Why ChatGPT Traffic Looks Low in Your Google Analytics

Your marketing review is on Monday. Someone pulls up Google Analytics and asks why visits from ChatGPT fell last week. Or why they barely show up at all, when your sales team keeps hearing "I asked ChatGPT and it suggested you."

Both of these are common, and neither one tells you on its own that AI assistants stopped recommending you. Visits from AI assistants are some of the hardest traffic to count correctly. This guide explains why, what happened to ChatGPT's tracking in October 2026, and how to sanity-check your own numbers in about an hour.

How analytics decides where a visitor came from

When someone lands on your site, Google Analytics 4 (GA4) looks for two clues:

A referrer. The browser often passes along the address of the page the visitor clicked from. If that page was chatgpt.com, GA4 can tell.
Tags in the link. Small labels at the end of a URL, such as utm_source=chatgpt.com, say where the click came from even when the referrer is missing.

GA4 then sorts each visit into a channel using fixed rules. Google's own documentation lists the rule for Direct: the source is "(direct)" and the medium is "(none)" or "(not set)" (Google Analytics Help). In plain words, Direct is where a visit goes when GA4 has no clue at all about where it came from. It does not mean the person typed your address.

So the question is simple. When someone clicks your link inside ChatGPT, does any clue survive the trip?

Why AI visits go missing

Several everyday things strip those clues away:

The link has no tag. ChatGPT adds utm_source=chatgpt.com to many links, but not to every link in every kind of answer. A test by the agency Seer Interactive found that cited links were the most likely to carry the tag, and that a link could still arrive without it (Seer Interactive).
The app hides the referrer. Links opened from inside a mobile app often reach your site with no referrer. In India, where most browsing happens on phones, this matters more than it does in most markets.
People copy and paste. A buyer reads an answer, copies your website name and types or pastes it into a new tab. There is no click to track.
People search you up afterwards. Very often, the buyer hears your name from ChatGPT and then searches for it on Google. That visit shows up as Organic Search or Direct, and ChatGPT gets no credit at all.

The last one is the biggest blind spot. An AI assistant can shape the shortlist without ever sending a click that analytics can see.

What happened between 8 and 16 October 2026

If your ChatGPT numbers dipped around 9 October, there is a likely explanation. One published analysis tracked a new answer format in ChatGPT, with interactive elements alongside text, and found that many links in that format were missing the utm_source=chatgpt.com tag (source).

The sequence it reported, in UTC:

8 October, evening: the share of answers with a tagged link began to fall.
9 October: about 75% of collected answers with cited links had at least one tagged link. Roughly a quarter had none.
14 October, evening: tagging fell much further.
16 October, morning: tagging was back to about 99% of answers with cited links.

Two points from that analysis are worth repeating to your team. First, a quarter of answers losing their tags does not mean a quarter of visitors stopped coming. It means those visits were harder to label. Second, the tags coming back does not fix the old data. Visits from that week that arrived without a clue stay filed as Direct or elsewhere.

The analysis also warns against treating its percentages as a correction factor for your own reports, because people click different links in different answers. That is sensible advice. Use the dates to explain a dip, not to rewrite your numbers.

What to do: add an annotation in GA4 for 8 to 16 October 2026 (the dates above are UTC, so in IST the window starts around midnight on 9 October). Note it in any monthly report that compares October with September. And be just as careful with a sudden jump in the week after 16 October, since part of it may be tags returning rather than more demand.

Is GA4 already showing AI traffic separately?

It may be. Google's channel list now includes an AI Assistant channel, described as visits from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. The same page says it excludes Google's own AI Overviews and AI Mode (Google Analytics Help).

Two things follow from that:

If you see an AI Assistant row in your Traffic acquisition report, start there. If you do not, your reports may still be grouping these visits under Referral.
Visits that come through Google's AI answers are counted with Google, not with AI assistants. You cannot see that slice in the AI Assistant channel.

The AI Assistant channel still depends on the same clues. A visit with no referrer and no tag will not land there, however it started.

A one-hour sanity check for your numbers

You do not need new software to get a better picture. Here is a check a marketing manager can run before the next review.

1. Look at the right report

In GA4, open Reports, then Acquisition, then Traffic acquisition. Switch the table to Session source / medium and search for "chatgpt", "openai", "perplexity", "gemini", "copilot" and "claude". Note what you find, including rows like chatgpt.com / (not set), which come from the tag without a matching medium.

2. Build a custom channel if you need one

If your property does not show AI visits clearly, create a custom channel group ([Google Analytics Help](https://support.google.com/ana