ChatGPT-attributed sessions in Google Analytics 4 fell 22% to 57% across sites tracked by Graphite starting Oct. 9, and OpenAI has since confirmed the drop was a bug in how ChatGPT tags its links, not a collapse in visitor interest.

Graphite CEO Ethan Smith, writing on LinkedIn, put the decline at roughly 45% from baseline for sites with meaningful traffic in the AI Assistant channel (GA4's reporting bucket for AI-driven referrals). LLM Pulse, monitoring its own site set, measured sessions from ChatGPT down about 46% during Oct. 9–11 versus the same days in the prior two weeks. Total traffic on those sites stayed flat over the same period — the first signal that visits weren't vanishing, just changing labels.

What broke in ChatGPT's links?

Smith's diagnosis: ChatGPT started opening links in answer text without the utm_source=chatgpt.com tag (the URL parameter that tells analytics where a visit came from) or a referrer. Sessions from those links therefore land in GA4's Direct channel, which is where unattributed traffic goes.

The timing tracks the GPT-6 rollout. OpenAI released GPT-6 to paid plans on Oct. 7 and to Free and Go users on Oct. 8; LLM Pulse saw no drop on Oct. 7. Adobe senior director of product management Cedric Huesler says that after the rollout, ChatGPT began opening links in a new tab with a browser setting that withholds the referrer, and dropped the utm_source tag its citations had carried since ChatGPT search launched in 2024.

Smith added a detail with real teeth for mobile publishers: 95% of mobile ChatGPT visits to Graphite's sites relied on the tag alone, making app traffic the most exposed. Links in the Sources panel and shopping links kept their tags.

Where did the visits go?

Huesler compared Oct. 9–13 with the prior week across roughly 500 pages where ChatGPT drove at least 5% of visits. ChatGPT visits fell by about 31,000 a day. Direct visits rose by about 26,000 a day. Search traffic fell 7%, in line with search trends elsewhere.

Direct rose on two of every three pages that lost ChatGPT traffic, versus about half the week before. The behavioral fingerprint matched too: on mobile, the share of Direct visitors seeing the consent banner — Huesler's proxy for first visits — climbed from 36% to 44%, moving toward the 55% he measures among ChatGPT visitors. He calls the findings early, built on four days of sampled data.

Profound, an AI search analytics vendor, ties the change to a new answer format. Among ChatGPT answers with citations that it collected, the share carrying at least one tagged source link fell from about 98% on Oct. 8 to 66% by Oct. 14. Over the same days, Intelligent UI — ChatGPT's format for charts, comparisons and small tools — jumped from 2% of collected answers to about 35%. Links opened through product or information panels kept their tags.

What the prompt-level testing showed

Tomek Rudzki at Peec AI ran the same 1 million prompts from Oct. 2–7 and again from Oct. 9–13. ChatGPT placed a direct URL for a webpage in 42% of responses, up from roughly 8–10% previously. He reports 54% of those URLs point to home pages, and that the links carry no tags or referrers. These figures count links displayed in answers, not actual user click-throughs.

Not every dataset agrees, though. LLM Pulse's authors said they did not observe a Direct increase, noting Direct is much larger than ChatGPT so a small influx would be hard to detect. Their aggregated data also couldn't separate traffic genuinely lost from traffic simply unlabeled. LLM Pulse and Profound did not publish sample sizes.

Was it confirmed and fixed?

Smith called the change a bug and wrote he expects OpenAI to fix it "very soon." Huesler said OpenAI confirmed the bug to him on Oct. 14, and in a follow-up post reported that a fix began rolling out early Oct. 15. By 9 a.m. PT, tagged ChatGPT clicks had returned to roughly 95–100% of the same hour a week earlier across thousands of domains he tracks.

The rollout isn't universal yet. SEO consultant Glenn Gabe replied that he wasn't seeing the fix on paid or free accounts. Profound says it has seen no official word on a fix, and OpenAI's ChatGPT release notes for Oct. 1–15 don't mention link tracking at all.

How should site owners handle the data?

Smith and Huesler both recommend annotating Oct. 8 — Huesler now frames the window as Oct. 8–15 — in analytics reports before anyone reads the dip as lost traffic. Until the fix fully lands, review ChatGPT and Direct traffic side by side for each page ChatGPT referred visitors to, plus the homepage, where Rudzki found more than half of direct links point.

Publishers running ChatGPT Ads with their own UTM parameters should be unaffected, per Smith. Worth remembering: when Google Analytics introduced the AI Assistant channel in May, unreferrer'd AI-assistant sessions were always destined for Direct. The tag was the only thing capturing that traffic, especially on mobile.

The open question now is how quickly tags return across every account, given OpenAI's silence in its release notes — Profound notes a restoration would aid future reporting but can't recover attribution already lost. Monitor your GA4 AI Assistant channel daily against Direct through the end of October.