Four Vendors Now Tie ChatGPT's Referral Drop to Missing Link Tags

Four analytics and AI-visibility vendors, Graphite, Peec AI, Profound, and LLM Pulse, have now published data pointing at the same culprit for ChatGPT’s sagging referral numbers: links that arrive without the tags analytics tools rely on. Search Engine Journal’s Matt G. Southern pulled the new measurements together on October 15, 2026. Our earlier report on Cedric Huesler of Adobe covered the Direct-traffic angle; this update covers what the other datasets add.

Graphite, an SEO firm, supplied the sharpest drop. CEO Ethan Smith posted on LinkedIn that sessions that GA4 credits to ChatGPT declined between 22 percent and 57 percent from October 9 across Graphite-tracked sites with meaningful AI Assistant channel traffic. Smith puts the overall decline near 45 percent against the earlier baseline. LLM Pulse measured a similar shape: ChatGPT sessions on its monitored sites ran about 46 percent below the comparable days of the prior two weeks for October 9 to 11, while total traffic on those sites held roughly steady.

Which links lost their tags matters for diagnosis. Smith reported that inline links inside answer text dropped the utm_source tag. Shopping links and anything in the Sources panel kept it. Profound, a vendor that tracks AI answers, ties this to a new answer format. In its sample of cited ChatGPT answers, the portion with a tagged citation link slid from about 98 percent on October 8 to 66 percent by October 14. Intelligent UI, ChatGPT’s layout for charts, side-by-side comparisons, and mini tools, rose from 2 percent of Profound’s collected answers to about 35 percent over the same stretch. Links opened from product or information panels still carried tags.

Peec AI’s Tomek Rudzki tested the link side directly. He sent an identical set of 1 million prompts twice, once from October 2 to 7 and again from October 9 to 13. A direct webpage URL appeared in 42 percent of responses in the second run, against roughly 8 to 10 percent before. Of those URLs, 54 percent pointed to home pages, and none carried tags or referrers. Those figures describe links displayed, not clicks made.

Mobile looks like the weak spot. Smith said on Graphite’s sites, 95 percent of ChatGPT visits from mobile depended on the tag alone, so the app is the most exposed surface. Without a referrer to fall back on, a stripped tag leaves the session with no source.

Whether any real visits vanished is still unsettled. Huesler’s analysis showed Direct gains offsetting ChatGPT losses on his page sample, and our earlier piece covered those numbers. LLM Pulse found no visible rise in Direct on its sites. Its authors explained that Direct is far larger than ChatGPT, so a modest influx would be hard to spot, and that aggregated data cannot separate lost visits from relabeled ones. Smith, for his part, said there is not yet enough data to rule out some genuine loss.

The status of a fix is contested. Smith called the change an OpenAI bug and expected a quick repair. Huesler says OpenAI confirmed the bug to him on October 14, and that clicks carrying the ChatGPT tag had returned to roughly 95 to 100 percent of their level one week prior by 9 a.m. PT on October 15, based on thousands of domains in his monitoring. Consultant Glenn Gabe answered that the fix had not yet shown up for him on either account tier. Profound has seen no official word, while the release notes OpenAI publishes for ChatGPT, covering October 1 through 15, do not mention link tracking. The report gives no OpenAI statement of its own.

Treat the vendor figures with the disclosure Search Engine Journal attached: all four companies sell SEO or AI search products, Huesler works at Adobe, and neither the LLM Pulse nor the Profound post states a sample size. Profound adds that restored tags would help future reporting but would not recover attribution already lost.

For reporting, annotate the October 8 start date, which Huesler now extends to a window running through October 15, before anyone reads the dip as lost demand. Compare ChatGPT and Direct sessions page by page, and check the homepage first, since that is where Rudzki found over half of the direct links pointing. Huesler recommends counting the UTM tag rather than the referrer, because the bulk of ChatGPT clicks continue to lack a referrer. Advertisers get one reprieve: Smith says UTM parameters you set yourself on ChatGPT Ads clicks should be unaffected.

Search Engine Journal, Matt G. Southern, October 15, 2026, drawing on posts from Graphite’s Ethan Smith, Adobe’s Cedric Huesler, Peec AI’s Tomek Rudzki, LLM Pulse, and Profound.