The Maze: Website owners are seeing fewer visits credited to ChatGPT, OpenAI’s AI assistant, after some links stopped identifying where visitors came from. Evidence from several countries points to missing tracking information. It does not establish that every apparent lost visit survived, or that a reported repair has reached everyone.
The Details:
The label can disappear before the visitor does. A website normally identifies a visitor’s source from information supplied with the click. A source marker appended to a link can name ChatGPT. A referrer tells the website which page sent the visitor. When both are absent, analytics software may classify the visit as “Direct”: an unidentified source, not necessarily someone typing the address. Profound, a company that monitors brands in AI answers, found missing markers in ChatGPT’s newer interactive answers. Links reached through additional product or information panels sometimes retained them. The route through an answer can therefore change how the same destination gets credit.
The measurements describe different things. LLM Pulse, an AI-search monitoring service, saw ChatGPT-attributed daily sessions fall about 46% on monitored sites during October 9–11, compared with the same weekdays in the previous two weeks. Total traffic stayed roughly stable. Its aggregate data cannot separate fewer visits from unidentified ones. Separately, Profound found that answers with at least one tagged source link fell from about 98% to 66% between October 8 and 14. That denominator is its collected answers containing citations, not all ChatGPT links or actual clicks. Neither vendor discloses its sample size. These figures cannot be combined into a single global traffic-loss estimate.
The repair evidence is encouraging but incomplete. Adobe product director Cedric Huesler examined about 500 pages where ChatGPT supplied at least 5% of visits. Comparing October 9–13 with the previous week, he found roughly 31,000 fewer ChatGPT visits and 26,000 more Direct visits per day. That supports relabeling, within an early sampled group. He later said OpenAI privately confirmed a bug. By 9 a.m. Pacific on October 15, his tagged-click telemetry across thousands of domains was back to roughly 95–100% of the same hour a week earlier. That is a comparison with earlier volume, not the percentage of links now carrying tags. Consultant Glenn Gabe still reported missing tags on paid and free accounts. OpenAI’s inspected release notes did not announce a tracking fix.
MarketMaze Take: A channel report can fall while the website still receives customers. Merchants should resolve this measurement break before treating it as a collapse in demand. But “it was all tracking” also goes beyond the evidence: Direct includes many sources, and some real decline remains possible.
A missing label can change investment decisions. The useful lesson across markets is that a referring platform controls some information used to judge its commercial value. Compare Direct and ChatGPT visits on the affected pages alongside total visits and orders. A similar movement strengthens the reporting explanation; it does not identify every visitor or prove unchanged sales.
Recovery is a forward-looking test. Restored tags would improve new reporting, while missing historical source information would not automatically return. A rebound on one provider’s domains does not settle every merchant’s account. Keep the October 8–15 disruption visible in reports and verify local recovery before joining the old and new numbers into one trend.
