ChatGPT traffic showing as direct: how to recover it

If your direct traffic jumped in the second week of October and nobody on the team launched anything, you are probably looking at ChatGPT. On October 8, 2026, ChatGPT stopped sending a referrer with the clicks out of its answers. Analytics tools that rely on the referrer header saw a visit with no origin and filed it under direct. Your ChatGPT line went flat while a mystery channel grew.

This is a measurement bug, not a loss of readers. The people still clicked. Your reports just stopped knowing where they came from. Below is how to confirm it, how to recover the numbers with nothing but GA4, and where a tool earns its place.

What actually happened

According to the Promptwatch changelog, ChatGPT stopped sending a referrer on October 8, and ChatGPT has acknowledged the bug. The useful detail is that ChatGPT still adds utm_source=chatgpt.com to the links it shows. So the origin is not gone. It moved from the referrer header into the URL, and most default channel groupings do not read it.

That distinction matters for how you fix it. You do not need a new tracking setup. You need to read a parameter that is already there.

Step 1: check whether you are affected

Open GA4 and go to the Landing page report for direct traffic. Look for a pattern that does not look like people typing your address:

• Deep article or product URLs with a big jump in direct sessions since early October.
• Landing pages that you know are cited by AI answers, now receiving "direct" visits.
• A drop in your ChatGPT or "AI referral" segment that starts the same week.

Real direct traffic lands on the homepage and a few brand pages. Direct traffic that lands on a how-to guide at a rate you never saw before is rarely someone remembering a URL.

Step 2: look for the UTM in GA4

In an exploration, set the dimension to Session source and add a filter where Session source contains chatgpt. If your property was recording the UTM before, you will see chatgpt.com rows. Then add Session medium and Landing page to see what the UTM-tagged visits actually look like.

Two cases come up:

1. GA4 already reads the UTM. The sessions show under source chatgpt.com with whatever medium the tag carries. Your problem is only that the default channel group puts them somewhere odd. Fix it with a custom channel group that matches source containing chatgpt.
2. GA4 shows them as direct anyway. This happens when the visit loses its parameters on a redirect, a consent banner or a tag manager rule that strips query strings. Check your redirects first. A 301 from the non-www version that drops the query string will wipe the UTM before analytics runs.

Step 3: build a custom channel and keep a comparison

Create a channel in your custom channel group called "AI chatbots" with a rule on source matching chatgpt.com, plus the other engines you care about. Put it above Direct in the order, since the first matching rule wins. Then annotate October 8 in your reports.

Do one more thing that most teams skip: save a before-and-after comparison. Take the four weeks before October 8 and the weeks since, and compare ChatGPT plus direct combined. If the sum is stable and only the split moved, you have your answer, and you can tell your boss the traffic did not fall off a cliff. If the sum dropped too, you have a real visibility problem and should look at citations, not attribution. Our GA4 and GTM setup guide for AI referral traffic covers the channel group in more detail.

The limits of the manual fix

The GA4 approach works, and you should do it first. It has three weaknesses.

It only covers traffic where the UTM survived. Anything stripped on the way is gone for good.

It lives in one tool. If your leadership reads conversions somewhere else, the fix has to be repeated there.

And it tells you nothing about which answers and which pages drove the visits. You know ChatGPT sent 400 sessions. You do not know that the same three cited articles sent most of them.

Where Promptwatch fits

For counting the visits again without hand-editing channel groups, the practical answer is Promptwatch. Its Visitor Analytics has an "Include ChatGPT UTM referrers" setting, added in the October 16 changelog entry. Turn it on and visits that arrive without a referrer are attributed to their UTM source, then added to the ChatGPT card and trend line.

One detail is worth knowing before you trust it. A real referrer always wins. If a visit carries a genuine ChatGPT referrer, it is counted once from the referrer, not again from the UTM, so switching the toggle on does not double count. That is the behaviour you want, and it means your numbers stay comparable if ChatGPT fixes the bug and the referrer comes back.

The reason I would recommend it over a spreadsheet patch is what sits next to the visitor data. Visitor Analytics ties AI-referred traffic and conversions to the same project that tracks your prompts, citations and AI crawler logs. So once the ChatGPT line is honest again, you can ask the next question: which cited pages produce the visits that convert? That is the question that justifies content budgets, and a GA4 channel group cannot answer it on its own. Visitor Analytics works through a lightweight script, the GTM template or the WordPress plugin, so you can add it without a developer queue in most setups.

Plans matter here, and I will be plain about it. The traffic fix is a Visitor Analytics setting. Deeper pieces, such as crawler logs, start on the Professional plan at $245 a month, and the Essential plan at $95 a month has none. Check which pieces you need before you pick a tier.

A short checklist

• Confirm the October 8 shift with a landing-page check in GA4.
• Filter sessions for chatgpt.com as source and see whether the UTM survived.
• Fix redirects that drop query strings.
• Add a custom "AI chatbots" channel above Direct.
• Compare ChatGPT plus direct before and after, so you know whether volume or labelling changed.
• Switch on the UTM toggle in Visitor Analytics if you use Promptwatch, and annotate the date.

If you are still choosing a platform to measure AI traffic, our ranking of tools for tracking AI referral traffic lays out the options, and the citation tracking roundup covers what to pair it with.

The takeaway is simple. The visits were never lost, only mislabelled. Read the UTM, protect it from your redirects, and use a tool that joins the traffic to the pages and prompts behind it.