utm_source=chatgpt.com tag, and because most ChatGPT clicks already arrive without a referrer, many of those visits were recorded as Direct. Adobe's Cedric Huesler says OpenAI told him it was a bug, and tagging was recovered by October 16. The visits most likely still happened, so October reports need an annotation to avoid undercounting ChatGPT traffic.Marketing teams opening their October analytics are finding a dip in ChatGPT referral traffic that starts around October 8 and recovers around October 16. Profound reported that teams saw declines of 30% to 40% in reported ChatGPT referrals during that window, and the evidence so far points to a labeling problem. ChatGPT appears to have kept sending visitors, but a growing share of those visits arrived carrying nothing that identified ChatGPT as the source.
The tag is back, so the immediate problem has passed, but two things have not. The October data will stay mislabeled unless someone corrects for it, and the episode exposed a more lasting weakness, because most ChatGPT clicks still arrive without a referrer and the UTM tag is now the only thing marking them as ChatGPT traffic.
In October, most of the missing ChatGPT traffic appears to have lost its label without leaving. For about eight days, many links in ChatGPT answers stopped carrying the utm_source=chatgpt.com tag, and without that tag or a referrer, analytics tools had no way to attribute the visit to ChatGPT.
Profound tracked the change by measuring the share of collected ChatGPT answers with cited links that included at least one tagged link. That share sat around 99% before October 8, fell to about 75% on October 9, and hovered around 5% to 12% for roughly 27 hours across October 14 and 15. It then recovered to about 99% by 06:30 UTC on October 16. Profound attributed the missing tags to Intelligent UI, an answer format introduced with GPT-6, which appeared in about a quarter of collected answers on October 9. The company also cautions that its percentages describe tagging in the answers it collected, so they should not be used as a multiplier to correct traffic or revenue figures.
OpenAI's explanation has reached the public only secondhand, through one person's account. Adobe's Cedric Huesler, who tracks ChatGPT clicks across thousands of domains, said OpenAI told him on October 14 that the missing tag was a bug, and that a fix began rolling out early on October 15, according to Search Engine Journal. By 9 a.m. Pacific that day, he reported tagged ChatGPT clicks at roughly 95% to 100% of the same hour a week earlier across the domains he tracks. Glenn Gabe replied that he was not yet seeing the change on paid or free accounts, which suggests the rollout took time to reach all users. OpenAI has not commented publicly, and its release notes for October 1 through 15 do not mention link tracking.
On the pages Huesler analyzed, Direct traffic rose by nearly as much as ChatGPT traffic fell. He looked at about 500 pages where ChatGPT had supplied at least 5% of visits. Comparing October 9 through 13 with the prior week, Search Engine Roundtable reported these results:
Typed and bookmarked visits tend to come from returning visitors, so a jump in first-time visitors within Direct fits Huesler's explanation that ChatGPT visits were being relabeled. The data still has limits, since it covers five days, comes from one analyst, and draws on pages chosen because ChatGPT already sent them meaningful traffic. The roughly 5,000 daily visits that did not appear in Direct may have landed elsewhere, since Profound notes that a missing UTM does not always mean a visit becomes Direct, or they may reflect a real decline. Sébastien Billard reported that ChatGPT human traffic and ChatGPT-User activity, the agent that fetches pages during a conversation, fell on his site after October 9, while visits from OpenAI's search crawler rose. He is still investigating, and one site's logs cannot settle whether citations fell.
The returning tag fixed the October problem, but the referrer gap means the same problem can recur whenever the tag slips. Huesler said that most ChatGPT clicks still arrive without a referrer, and he reported that ChatGPT opens links with a browser instruction that tells the browser to withhold the referring page. When that happens, the UTM tag is the only signal left that names ChatGPT as the source.
That has a practical consequence for how AI traffic is classified. Google added an AI Assistant channel to GA4's Default Channel Group in May 2026, and Google's channel definitions place a visit there when its medium is ai-assistant or when its referrer matches Google's list of AI assistants. The rule does not mention utm_source, and the tag ChatGPT adds sets the source rather than the medium, so a ChatGPT visit that arrives tagged but without a referrer may never reach the AI Assistant channel. Custom reports built on a chatgpt.com referrer share that blind spot, whereas rules that read the session source, which GA4 takes from utm_source when the tag is present, will pick tagged visits up. Teams that report ChatGPT traffic should check where tagged ChatGPT sessions land in their own setup.
Gemini's links are changing as well, though in the opposite direction. On October 1, only a week before ChatGPT's tags began to slip, Search Engine Journal reported that Gemini appears to have started adding UTM parameters to some referral links, a change Google has not documented. Each assistant controls how its links are labeled, and any of those choices can change without notice.
The correction has two goals: making sure nobody reads the October dip as lost visibility, and confirming that ChatGPT tracking has recovered in your own data.
chatgpt.com if reports rely on the referrer alone.Reading the results of those checks comes down to a few common patterns.
| What the data shows after October 16 | Likely explanation | What to do next |
|---|---|---|
| ChatGPT sessions back near pre-October 8 levels, citations steady | The dip was a labeling gap | Annotate October and move on |
| ChatGPT sessions still low, Direct elevated on the same pages | Reports may rely on the referrer, or the fix has not reached your visitors yet | Check channel rules, then recheck in a week |
| ChatGPT sessions still low, Direct flat, citations steady | Visits may be landing in another channel, or volumes are too small to read | Review other channels and extend the window |
| ChatGPT sessions still low, citations down | Likely a real decline in ChatGPT visibility | Review which pages lost citations and why |
| ChatGPT sessions back to normal, citations down | Tracking recovered, but visibility may be slipping | Review lost citations before the decline reaches traffic |
| Direct also up on pages ChatGPT never sent visitors to | Something else is driving Direct traffic | Rule out campaigns, offline promotion, or tracking changes first |
The episode shows how fragile any AI measurement plan becomes when it depends on one platform's tags. For about eight days, reports that identified ChatGPT visits by tag or referrer were likely to show a drop, and the larger risk sat in revenue reporting, where credit for a sale could quietly shift to another channel. When we traced ChatGPT-assisted revenue for a B2B distributor, 49% of it came from orders where ChatGPT was not the final touch, which is why we read revenue on a multi-touch basis and never from referral reports alone.
For our clients, we are handling the gap in four ways:
That is why we measure AI search in three layers that check one another: exposure in AI answers, behavior on the site, and the revenue that follows. Exposure is measured inside the answers themselves, and Galileo, our multi-touch attribution platform, connects site behavior to the revenue those journeys produce.
October's dip in ChatGPT traffic most likely reflects about eight days of missing tags, a bug in Huesler's account that has since been fixed, although part of the drop may still turn out to be real for some sites. The practical response is to annotate October 8 through 16, confirm that tagged ChatGPT traffic has recovered in your own data, and check that reports identify ChatGPT by source instead of referrer. With the referrer still missing on most clicks, the tag is the only label ChatGPT traffic carries, which is why AI search measurement needs signals that do not depend on any one platform's choices.
To tie AI visibility to pipeline and revenue, see how Galileo maps AI-assisted journeys. To find out whether ChatGPT and other AI assistants are recommending your brand, talk to our SEO, AEO, and GEO team.