ChatGPT referral traffic now shows up as Direct in most analytics tools because, since the GPT-6 rollout in October 2026, links clicked in ChatGPT reportedly arrive without the utm_source=chatgpt.com tag and without a referrer. The visitors still come, but your reports can no longer tell where they came from.
OpenAI has not commented as of October 15, 2026, so treat the cause as reported, not confirmed. This guide, part of the analytics, tracking and privacy hub, separates the evidence, explains the browser mechanism, and gives a method to estimate how many Direct visits are really ChatGPT clicks. ChatGPT referral traffic is the set of visits that analytics attributes to chatgpt.com, through the referrer or a utm_source tag. Direct is the GA4 channel for sessions with source (direct) and medium (none) or (not set), which means no referrer and no campaign tag arrived.
What happened to ChatGPT referral traffic after GPT-6?
ChatGPT referral traffic dropped sharply in analytics from about October 9, 2026, while Direct traffic rose on the same pages. Practitioners trace the shift to GPT-6, which OpenAI rolled out on October 7, and report that ChatGPT links lost both their UTM tag and their referrer, so the clicks were relabeled rather than lost.
According to OpenAI's GPT-6 announcement, GPT-6 started rolling out to Plus, Pro, Business and Enterprise users on October 7, 2026 and to Free and Go users the next day. The announcement does not mention links or referral traffic.
On October 14, 2026, Search Engine Roundtable's Barry Schwartz reported that OpenAI appears to have removed the UTM parameters from links in ChatGPT results, and that it is unclear whether the change was intended. The report rests on an analysis by Cedric Huesler of Adobe. On about 500 pages where ChatGPT had driven 5% or more of visits, he compared October 9 to 13, 2026 with the week before:
- ChatGPT visits fell by about 31,000 a day.
- Direct visits rose by about 26,000 a day on the same pages.
- Search visits fell 7%, in line with other sources.
- Direct rose on two in three pages that lost ChatGPT traffic, against roughly a coin flip the week before.
What is confirmed, what is reported and what is speculation?
Only two facts are officially documented: GPT-6 rolled out on October 7 and 8, 2026, and OpenAI's own help center says ChatGPT tags referral URLs with utm_source=chatgpt.com. The missing tag, the missing referrer and the cause are reported by practitioners, and whether the change is permanent is speculation.
OpenAI's publishers and developers FAQ still says, as of October 15, 2026, that "ChatGPT automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs." The tagging goes back further: in June 2025, Search Engine Roundtable reported that links in the "more" section of ChatGPT answers gained the UTM tags that citation links already carried.
- Officially confirmed: the GPT-6 rollout dates, and the documented
utm_source=chatgpt.combehavior that observed traffic now contradicts. - Reported: links without the tag and without a referrer, the
noopener,noreferrermechanism, and the Adobe traffic figures (Huesler via Search Engine Roundtable). Sébastien Billard reported on October 14, 2026 a drop in human ChatGPT visits since October 9 and more OAI-SearchBot crawling from about October 7. - Not published: which ChatGPT surfaces are affected (web, iOS app, Android app, desktop app) and whether all link types lost the tag.
- Speculation: whether OpenAI meant to drop attribution or will restore it. Dalin de Graff notes that OpenAI has not confirmed whether the change is deliberate or a side effect.
Why do noopener and noreferrer turn ChatGPT clicks into Direct?
The noreferrer window feature tells the browser to open a link without sending the Referer header, and it also switches on noopener. Without a Referer header and without a UTM tag in the URL, the landing page receives no information about the previous site, so analytics files the visit as Direct.
The Referer header is the HTTP header in which a browser tells the destination which page the visitor came from. Huesler, quoted by Search Engine Roundtable, wrote that GPT-6 opens links with window.open(url, "_blank", "noopener,noreferrer") and drops the utm_source=chatgpt.com tag. MDN's documentation of window.open() describes the effect: with noreferrer set, "the browser will omit the Referer header, as well as set noopener to true." With noopener, the new tab also cannot access the ChatGPT window through window.opener.
Either signal alone would have been enough for attribution. Before the change, a missing referrer (common in apps) was rescued by the UTM tag, and a missing tag was rescued by the referrer. Losing both at once removes the last trace.
The new visitors also look like ChatGPT visitors. In the same October 2026 comparison, Huesler reported via Search Engine Roundtable that on mobile the share of Direct visitors who saw a consent banner, a proxy for first-time visitors, rose from 36% to 44%, against 55% for known ChatGPT visitors.
How does GA4 classify a ChatGPT visit with and without the tag?
GA4 classifies a ChatGPT visit by the signal it receives: a recognized referrer puts it in the AI Assistant channel, a UTM tag sets the source from the tag, and a visit with neither becomes Direct. Since the reported GPT-6 change, most ChatGPT clicks fall into the last case.
Google's default channel group documentation sets the medium to "ai-assistant" and the campaign to "(ai-assistant)" when the referrer matches GA4's list of AI assistants, and names ChatGPT as an example. Direct is defined as source "(direct)" with medium "(not set)" or "(none)". When any UTM parameter is present, GA4 derives all traffic source values from the UTM parameters, and missing ones show as "(not set)", according to Google's traffic-source dimensions page. A tag with only utm_source=chatgpt.com therefore leaves the medium "(not set)", which matches neither the AI Assistant rule (medium "ai-assistant") nor Direct (source "(direct)"), so it falls to Unassigned, the value GA4 uses when no channel rule matches.
| Signals on arrival | Session source / medium | Default channel | When it happens |
|---|---|---|---|
| Referrer chatgpt.com, no UTM | chatgpt.com / ai-assistant | AI Assistant | Referrer kept, tag absent |
utm_source=chatgpt.com only | chatgpt.com / (not set) | Unassigned (no rule matches) | Tag kept, referrer lost |
Referrer and utm_source=chatgpt.com | chatgpt.com / (not set) | Unassigned (no rule matches) | UTM values replace the referrer |
| Neither | (direct) / (none) | Direct | Reported default since the GPT-6 rollout |
| Your own UTM in a ChatGPT Ads URL | Your tagged values | Depends on your medium | Paid clicks, tags set by the advertiser |
Before October 2026 the tag already beat the referrer, which split ChatGPT across channels. The Gemini referral traffic guide shows a custom AI channel that catches both forms. That channel cannot catch the bottom rows, because no ChatGPT signal remains.
How can you spot ChatGPT referral traffic hiding in Direct?
You can spot ChatGPT referral traffic in Direct by comparing, page by page, the change in ChatGPT sessions with the change in Direct sessions around October 7 to 9, 2026. Pages where ChatGPT fell and Direct rose by a similar amount, while other channels stayed flat, are the ones where ChatGPT clicks now hide.
- Pick two date ranges. Use a baseline of September 30 to October 6, 2026 and a test range starting October 9, 2026, both of the same length and the same weekdays.
- Build the page list. In "Explore", create a "Free form" exploration with "Landing page" as rows and "Session source / medium" as columns, with "Sessions" as the metric.
- Select ChatGPT-exposed pages. Keep landing pages where sources containing
chatgpthad a meaningful share of sessions in the baseline, for example 5% or more, the cutoff Huesler used in October 2026. - Compare the deltas. For each page, put the change in ChatGPT sessions next to the change in "(direct) / (none)" sessions.
- Check a control channel. Compare with organic search on the same pages, so a sitewide dip is not mistaken for a shift.
- Look at deep landing pages. Direct visits to long, specific URLs that nobody types or bookmarks, with no internal entry path, are a strong hint of a lost referrer.
How much ChatGPT referral traffic moved to Direct? A worked estimate
The Direct shift estimate puts a range on the misattributed visits: for ChatGPT-exposed pages, take the rise in Direct above the trend of a control channel, and cap it at the drop in ChatGPT sessions. The smaller of the two numbers is a cautious estimate of ChatGPT visits now counted as Direct.
The decision rule:
- Expected Direct is baseline Direct sessions multiplied by the change in a control channel, such as organic search on the same pages.
- Excess Direct is the actual Direct sessions minus expected Direct.
- Shifted ChatGPT visits is the lower of excess Direct and the ChatGPT drop. If excess Direct is zero or negative, report no shift.
Hypothetical example, with invented numbers for illustration only: a page set had 1,000 ChatGPT and 2,000 Direct sessions in the baseline week, then 200 ChatGPT (a drop of 800) and 2,650 Direct in the test week. Organic search fell 5%, so expected Direct is 2,000 × 0.95 = 1,900 and excess Direct is 750. The Direct shift estimate is the lower of 750 and 800: about 750 ChatGPT visits a week sit in Direct, for an estimated ChatGPT total of 950.
For conversion credit, the attribution models guide explains how GA4 credits Direct in key event reports.
What can and cannot fix ChatGPT referral tracking?
No setting on your side can restore the referrer or the UTM tag, because ChatGPT decides how links open. What you control is how you report: estimate the shifted share, watch server logs for ChatGPT activity, keep your own tags on links you place, and recheck the OpenAI documentation for a fix.
Things that do not work:
- Adding UTM parameters to your canonical URLs. ChatGPT links to the URL it found, and every other source would carry the same tag.
- A new channel group or a GTM rule. Both classify signals that never arrive.
- Asking the browser for the referrer. The
noreferrerfeature is set by the opening page, ChatGPT, not by yours.
Things that help:
- Server or CDN logs. Billard's October 2026 report of more OAI-SearchBot crawling shows that bot activity stays visible even when human referrals do not.
- Tags on links you control. Links in ChatGPT Ads, custom GPTs or apps carry the UTM values you set. The ChatGPT Ads conversion tracking guide covers the paid side.
- Visibility tracking. AI visibility tools measure how often ChatGPT cites you, which no longer depends on referrers.
Common mistakes when reading ChatGPT referral traffic after GPT-6
The most common mistakes with ChatGPT referral traffic after GPT-6 come from reading a tracking change as a demand change. Since the reported change of October 2026, a falling ChatGPT line says little on its own, so every conclusion needs the Direct trend on the same pages next to it.
- Reporting the drop as lost demand. A fall in ChatGPT sessions after October 9, 2026 with a matching rise in Direct is a labeling change. Report ChatGPT and Direct together before anyone cuts budget for AI visibility work.
- Crediting Direct growth to brand campaigns. The same shift can make a brand campaign or a PR push look stronger than it was. Check the timing against October 7 to 9, 2026 first.
- Comparing AI traffic year over year without a note. Any period that crosses the GPT-6 rollout mixes two tracking methods. Annotate the break and show the Direct shift estimate separately.
- Treating the Adobe figures as a benchmark. Huesler's numbers describe one sample of pages. Your own ratio between lost ChatGPT and gained Direct can differ, so measure it.
- Assuming the change is permanent. OpenAI has not confirmed it, and its FAQ still documents the tag. Keep the AI channel in place so it picks up tagged visits again if they return.
