thirds of the pages that lost ChatGPT referrals, against about half the week before. Smith saw ChatGPT-attributed GA4 sessions fall between 22% and 57% on sites Graphite tracks, roughly 45% overall.
The data has limits. Huesler himself called it early, based on four days of sampled data. About 5,000 daily visits are unaccounted for between the 31,000 lost and 26,000 gained. LLM Pulse noted that direct is so much larger than ChatGPT that a small shift is hard to see, and aggregated data can’t separate lost visits from unlabeled ones. Several of the firms behind these measurements sell AI search tools, so the figures are best read as directional.
When Direct Traffic Rises for the Wrong Reasons
Direct has always been the junk drawer of analytics. Typed URLs, bookmarks and clicks from apps and documents that strip referrer data all end up there, along with anything else the platform can’t explain. That makes AI traffic attribution particularly vulnerable when tracking signals disappear. When ChatGPT visits join the pile, direct looks healthier than it is and ChatGPT looks weaker than it is.The two lines move in opposite directions, which makes the pattern look like a genuine change in behavior and invites a story to explain it. Brand strength is up. AI discovery has cooled.
Tracking changes rarely announce themselves. No alert fires, the dashboard stays tidy and the report goes out on schedule. The cost arrives later, when content gets cut because it seems to have lost its audience, budget moves away from work that was performing, or a board hears an explanation for a decline that never happened.
Any AI referral line read on its own is a floor. The more dependable signal is referral and direct read together on the same landing pages, and that comparison is what pointed to a tracking problem in October.
The Tracking Fix Restored Visibility, Not the Missing History
Huesler says OpenAI told him on October 14 that the missing tag was a known bug, and a fix began rolling out early on October 15. By 9am Pacific that day he saw tagged ChatGPT clicks back to roughly 95 to 100% of the same hour a week earlier across thousands of domains. Profound’s sampling of cited answers reached about 99% tagged by the morning of October 16.
The recovery was gradual. SEO consultant Glenn Gabe reported he wasn’t yet seeing the change on paid or free accounts. OpenAI’s release notes for October 1 through 15 don’t mention link tracking, so the confirmation remains Huesler’s account.
Two issues survive. Profound notes that restored tags help future visits but don’t automatically recover source information missing from earlier ones, so any report spanning October 8 to 15 carries the distortion. Huesler also says most ChatGPT clicks still arrive without a referrer, which is why he recommends counting by the UTM tag.
How to Protect Reporting from Attribution Gaps
Teams can annotate October 8 to 15 in GA4 and any other reporting tool, converting the dates to the property’s time zone. They can test their own pages by checking whether direct rose where ChatGPT fell, especially on deeper pages ChatGPT used to send people to. Tracking the share of ChatGPT sessions carrying the tag gives an early warning, since a repeat would show up there first. A dated log of platform changes next to the reporting is what explains the gap to someone reading the trend six months from now.
AI Search Needs Measurement Leaders Can Trust
AI assistants have become a real discovery channel, and one change in how a platform opens links can distort a quarter of reporting. Platforms are also diverging. Search Engine Journal reported on October 15 that Gemini had added UTM parameters to its links about two weeks earlier, so one assistant was adding attribution while another lost it.
That fragility won’t resolve soon. Marketers can’t control how platforms handle links, but they do control how the data is read. Annotating changes, reading referral and direct as a pair, tagging their own links consistently and holding decisions during anomalies are all within reach. An unexplained shift deserves an investigation before anyone acts on it. Teams that work that way stay credible with finance while the instruments keep moving.
Frequently Asked Questions
What happened to ChatGPT referral traffic in October 2026?
Starting around October 8 to 9, many ChatGPT links began reaching websites without the utm_source=chatgpt.com tag and without referrer data, so analytics recorded the visits as direct. Graphite found ChatGPT-attributed GA4 sessions down 22% to 57% on the sites it tracks, roughly 45% overall.
Did ChatGPT lose visitors, or were they mislabeled?
The evidence points mostly to mislabeling. Adobe found ChatGPT visits down about 31,000 a day and direct up about 26,000 on the same pages, with search down only 7%. Huesler also found the new direct visitors behaved like past ChatGPT visitors. Search Engine Journal cautioned that there isn’t yet enough data to rule out some lost visitation, and LLM Pulse noted its data couldn’t separate lost visits from unlabeled ones.
Did OpenAI confirm it was a bug?
Huesler says OpenAI told him on October 14 that it was a known bug, and he posted that ChatGPT was tagging links again on October 15. Search Engine Journal reports that OpenAI’s release notes for October 1 through 15 don’t mention link tracking, and Profound said it hadn’t seen official word. As far as published reporting shows, the confirmation is Huesler’s account.
Are UTM tags back on ChatGPT links?
Yes, according to Huesler and Profound. Huesler saw tagged clicks back to roughly 95 to 100% of the prior week’s level by 9am Pacific on October 15, and Profound found about 99% of cited answers carrying a tagged link by the morning of October 16. The return was gradual, and at least one consultant had not yet seen it on some accounts on October 15.
How long has ChatGPT tagged its links?
Citation links have carried utm_source=chatgpt.com since ChatGPT search launched in October 2024, according to Search Engine Journal’s testing at the time. Search Engine Roundtable reported in June 2025 that OpenAI extended the tag to the “More” sources, which previously had none.
Which ChatGPT links lost the tag?
Smith reported that inline links in answer text lost it, while links in the Sources panel and shopping links kept it. Profound found that links reached through an additional product or information panel still carried tags. Attribution could therefore depend on which link a person clicked, even within the same answer.
Does restoring the tags fix the October numbers?
No. Profound notes that the return of the tags improves reporting for new visits but doesn’t automatically recover source information missing from earlier ones. Huesler places the affected window at October 8 to 15 and recommends annotating it in reports so no one mistakes the dip for lost traffic.
How can a site tell whether it was affected?
Start with landing pages where ChatGPT drove a meaningful share of visits. Compare ChatGPT and direct sessions on those pages before and after October 9. A ChatGPT drop paired with a direct rise while search holds steady matches the pattern Adobe found. Huesler also suggests watching direct on the deeper pages ChatGPT used to send people to.
Were ChatGPT Ads or other platforms affected?
Smith said clicks on ChatGPT Ads campaigns with their own UTM parameters should be unaffected. Search Engine Journal also reported on October 15 that Gemini had added UTM parameters to its links about two weeks earlier.
Should marketers change their AI strategy because of this?
Not on this data alone. Search Engine Land advised treating sudden ChatGPT referral drops with caution until the tracking behavior is clear. Profound said it wouldn’t change strategy because the ChatGPT referral number alone fell, and would apply the same caution to a sudden increase as tracking returned.
Could it happen again?
It’s possible. The referrer is still missing on most ChatGPT clicks, according to Huesler, so the UTM tag is the signal separating ChatGPT visits from direct traffic. Tracking the tagged share of ChatGPT sessions would give an early warning, and a dated log of platform changes would explain any future break
Sources:
Search Engine Roundtable, OpenAI Drops UTM Parameters From ChatGPT Results (GPT-6), Barry Schwartz, October 14, 2026, updated October 15.
Search Engine Roundtable, ChatGPT Updates To Improve Tracking In Analytics With More UTM Parameters, Barry Schwartz, June 16, 2025.
Search Engine Land, ChatGPT appears to drop UTM tracking, making AI referral traffic harder to measure, Anu Adegbola, October 15, 2026.
Search Engine Journal, ChatGPT Traffic Drop Appears To Be A Tracking Problem, Matt G. Southern, October 15, 2026.
Search Engine Journal, Rank Tracker Data, ChatGPT Tags & AI Search Links (SEO Pulse), Matt G. Southern, October 17, 2026.
Search Engine Journal, SearchGPT Launches As ChatGPT’s Real-Time Search Feature, Matt G. Southern, October 31, 2024.
Profound, ChatGPT’s missing UTM tags: what happened from October 8 to October 16, Josh Blyskal, October 2026.
Cedric Huesler, Adobe, and Ethan Smith, Graphite, LinkedIn posts, October 2026, as reported by the outlets above.
Comments