ObsurfableIf your analytics show ChatGPT referral traffic falling off a cliff after October 9, 2026, check Direct before you rewrite your AEO strategy.
Search Engine Roundtable reports that OpenAI appears to have removed or replaced the UTM parameters ChatGPT used to attach to outbound links — a change that lines up with the GPT-6 consumer rollout a week earlier. Without the tag, and with a stricter link open that suppresses the referrer, those visits land in Direct.
That is an attribution problem, not proof that ChatGPT stopped sending people to your site.
What changed
In June 2026, OpenAI added UTM tracking to ChatGPT links — typically utm_source=chatgpt.com — so analytics platforms could separate ChatGPT clicks from other traffic.
With GPT-6, that signal appears gone. Cedric Huesler at Adobe described the new behavior on LinkedIn: GPT-6 opens links with window.open(url, "_blank", "noopener,noreferrer") and drops ?utm_source=chatgpt.com. No UTM tag. No referrer. Analytics files the visit under Direct.
Barry Schwartz's write-up frames it as a tracking change that started around the GPT-6 release. Whether it was deliberate product policy or a side effect of how GPT-6 opens links, the measurement outcome is the same: ChatGPT-sourced clicks no longer announce themselves.
The Adobe numbers
Huesler checked roughly 500 pages where ChatGPT previously drove 5%+ of visits, comparing October 9–13 with the week before:
| Signal | Change |
|---|---|
| ChatGPT-attributed visits | −31k per day |
| Direct visits (same pages) | +26k per day |
| Search on those pages | −7% (matched the broader site) |
Roughly 84% of the lost ChatGPT volume shows up as new Direct traffic on the same URLs. Direct rose on 2 in 3 pages that lost ChatGPT traffic — the week before, that split was closer to a coin flip.
Behavior matches too. On mobile, the share of Direct visitors seeing a consent banner (new to the site) rose from 36% to 44%. For known ChatGPT visitors, that share sits around 55%. The new Direct cohort looks more like ChatGPT traffic than like classic bookmark/type-in Direct.
Other practitioners reported the same week: a sharp drop in ChatGPT human traffic (and in some cases chatgpt-user robot traffic), while OAI-SearchBot crawl volume increased. The human-click path went dark in analytics; the crawl path did not.
Why this matters for AEO
AI referral traffic was never a perfect proxy for visibility. Most users get the answer in-chat and never click. But UTM-tagged ChatGPT sessions were one of the few downstream signals teams could put next to mention and citation rates.
That signal just got noisier:
- Dashboards will undercount ChatGPT. Any KPI built on
utm_source=chatgpt.comor achatgpt.comreferrer will show a false decline after ~October 9. - Direct will absorb the residue. Teams that treat Direct as "unattributable brand" will misread a GPT-6 tracking change as brand lift or as organic demand.
- Cross-period comparisons break. Week-over-week ChatGPT traffic before and after GPT-6 is not an apples-to-apples series unless you rebuild attribution.
- Citation and mention tracking matter more. When click attribution fails, prompt-level mention and citation measurement is the layer that still tells you whether ChatGPT recommends you.
This is the same measurement trap we already see elsewhere in AI search: the system influences the buyer, then hides or softens the trail. GPT-6 just made the trail darker on the click path.
What to do now
1. Do not treat the ChatGPT traffic drop as a visibility collapse.
Reconcile ChatGPT and Direct on URLs that historically received ChatGPT clicks. If Direct rises in lockstep with ChatGPT falling, you are looking at reclassification.
2. Rebuild AI referral detection.
UTM tags were convenient. Without them, options are thinner:
- Watch for residual referrer or landing-page patterns if any still leak through (do not assume they will)
- Segment Direct by landing pages that previously converted ChatGPT traffic
- Compare consent-banner / new-visitor rates on those pages against historical ChatGPT cohorts
- Keep platform-specific UTM tracking where other AI engines still pass tags (e.g. Perplexity), and document ChatGPT as a special case
3. Separate crawl signals from human click signals.
An uptick in OAI-SearchBot does not replace lost referral reporting. Crawl eligibility still matters for retrieval; human click attribution answers a different question.
4. Keep answer-layer KPIs as the primary AEO scoreboard.
Track mention rate, citation rate, share of voice, and recommendation framing on a fixed prompt set. ChatGPT can influence a purchase without ever sending a tagged session. Referral traffic is a secondary, incomplete readout — and after GPT-6, even more incomplete.
Bottom line
GPT-6 did not just change how ChatGPT searches and answers. It appears to have changed how ChatGPT hands off to your site. The June UTM layer made ChatGPT clicks visible. The October behavior makes many of those same clicks look like Direct.
If ChatGPT "traffic" cratered after October 9 while Direct rose on the same pages, measure the reallocation before you change your content strategy. Visibility and attribution are not the same thing — and ChatGPT just reminded the market how quickly the click trail can disappear.