# ChatGPT traffic suddenly collapsed? Lost visits or just lost labels

> From October 8 to 15, 2026, ChatGPT links arrived without utm_source and GA4 filed them as Direct. How to check your own data and what to annotate.

URL: https://geosnap.ai/en/blog/traffico-chatgpt-crollato-utm-visite-perse-o-etichette-perse
Language: English
Version IT: https://geosnap.ai/blog/traffico-chatgpt-crollato-utm-visite-perse-o-etichette-perse
Author: Rinald Sefa, CMO Geosnap · Category: Guides
Publisher: Geosnap (Maind Group S.r.l.), https://geosnap.ai

## In short

In most cases the visits did not disappear: from October 8 to 15, 2026, many ChatGPT links reached websites without the utm\_source=chatgpt.com parameter and without a referrer, and GA4 classified them as direct traffic. OpenAI informally confirmed it was a bug, fixed on October 15. To tell whether you lost visits or just labels, compare ChatGPT and Direct on the same landing pages.

- Between October 9 and 11, 2026, ChatGPT-attributed sessions fell by about 46% on the sites monitored by LLM Pulse, while total traffic stayed flat.
- On 500 pages analysed by Adobe, ChatGPT lost 31,000 visits a day and direct traffic gained 26,000: the visits had only moved channel.
- Links in the text of GPT-6 answers had lost the utm\_source=chatgpt.com parameter; links in the Sources panel and shopping links had not.
- The fix arrived on October 15, but the October 8-15 data is not being recomputed and the referrer is still missing on most clicks.
- Annotate the window in GA4, count the UTM parameter rather than the referrer, and keep citations in answers separate from clicks.

Written by the Geosnap AI agent, reviewed and approved by Rinald Sefa.

Translated from the Italian original. [Read the original](https://geosnap.ai/blog/traffico-chatgpt-crollato-utm-visite-perse-o-etichette-perse)

If your ChatGPT traffic in Google Analytics collapsed overnight in the first half of October 2026, the visits almost certainly did not disappear: they lost their label. From October 8 to 15, many links clicked inside ChatGPT reached websites without the `utm_source=chatgpt.com` parameter and without a referrer, and GA4 filed them as direct traffic. The episode is over, but the lesson stays valid: traffic from AI assistants rests on a fragile tag, and before you draw conclusions from a sudden drop you need to check the tracking. Here is what happened, how to tell lost visits from lost labels, and what to check in your own data.

## What happened to ChatGPT traffic on October 8, 2026?

On October 7, OpenAI started rolling out [GPT-6 with Intelligent UI](https://openai.com/index/gpt-6-for-everyone/) to paid plans, and on October 8 to Free and Go plans. Intelligent UI composes answers with charts, buttons, side-by-side comparisons and small interactive tools instead of plain text. The announcement says nothing about how links and sources change, but within hours many sites saw their ChatGPT-attributed sessions halve.

The numbers collected by people who track this traffic tell the same story. [LLM Pulse](https://llmpulse.ai/blog/chatgpt-referral-traffic-fell/) measured, across the sites it monitors, a drop of around 46% in ChatGPT sessions between Friday October 9 and Sunday October 11 compared with the same days in the two previous weeks: the typical site lost roughly half of its ChatGPT traffic, and more than eight in ten sites lost over 30%. Over the same days, total traffic on those sites stayed flat (around -2%), and referrals from Perplexity and Claude showed no break. According to [Search Engine Journal](https://www.searchenginejournal.com/chatgpt-traffic-drop-appears-to-be-a-tracking-problem/593157/), Graphite saw drops between 22% and 57% on sites with meaningful traffic in GA4's AI Assistant channel, with an estimated 45% fall overall.

The explanation came from looking at where the visits went. Cedric Huesler, senior director of product management at Adobe, compared October 9-13 with the previous week on about 500 pages where ChatGPT drove at least 5% of visits: ChatGPT lost about 31,000 visits a day, direct traffic gained about 26,000 a day on the same pages, and two out of three pages losing ChatGPT referrals saw direct rise. Search traffic, meanwhile, fell 7%, in line with the broader trend. [Search Engine Land](https://searchengineland.com/chatgpt-appears-to-drop-utm-tracking-making-ai-referral-traffic-harder-to-measure-496552) reports the details.

## Why did GA4 count those visits as direct?

Because GA4 assigns a session's source first from UTM parameters, then from the referrer: when both are missing, the visit is "Direct". Since ChatGPT search launched, links in answers carried the `utm_source=chatgpt.com` parameter, and that parameter was what made the visits recognisable, above all from the mobile app, which often passes no referrer. According to Graphite's Ethan Smith, quoted by Search Engine Journal, 95% of mobile ChatGPT visits to the sites his agency tracks relied on that tag alone.

With GPT-6, two things changed at once. Links placed in the answer text stopped including the parameter, while links in the Sources panel and shopping links kept it. On top of that, as Huesler observed, ChatGPT started opening links in a new tab with a browser setting that withholds the referrer. No tag and no referrer: as far as GA4 is concerned, that visit came from nowhere.

A third data point shows the scale of the change. Profound, again according to Search Engine Journal, measured that the share of ChatGPT answers with at least one tagged source link fell from about 98% on October 8 to 66% on October 14, while answers in the Intelligent UI format went from 2% to about 35%. Peec AI re-ran the same one million prompts before and after the release: ChatGPT put a URL directly into the answer in 42% of responses, against 8-10% before, and 54% of those URLs pointed to home pages. More links in the text, fewer tagged links: the traffic did not shrink, it became invisible.

## Was it a bug? What OpenAI said

Yes, according to the only confirmation available, which is secondhand. Huesler wrote that on October 14 OpenAI confirmed to him that it was a known bug with a fix due that week; on October 15 the fix started rolling out and by 9 a.m. Pacific time, tagged ChatGPT clicks were back to roughly 95% to 100% of the same hour a week earlier across the thousands of domains he tracks. [Relevant Audience](https://www.relevantaudience.com/analytics/chatgpt-utm-tag-bug-fixed-openai-confirmed-adobe/) reconstructed the full timeline and makes an important point: OpenAI has published no statement, and its release notes for October 1 to 15 do not mention tags or link tracking.

Three things remain open even after the fix. The referrer is still missing on most clicks, so attribution keeps depending on the UTM parameter. Sessions from October 8 to 15 are not being recomputed: nobody has mentioned a backfill, and those visits stay classified as direct. Finally, other assistants change their labels without notice: LLM Pulse reports that between September 28 and October 2 visits with source "gemini" nearly disappeared while visits from gemini.google.com held steady, a sign of a change in how Gemini tags its links.

## How to tell whether you lost visits or just labels

Compare ChatGPT and Direct on the same pages over the same period. The method used by the analysts above works for any site and takes ten minutes in GA4.

1. **Take the landing pages that received the most ChatGPT visits** in the two weeks before October 8. Add the home page: according to Peec AI, more than half of the direct links pointed there.
2. **For each one, line up ChatGPT sessions and Direct sessions** day by day from October 1 to 20. If Direct rises as ChatGPT falls and the sum stays stable, those are lost labels.
3. **Look at behaviour, not just volumes.** Huesler used the share of visitors who see the cookie banner as a first-visit proxy: among new mobile "direct" visitors it went from 36% to 44%, moving toward the 55% typical of ChatGPT visitors. You can do the same with the share of new users or average session duration.
4. **Check your property's time zone.** The drop began around 2 p.m. Pacific on October 8 and the fix around 2 a.m. Pacific on October 15: in Central European time your reports break overnight between the 8th and the 9th and return to normal on the afternoon of the 15th.
5. **Confirm the recovery** by comparing sessions with source chatgpt.com from October 16 onward with the same weekdays in late September.

If instead Direct did not rise and the sum of the two rows really fell, then something changed in ChatGPT's answers for your brand. Even then, the traffic figure alone is not enough: the first thing to check is whether the answers still cite you. Geosnap measures citations in ChatGPT, Gemini and Google AI Mode answers independently of UTMs and referrers, and with the [Google Analytics integration](https://geosnap.ai/en/features) you can put visibility and visits side by side for the same period.

## What to do in your reports, and what to avoid

Annotate the window and do not touch your strategy because of that number. Concretely:

- **Add an annotation from October 8 to 15, 2026** in GA4 and in your reporting tools: six months from now, whoever looks at the chart will not know what happened.
- **Count the UTM parameter, not the referrer.** If you have a custom channel group based on source, like the one described in our [guide to ChatGPT traffic in GA4](https://geosnap.ai/en/blog/traffico-da-chatgpt-come-misurarlo-google-analytics-4), leave it as it is: it remains the most reliable approach.
- **Do not rebuild attribution models around the gap.** The visits never went away; a model calibrated on that week is wrong for every other week.
- **Keep visibility and clicks separate.** Being cited in an answer and receiving a click are two different measures: the first can grow while the second falls, and the other way round.
- **Treat ChatGPT Ads separately.** Ad clicks carrying your own custom UTM parameters were not affected, because you set the label yourself.
- **Do not mistake bots for people.** One site recorded, over the same days, a drop in the ChatGPT-User bot and a rise in OAI-SearchBot: those are crawlers, not visitors. To understand who they are, read our [guide to AI crawlers](https://geosnap.ai/en/blog/il-tuo-sito-blocca-chatgpt-guida-crawler-ai).

The most useful lesson is the last one: AI traffic attribution depends on choices made by the assistants, and those can change overnight, through a bug or a decision. Whoever measures both citations and visits notices immediately; whoever looks only at Google Analytics risks making decisions on a number that does not exist.

## Sources

- [GPT-6 and Intelligent UI for everyone](https://openai.com/index/gpt-6-for-everyone/) (OpenAI, 2026-10-07)
- [ChatGPT Referral Traffic Fell by Almost Half on October 9, 2026. Here's What We Know](https://llmpulse.ai/blog/chatgpt-referral-traffic-fell/) (LLM Pulse, 2026-10-13)
- [ChatGPT appears to drop UTM tracking, making AI referral traffic harder to measure](https://searchengineland.com/chatgpt-appears-to-drop-utm-tracking-making-ai-referral-traffic-harder-to-measure-496552) (Search Engine Land, 2026-10-15)
- [ChatGPT Traffic Drop Appears To Be A Tracking Problem](https://www.searchenginejournal.com/chatgpt-traffic-drop-appears-to-be-a-tracking-problem/593157/) (Search Engine Journal, 2026-10-15)
- [ChatGPT UTM tag loss was a bug, Adobe's Huesler says; fix rolling out](https://www.relevantaudience.com/analytics/chatgpt-utm-tag-bug-fixed-openai-confirmed-adobe/) (Relevant Audience, 2026-10-16)
- [The Day ChatGPT Traffic "Disappeared" From Analytics, and Where It Went](https://thepromptinsider.com/news/chatgpt-missing-utm-tags-referral-traffic/) (The Prompt Insider, 2026-10-14)

## Frequently asked questions

### Did ChatGPT traffic really fall in October 2026?

For most sites, no. Analyses by LLM Pulse, Adobe and Graphite show ChatGPT-attributed sessions down 45-46% on average, but on the same pages direct traffic rose by almost as much. The visits were there; what was missing was the label that made them recognisable.

### Why does GA4 classify ChatGPT visits as Direct?

GA4 derives the source first from UTM parameters and then from the referrer. ChatGPT, especially from the mobile app, often passes no referrer, so recognition depended on the utm\_source=chatgpt.com parameter. When GPT-6 stopped adding it to links in the answer text, those visits ended up in Direct.

### Has OpenAI confirmed the bug?

Not publicly. Adobe's Cedric Huesler wrote that OpenAI confirmed to him on October 14 that it was a known bug and that the fix started rolling out on October 15. OpenAI's release notes for October 1 to 15 mention neither the problem nor the fix.

### Can I recover the October 8-15 data?

No. Information that never reached your site cannot be rebuilt by any analytics tool, and nobody has announced a backfill. The most useful thing is to annotate the window and read ChatGPT and Direct together for that week.

### Were ChatGPT Ads affected by the problem?

According to Graphite's Ethan Smith, no, as long as you use custom UTM parameters on your ads: in that case you set the label yourself and it does not depend on the links ChatGPT generates. It is still worth checking the clicks for that period in your own data.