Chatgpt Is Still Sending You Traffic. It Just Stopped Saying So.

Key takeaways

• When GPT-6 rolled out on October 7–8, 2026, links inside ChatGPT answers lost their ?utm_source=chatgpt.com tag, and ChatGPT started withholding the referrer. Those clicks looked exactly like someone typing your URL.
• ChatGPT referrals fell by almost half within days, but total traffic barely moved. The visitors are still coming. They’re being counted as Direct.
• That breaks more than one chart. It undercounts AI search, inflates Direct, makes AI visibility work look like it stopped paying off, and feeds a fake collapse into forecasts and models.
• Gemini started tagging some of its links a week earlier, while Claude doesn’t tag its links at all.
• Update, October 15: OpenAI reportedly told Adobe’s Cedric Huesler the missing tag was a bug, and tagged clicks are coming back. The referrer isn’t, and October 8–15 stays mislabelled in your data for good.
• Annotate the date, read ChatGPT and Direct together, and fix your model inputs before October’s data reaches them.

As an AI engineer at ELIYA, most of my work is building the pipelines that carry marketing data from ad platforms and analytics tools into the models our clients use to decide where their budget goes. A traffic source that halves overnight for a technical reason is exactly the kind of thing those models need protecting from, so this one landed on my desk fast.

What changed in ChatGPT?

On October 14, Barry Schwartz at Search Engine Roundtable reported that OpenAI had removed the UTM parameters from links in ChatGPT answers. The timing matches GPT-6, which reached paid plans on October 7 and Free and Go users a day later, together with a new “Intelligent UI” that puts links and interactive elements directly inside answers.

Cedric Huesler from Adobe found the cause. GPT-6 now opens links like this:

And the ?utm_source=chatgpt.com tag disappeared from links inside the answer text. Profound found that the share of ChatGPT answers with at least one tagged source link fell from about 98% on October 8 to 66% by October 14. Links in the sources panel kept their tag. Each change alone would be manageable. Together they remove every trace of the source:

• noreferrer tells the browser not to say which site the visitor came from.
• With the tag gone too, nothing else in the visit points back to ChatGPT.

Google Analytics has nothing to go on, so it reports the visit as (direct) / (none), as if the person had typed your address from memory.

Figure 1. Same visitor, same page. The only thing that changed is what arrives with the click. Since October 15, the tag has started returning; the referrer is still withheld.

Alt text: Diagram comparing a ChatGPT click before and after GPT-6. Before: the link carries utm_source=chatgpt.com and a chatgpt.com referrer, so analytics reports ChatGPT. After the October 7–8, 2026 rollout: the link opens with noreferrer and no tag, so analytics reports Direct.

OpenAI hasn’t said anything publicly. But on October 15, Huesler reported that OpenAI had told him the missing tag was a known bug, and that tagged ChatGPT clicks were already back to roughly 95–100% of the previous week’s level on the sites he tracks (Search Engine Journal). The fix may not have reached every account yet, and it brings back the tag, not the referrer. To be fair, noopener is standard security practice for links that open in a new tab, and noreferrer is a common privacy setting. Dropping the UTM tag at the same time is what turned a technical detail into a reporting problem. With the tag back, a ChatGPT click is traceable again. Without the referrer, any click where the tag goes missing still lands in Direct.

How do we know the visitors are still coming?

Two independent datasets point the same way.

Huesler looked at individual pages. He took about 500 pages where ChatGPT usually brings in at least 5% of visits and compared October 9–13 with the week before:

• ChatGPT visits fell by 31,000 a day.
• Direct visits on the same pages rose by 26,000 a day. By our math, that’s about 84% of what ChatGPT lost.
• Direct went up on two out of every three pages that lost ChatGPT traffic. The week before, it was a coin flip.
• Search traffic on those pages fell 7%, about the same as everywhere else, so this wasn’t a general slump.

Even the behaviour matched. On mobile, the share of Direct visitors who saw the cookie consent banner, meaning they were new to the site, rose from 36% to 44%. For ChatGPT visitors it’s 55%. The new Direct visitors behave like ChatGPT visitors because that’s very likely what they are.

Figure 2. The visitors didn’t vanish. They moved to Direct.

Alt text: Chart of about 500 pages, October 9–13 vs. the week before: ChatGPT visits fell by 31,000 a day while Direct rose by 26,000 a day. Direct rose on 2 in 3 pages that lost ChatGPT traffic, and the share of new mobile visitors in Direct rose from 36% to 44%, close to ChatGPT’s 55%.

LLM Pulse looked at whole sites. Across the sites it monitors, ChatGPT referral sessions fell 46% between October 9 and 11 compared with the previous two weeks, and more than eight in ten sites lost over 30% of their ChatGPT traffic. Total traffic over the same days moved by only 2%.

LLM Pulse didn’t see a clear jump in Direct at site level, and that’s expected. On most sites Direct is many times bigger than ChatGPT, so a few thousand extra visits disappear into the noise. You see the shift when you look at the pages ChatGPT used to send people to, which is what Huesler did.

What does this break?

Nothing about your audience changed on October 8. The labels on the clicks did, and a lot of marketing reporting quietly depends on those labels.

Figure 3. What your reports show, and what’s actually happening.

Alt text: Table of six reporting effects and their real cause: ChatGPT referrals down almost half, Direct up on long URLs, GA4’s AI Assistant channel shrinking, Direct converting better, AI search work looking like it stopped paying off, and forecasts showing a step change on October 8. In each case the audience is the same and only the click labels changed.

Your AI channel undercounts. On May 13, 2026, Google Analytics added a built-in AI Assistant channel that groups traffic from ChatGPT, Gemini, Claude and others automatically. It recognises that traffic by its referrer. A ChatGPT visit with no referrer and no tag never reaches the channel, so the channel shrinks just as AI search keeps growing.

Direct gets credit it didn’t earn. Similarweb found ChatGPT referrals converting at 7.1%, second only to paid search. Those sales still happen, but they’re now attributed to Direct. Expect Direct to look healthier, and your brand to get credit for demand that AI answers created.

AI visibility work looks like it stopped working. Many teams spent this year getting their brand cited in AI answers. The citations can still be there. The clicks just lost the label that proved it. A dashboard that shows ChatGPT traffic halving the week after a content push is an easy way to get a budget cut that nobody deserved.

Models and forecasts see a collapse that never happened. This is the one I worry about most. Any forecast, attribution model or AI agent that reads AI referral traffic as its own input now sees a step change on October 8. The model can’t tell a real drop in demand from a tracking change unless someone tells it.

Why is this a step backwards?

Because the last sixteen months were going the other way. OpenAI had been making ChatGPT easier to measure, not harder:

• Before June 2025: links in ChatGPT’s citations already carried utm_source=chatgpt.com.
• June 2025: the extra “More” sources got the tag too.
• May 2026: ChatGPT started putting clickable brand links directly inside its answers, every one of them tagged. According to Qwairy, the share of answers with a brand link jumped from 0.4% to 6.2% in a single day.
• May 13, 2026: GA4 gave AI assistants their own default channel.

Figure 4. Sixteen months of easier AI tracking, then one step back, partly undone when the tag started returning on October 15.

Alt text: Timeline from June 2025 to October 2026: ChatGPT tags its extra sources, adds tagged brand links in May 2026, GA4 adds a built-in AI Assistant channel on May 13, 2026, Gemini starts adding UTM tags around September 30, and GPT-6 drops both the tag and the referrer on October 7–8, reported on October 14.

That’s traffic marketing teams have been working hard to grow. For a week, it looked as if it had vanished, and for anyone who doesn’t annotate that week, it still will.

What are Gemini and Claude doing?

Gemini went the other way. Around September 30, 2026, Gemini started adding UTM parameters to some outbound links. A Reddit user spotted it within a day, and Google’s John Mueller responded in the thread, asking for reproducible examples of what happens to the referrer.

The referrer question matters. On desktop, Gemini already passes a referrer. But mobile apps and in-app browsers often strip it, and that’s exactly where a tag helps. Google hasn’t documented the change, so we don’t know the exact parameter values or which answers carry them. Some Gemini visits will still arrive untagged.

Claude never tagged its links. Anthropic’s assistant doesn’t add UTM parameters by default, so its visits can only be recognised by the claude.ai referrer. There was no tag for Anthropic to take away, but there’s no backup either. When the referrer doesn’t come through, which some tracking vendors say happens with links inside answers, a Claude visit lands in Direct, just as ChatGPT visits do now.

Google’s search AI is a separate case. Clicks from AI Overviews and AI Mode count as normal Google organic search, not as AI assistant traffic. Google had its own noreferrer slip in 2025, when AI Mode clicks showed up as Direct for a short time, and fixed it within days.

Figure 5. How each AI assistant labels the clicks it sends, as of mid-October 2026. ChatGPT’s tag started returning on October 15; its referrer is still withheld.

Alt text: Table comparing how AI assistants label outbound clicks. ChatGPT: UTM tag removed with GPT-6 and referrer blocked, so visits land in Direct. Gemini: tags some links since about September 30, 2026 and usually passes a referrer on desktop. Claude: no tag by default, recognised by the claude.ai referrer when it’s passed. AI Overviews and AI Mode: count as Google organic search.

So within two weeks, Google started labelling its AI clicks, OpenAI stopped (by accident, it now seems), and Claude still doesn’t label them at all. A month ago, most people would have bet on the opposite.

What should you do this week?

1. Mark the dates. Add an annotation for October 8–15 in GA4 and in every report you share. That’s the window between the GPT-6 rollout and the reported fix. Explain the drop before someone asks you about it. The fix only changes new clicks, so visits from that week stay in Direct.

2. Read ChatGPT and Direct together. For the pages ChatGPT used to send people to, add the two numbers and watch the total. Deep pages give it away: nobody types a long blog URL from memory, so a sudden Direct spike on one almost certainly came from an AI answer.

3. Fix the data before it reaches a model. If you feed AI referral traffic into a forecast, an attribution model or an AI agent as its own input, merge it with Direct from October 8 until tagged ChatGPT traffic is back to normal in your own data, or flag the break explicitly. Otherwise the model learns from a collapse that never happened.

4. Keep visibility and clicks separate. Being cited in AI answers and getting attributed clicks from them are two different things. Track how often you’re mentioned and cited separately from the traffic.

5. Ask people. Add a “How did you hear about us?” question to forms and checkout, with ChatGPT, Gemini and Claude as options. Self-reported answers are messy, but they don’t depend on anyone’s link code.

6. Check the fix in your own data. Tagged ChatGPT visits should climb back from October 15. If they don’t, keep the merged view running. Also check which channel the returning visits land in: GA4’s AI Assistant channel was built around the referrer, which ChatGPT still withholds.

How does ELIYA measure traffic you can’t tag?

UTM tags and referrers are a courtesy from the platform. One product update switched them off for the biggest AI assistant in the world, and another switched them on for Gemini. Neither was announced, and the fix for one was confirmed only through a third party. If your reporting depends entirely on click-level tracking, your view of what works is only as stable as someone else’s release notes.

This is the problem ELIYA works on every day. We measure the real, incremental impact of every marketing channel and campaign, and turn it into budget recommendations:

• Marketing mix modelling (MMM) looks at how spend, activity and revenue move together over time. It doesn’t need a tag to survive the trip from one website to another, and because it works on aggregate, non-personal data, it fits GDPR and Switzerland’s FADP without depending on cookies, consent rates or referrers.
• Incrementality testing checks what actually changes when you turn a channel up or down.
• Agentic AI does the heavy lifting, and ELIYA’s data scientists review every output, so it’s fast and never a black box.

If you’re still weighing click-based attribution against modelling, see our guide to MMM vs multi-touch attribution.

From the engineering side, I see this approach pay off in weeks like this one. When a channel’s tracking breaks, click data goes haywire. A model built on overall patterns, checked by people who know where the data came from, keeps telling you what’s really driving revenue.

Try ELIYA AI on your own marketing data. ELIYA Studio is ELIYA’s AI analyst for your marketing mix model. Ask it in plain language which channels really drove last quarter’s revenue, where your next euro of budget should go, or what happens if you move 10% from Meta to Google, and get an answer in minutes instead of weeks. Every answer comes from your model’s channel returns and saturation curves, not from click labels that can disappear overnight.

Try ELIYA AI

The bottom line

ChatGPT didn’t stop sending you visitors. For a week, it stopped telling you about them. The tag is coming back, the referrer isn’t, and Gemini’s new tags could change without notice. Build your measurement so the next disappearance is a footnote, not a crisis.

If you’re an e-commerce or retail brand spending across several channels and markets, and you need to know what’s really driving revenue when click labels come and go, ELIYA AI is built for this. Book a meeting with ELIYA and see what your own marketing mix model says about last quarter.

Frequently Asked Questions

Why did ChatGPT traffic drop in October 2026?

Mostly because of labelling, not demand. From October 8, links inside ChatGPT answers arrived without the utm_source=chatgpt.com tag and without a referrer, so analytics counted them as Direct. ChatGPT referral sessions fell about 46%, while total traffic on the same sites barely moved.

Is ChatGPT adding the utm_source tag to its links again?

It appears so. On October 15, 2026, Adobe’s Cedric Huesler reported that OpenAI called the missing tag a known bug and that tagged clicks were recovering. OpenAI hasn’t confirmed this publicly, and ChatGPT still withholds the referrer.

Why does ChatGPT traffic show up as Direct in GA4?

GA4 needs a UTM tag or a referrer to identify where a visit came from. When a link opens with noreferrer and carries no tag, GA4 has nothing to go on and reports the visit as (direct) / (none).

How can I track ChatGPT traffic without a referrer?

Combine signals. Read ChatGPT and Direct together on deep pages, ask “How did you hear about us?” with AI assistants as options, track AI citations separately from clicks, and use aggregate methods like marketing mix modelling that don’t depend on click labels.

Will the October 8–15 data be corrected?

No. A fix only changes new clicks. Visits that already landed in Direct stay there, so annotate the period and adjust any forecast or model input that covers it.