ChatGPT UTM tag loss was a bug, Adobe's Huesler says; fix rolling out

The drop in ChatGPT traffic that analytics tools recorded from 8 October 2026 was a tagging bug, not lost visitors, according to Adobe's Cedric Huesler, who said on 15 October that OpenAI confirmed the problem to him and that a fix began rolling out early that day. By 9am Pacific time on 15 October, Huesler said, links from ChatGPT were again carrying utm_source=chatgpt.com, and tagged ChatGPT clicks were back to roughly 95% to 100% of the same hour a week earlier across thousands of domains in Adobe's data. The confirmation is secondhand: OpenAI has not said anything publicly, and its ChatGPT release notes do not mention the bug or the fix. The referrer is still missing on most ChatGPT clicks, so the UTM tag remains the signal to count.

What is new since Relevant Audience's 15 October report

Relevant Audience covered the first stage of this story on 15 October 2026 in "ChatGPT appears to drop UTM tags and referrer, sending clicks to Direct (unconfirmed)". That post was built on a Search Engine Roundtable report from 14 October and on Huesler's first LinkedIn analysis of about 500 pages. At that point nobody knew whether the change was intentional, permanent or a bug, and OpenAI had said nothing.

Three things have changed since then:

• The cause: Huesler now says OpenAI told him on 14 October that it was a known bug. Before, the cause was open.
• The fix: Huesler says a fix started rolling out at about 2am Pacific time on 15 October, in stages at first and steady by the morning, and that the UTM tag is back on ChatGPT links.
• The annotation window: the period to mark in reports is now 8 to 15 October, as Huesler puts it. The earlier post suggested annotating around 8 to 10 October, because nobody knew then when or whether the tag would return.

What has not changed: the referrer. Huesler wrote that it is still missing on most ChatGPT clicks. And the confirmation still does not come from OpenAI directly.

How the confirmation reached the public

Search Engine Journal's Matt G. Southern pulled the strands together on 15 October in Search Engine Journal, under the headline that the ChatGPT traffic drop appears to be a tracking problem. SEJ reports that Huesler "now says OpenAI confirmed to him on Oct. 14 that it was a bug", and that in a follow-up post he wrote that a fix began rolling out early on 15 October.

Huesler's own follow-up on LinkedIn, published on 15 October under the opening line "Relieved. It was a bug.", gives the fuller timeline. He wrote that he asked OpenAI what was going on at 9am Pacific time on Wednesday 14 October, and that by 11am OpenAI confirmed it was a known bug with a fix due that week. He thanked "the folks at OpenAI" for the quick turnaround.

That is the whole chain of evidence. OpenAI told one person at Adobe, and that person said so publicly. It is credible, because Huesler's earlier data matched what several other trackers saw and his recovery figures come from Adobe telemetry, but it is not a statement from OpenAI. SEJ notes that Profound, one of the AI-visibility tool vendors tracking the issue, said it had not seen official word on a fix, and that OpenAI's ChatGPT release notes for 1 to 15 October do not mention link tracking. Relevant Audience checked the release notes page on 16 October and found the same: the entries dated 1 to 15 October cover GPT-6, Intelligent UI and other launches, but none mentions UTM tags or link tracking.

The timeline, as Huesler gives it

Huesler's LinkedIn post sets out the week hour by hour. All times are Pacific time, as he wrote them:

• Thursday 8 October, about 2pm: GPT-6 stops adding the tag. ChatGPT visits that could be attributed fall to about a third of normal within hours.
• Tuesday 13 October: Huesler posts about a fall of about 60%. His first guess was that GPT-6 shows fewer links.
• The same evening: he concludes the visits were hidden rather than lost, with no tag and no referrer, so analytics counted them as Direct.
• Wednesday 14 October, 9am: he asks OpenAI. By 11am, he says, OpenAI confirms a known bug with a fix due that week.
• Thursday 15 October, about 2am: the fix starts rolling out, in stages at first and steady by the morning.
• Thursday 15 October, 9am: tagged ChatGPT clicks back to roughly 95% to 100% of the same hour a week earlier, across thousands of domains.

SEJ adds that GPT-6 began rolling out to paid ChatGPT plans on 7 October and to Free and Go users on 8 October, and that LLM Pulse saw no drop on 7 October. That fits Huesler's 8 October start time.

What the other trackers saw during the week

The confirmation carries weight because the drop was large and showed up in several independent datasets. SEJ's report lists them, and the table below sets out the main figures with who reported each one. All of these come from companies that sell SEO or AI search tools, or in Huesler's case from Adobe, which SEJ points out; LLM Pulse and Profound did not give sample sizes.

SEJ also reports details on which links lost the tag. One analyst it quotes found that inline links in answer text lost the tag while links in the Sources panel and shopping links kept it, and that 95% of mobile ChatGPT visits to the sites in that dataset relied on the tag alone, which made app traffic the most exposed. Profound tied the change to Intelligent UI, ChatGPT's format for charts, comparisons and small tools. Rudzki's figures are about links ChatGPT displayed, not clicks.

LLM Pulse added a caution that still applies: it did not see a measurable rise in Direct traffic, because Direct is much larger than ChatGPT, and its aggregated data could not separate lost visits from relabelled ones. Huesler's page-level data, which showed Direct rising on two of every three pages that lost ChatGPT traffic, is the strongest evidence that the visits were relabelled. The bug confirmation is consistent with that reading.

What is still not settled

Four points are still open, and each one affects October reporting.

The referrer is still missing

Huesler wrote on 15 October that the referrer is still missing on most ChatGPT clicks, so the UTM tag is what analysts can rely on. Some background, which is general web behaviour rather than anything in the sources: GA4 assigns a session's source first from campaign parameters such as utm_source and then from the referrer. With the tag back, ChatGPT sessions should again be labelled as ChatGPT even without a referrer. Any link that loses its parameters, for example when a user copies and shares a stripped URL, would fall back to Direct. Nobody has said whether OpenAI plans to restore the referrer.

Not every account may have the fix yet

SEJ reports that SEO consultant Glenn Gabe replied to Huesler that he was not yet seeing the change on paid or free accounts. Huesler described a staged rollout. SEJ calls the open question how quickly the tags return for every account. A site that still sees a low ChatGPT line after 15 October may be waiting on the rollout rather than seeing a different problem.

Lost attribution does not come back

Profound notes, according to SEJ, that restoring the tags helps future reports but does not bring back attribution already lost. Sessions recorded as Direct between 8 and 15 October stay Direct in GA4. No backfill has been mentioned by anyone.

OpenAI has said nothing in public

There is no OpenAI blog post, help article or release note about the bug. If OpenAI later describes it differently, the record will need updating.

What to do in your October reports (analysis)

The steps below are Relevant Audience analysis based on Huesler's three recommendations and SEJ's reporting. They are not instructions from OpenAI.

1. Annotate 8 to 15 October in GA4 and any reporting tool. Huesler's first recommendation is to mark the window so the ChatGPT dip and the Direct bump read as one event, not a trend.
2. Check the dates in your property's time zone. Huesler's times are Pacific. For a GA4 property set to Bangkok time, which is 14 hours ahead of Pacific daylight time in October, 2pm on 8 October Pacific is 4am on 9 October in Bangkok, and 2am on 15 October Pacific is 4pm on 15 October in Bangkok. So a Bangkok-time report should show the break from 9 October, with 16 October the first full day after the fix began. This is a conversion, not something the sources state.
3. Do not rebuild attribution models around the dip. Huesler's second point is that the visits never went away. Changing channel groupings, AI-traffic definitions or targets because of one bad week would bake a bug into the reporting.
4. Keep counting the UTM, not the referrer. If a custom channel group or Looker Studio filter for AI traffic depends on the referrer alone, it will keep under-counting. Matching on the source value chatgpt.com, which comes from the tag, is the more reliable route while the referrer is missing.
5. Report ChatGPT and Direct together for the affected week. SEJ suggests reviewing ChatGPT and Direct side by side for each page ChatGPT referred visitors to, plus the home page, where Rudzki found more than half of the direct links pointed. For a month-on-month comparison, showing the combined figure for those pages is more honest than a ChatGPT line with a hole in it.
6. Confirm the recovery on your own site. Compare chatgpt.com sessions from 16 October onward with the same weekdays in late September. If they have not recovered, Gabe's comment suggests the rollout may simply not have reached every account yet.
7. Treat ChatGPT Ads separately. SEJ reports one analyst's view that clicks from ChatGPT Ads carrying the advertiser's own UTM parameters should be unaffected. Advertisers who use their own tags can check that their paid ChatGPT line held steady through the week.

Teams that report AI visibility as a business result, for example as part of ChatGPT SEO work, have a specific reason to fix the record. A month that shows ChatGPT referrals down by half looks like a failure of the work when it was a labelling fault.

What the week showed about AI referral tracking

This is analysis. The week showed how much AI referral reporting depends on one parameter that a platform adds and can stop adding without notice. SEJ notes that when Google Analytics introduced its AI Assistant channel in May, sessions from an AI assistant without a referrer would also land in Direct; the ChatGPT tag was what let analysts catch that traffic, especially on mobile. SEJ also notes that Gemini went the other way about two weeks earlier, adding UTM parameters to its links.

For any site that tracks AI answer engines as a channel, a simple weekly check of chatgpt.com sessions against Direct on deep pages would have flagged this break within days. That kind of check belongs in the same routine as the rest of GEO reporting, because the platforms that cite a site are also the ones that decide whether the visit can be counted.

What this means for Thai marketers

None of the sources names Thailand or any other country, and Huesler's data is described as thousands of domains, not a regional sample. Thai sites that track ChatGPT referrals in GA4 should assume they were exposed to the same break unless their own data says otherwise, and should annotate the window in the property's own time zone, which for most Thai properties means 9 to 15 October. Thai-language reports that compare October with September, or that present AI traffic as its own line for management, need the footnote before they go out.

FAQ

Did OpenAI confirm the ChatGPT UTM bug?

Not publicly; the confirmation is secondhand. Adobe's Cedric Huesler said on 15 October 2026 that OpenAI confirmed to him on 14 October that the missing utm_source=chatgpt.com tag was a known bug, and Search Engine Journal reported his account. OpenAI's ChatGPT release notes for 1 to 15 October do not mention it, according to SEJ.

Is ChatGPT adding utm_source=chatgpt.com to links again?

Yes, according to Huesler, who said a fix began rolling out at about 2am Pacific time on 15 October and that by 9am tagged ChatGPT clicks were back to roughly 95% to 100% of a week earlier across thousands of domains. Glenn Gabe said he was not yet seeing the change, so the rollout may be uneven.

Which dates should I annotate in GA4?

Annotate 8 to 15 October 2026, the window Huesler recommends. His times are Pacific, so in a GA4 property set to Bangkok time the break starts on 9 October and the fix began on the afternoon of 15 October.

Will GA4 recover the ChatGPT sessions that were counted as Direct?

No source says it will. Profound noted, according to SEJ, that restoring the tags helps future reports but does not bring back attribution already lost, so those sessions stay in Direct.

Is the referrer back too?

No, Huesler wrote on 15 October that the referrer is still missing on most ChatGPT clicks. He recommends counting the UTM tag, not the referrer.

The ChatGPT dip in October reports was a labelling fault that has now been fixed, at least for most traffic, and the right response is a note on the chart rather than a new attribution model. If your AI-traffic reporting depends on referrers or needs a cleaner channel setup, Relevant Audience's GA4 team can review it.