ChatGPT traffic showed up as Direct for a week

The gap began as ChatGPT moved its users to GPT-6, which we covered in ChatGPT is moving every user to GPT-6. Huesler looked at about 500 pages where ChatGPT normally brought at least 5% of visits. Comparing 9 to 13 October with the week before, he found ChatGPT visits down by about 31,000 a day and Direct visits on the same pages up by about 26,000 a day, while search traffic fell by the same 7% it fell elsewhere. He cautioned that the figures were sampled, covered four days and only show on pages where ChatGPT is a large share of traffic. His explanation was that the new links opened in a way that passed neither the tag nor a referrer.

In a follow-up post, Huesler wrote that attributable ChatGPT visits had dropped to about a third of normal from 8 October, that OpenAI told him on 14 October it was a known bug, and that by the morning of 15 October tagged ChatGPT clicks were back to roughly 95 to 100% of the previous week. The confirmation is Huesler's account of what OpenAI told him.

ChatGPT's citation links have carried a tracking tag for some time. In June 2025 Search Engine Roundtable noted that OpenAI had added one to the links in ChatGPT's "More" section of sources too, which until then had carried none.

Before the bug, the tag was on every link we saw. Our UK AI Shortlist Study kept the full ChatGPT answers for its three worked examples, three runs of one buying question each, collected on 6 and 7 September 2026. Eight of the nine answers contained links, and all 64 citation links in those eight answers, and all 22 entries in their source lists, ended in utm_source=chatgpt.com. That is a small sample from before GPT-6. It shows what normal looked like, and says nothing about how many links the bug reached.

The tag matters more than the referrer now, because Huesler wrote that the referrer was still missing on most ChatGPT clicks after the fix. That has a knock-on effect in Google Analytics. Google's channel definitions put a visit in the AI Assistant channel when the referrer matches a list of AI assistants or the medium is exactly "ai-assistant". A ChatGPT visit that arrives with the tag but no referrer may therefore not appear in that channel, so look for chatgpt.com as a session source as well as at the channel total.

Some visits never carry either signal: a link someone copies out of ChatGPT into a browser, or one opened inside another app, arrives as Direct whatever OpenAI does. Analytics undercounts AI referrals for that reason, and last week the gap grew far larger for a few days.

Mark 8 to 15 October 2026 in your analytics as one event, which is Huesler's advice too, so nobody reads the dip as lost demand or rebuilds a report around it. If you need an estimate for those days, compare Direct visits to the pages ChatGPT usually sends people to with the same pages the week before; a rise there that mirrors the ChatGPT fall is the misfiled traffic.

Referral traffic only counts the people who clicked. Whether ChatGPT names you at all is decided before any click, and a tracking bug cannot touch it. To see that directly, check if your business is on ChatGPT: our free checker puts 3 buyer questions about your category to ChatGPT in a logged-out session and shows whether each answer names you and which pages it cites. Our ChatGPT visibility tracker page explains how we measure it over time.

We counted the links in the nine ChatGPT answers saved in the UK AI Shortlist Study evidence bundle, collected from ChatGPT on 6 and 7 September 2026 for three UK buying questions, each asked three times. One answer had no links. In the other eight we checked every citation link and every source-list entry for utm_source=chatgpt.com. The bundle is free to download from the study page, so anyone can repeat the count.