Analytics / November 2026

LLM Traffic in GA4: How to Track AI Referrals Before They Disappear Into Direct

AI assistants change how they tag links without notice. Here is a GA4 setup that holds up anyway.

Maxim Baeten
Maxim Baeten

10 min read

LLM traffic is the set of visits that reach your site from AI assistants such as ChatGPT, Gemini, Perplexity, Copilot and Claude. In GA4 it is harder to count than it should be. In the last two months, Gemini started and then stopped tagging cited links, ChatGPT lost its UTM tag for about a week, and the Gemini mobile apps send no referrer at all. A share of those visits now sits in Direct, and the AI line in your report looks flat while Direct climbs.

If you run paid media for a SaaS or e-commerce brand, this matters twice. You want to know whether AI answers send buyers, and since ChatGPT Ads opened up, you may also be paying for clicks from the same interface. Both need tracking you can trust.

This guide covers how AI referral traffic shows up in GA4, why part of it disappears into Direct, and the setup I use on client properties: a custom AI channel, clear UTM rules for AI ad links, and a short weekly check on Direct.

What LLM traffic looks like in GA4

GA4 has no default channel for AI assistants. A visit from an assistant lands in one of three places, depending on what the click carries with it.

What the click carries Example source / medium Default channel in GA4
A referrer only gemini.google.com / referral Referral
A UTM tag added by the assistant chatgpt.com / (not set) or referral Usually Referral or Unassigned
Neither (direct) / (none) Direct

The first two rows are recoverable. You can group them into one AI channel with a regex on the source. The third row is lost at the session level. GA4 cannot tell a Gemini app visit from someone who typed your URL, so the best you can do there is estimate.

A practical definition helps when you report on it: AI search traffic is any session whose source is a known AI assistant domain or an assistant-set UTM source. Everything else is a guess, and should be reported as one.

Why AI referral traffic ends up in Direct

Google defines (direct) / (none) as traffic that does not have a clear referral source. Its help page on direct traffic lists missing UTM parameters, redirects that strip them, and links opened from documents among the causes. AI assistants hit several of these at once.

smartphone

Apps that send no referrer

The Gemini iOS app opens links in its own WebView and sends no referrer, so the visit lands in Direct. Android app traffic landed in Direct in nearly every case in the server logs analysed by Natzir Turrado.

link_off

Links opened with noreferrer

When an assistant opens a cited link with noreferrer, the referrer is gone. If the UTM tag is missing as well, there is nothing left for GA4 to classify.

sell

UTM tags that come and go

Assistants add their own tags to cited links, but none of this is a documented contract. Tags have appeared and disappeared within weeks, with no announcement either way.

edit_note

Ad URLs the platform may rewrite

On the paid side, a platform feature that rewrites destination URLs can drop the UTMs you set. Then your paid clicks become Referral or Direct sessions.

Copy and paste adds to this. A user who copies your URL out of an answer and pastes it into a new tab arrives as Direct, whatever the assistant does with its links. No setup fixes that part.

What changed in September and October 2026

These are the changes that moved ChatGPT referral traffic and Gemini traffic between channels this autumn. None of the tagging changes was announced by the platform. The dates below are as reported by the sources named, and the Gemini dates in particular come from independent server logs, not from Google.

When (2026) What happened Reported by
From about 9 September Gemini starts appending utm_source=gemini to some cited links. Natzir Turrado's server logs, via PPC Land
Around 28 September Gemini stops adding the tag. Visits with a gemini.google.com referrer stay flat. PPC Land; Search Engine Roundtable (30 October)
8 to 15 October ChatGPT links lose the utm_source=chatgpt.com tag and the referrer after the GPT-6 release. A fix rolls out from 15 October; Huesler says OpenAI confirmed a bug to him. Cedric Huesler (Adobe), via Search Engine Roundtable and Relevant Audience
From 30 October ChatGPT Ads creative enhancement runs by default and can rewrite headlines, images and destination URLs. Relevant Audience, via a 2 November news roundup

The ChatGPT gap shows how large the shift can be. Across roughly 500 pages where ChatGPT drove at least 5% of visits, Huesler measured about 31,000 fewer ChatGPT visits and about 26,000 more Direct visits a day, as reported by Search Engine Roundtable. Sessions counted as Direct during that week stay in Direct. GA4 does not reprocess them.

The Gemini tag was useful while it lasted. According to PPC Land's write-up of Turrado's data, it surfaced 16 to 65 extra Gemini visits for every 100 visits that already had a Gemini referrer. Turrado treats that as a floor and puts Gemini's minimum share of Direct at 0.05% to 0.23% in a normal week. Small, but it is invisible in every standard report.

How to track LLM traffic with a custom channel group

The cleanest way to see AI traffic in GA4 is a custom channel group with an AI assistants channel. Google's own traffic source documentation describes this setup. Custom channel groups apply to your data retroactively, so you see history too.

1

Start from the default channel group

In Admin, open Data display, then Channel groups. Create a new group. GA4 starts it as a copy of the default group, so every other channel keeps its standard rules. Name it something like "Default + AI".

2

Add an "AI assistants" channel

Set one condition: Source matches regex, with the pattern below. Your own paid AI ad clicks stay out of this channel as long as you tag them with a source the pattern does not match, such as chatgpt_ads (see the UTM rules below).

3

Move it above Referral

GA4 assigns a session to the first channel whose rules match. If Referral sits above your AI channel, it catches chatgpt.com and gemini.google.com first and your new channel stays empty.

4

Check it in Traffic acquisition

Switch the primary dimension to your new channel group. Then add Session source as a secondary dimension and confirm only assistant domains sit in the AI row.

A starting regex for AI traffic in GA4

Source matches regex

^(.*\.)?(chatgpt\.com|openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai|deepseek\.com)$|^gemini$

The last part, ^gemini$, catches the utm_source=gemini tag Gemini used in September. Keep it in. If the tag returns, you will see it without touching the setup.

Treat the list as a start. Open your Referral report once a month, sort by sessions, and add any assistant domain you find. I prefer exact domains over a loose word match like "copilot" or "gemini" anywhere in the source, because loose matches pull in unrelated sites and make the channel harder to trust.

  • infoLimits. A standard GA4 property allows 2 custom channel groups and a GA4 360 property allows 5, each with up to 50 channels.
  • infoScope. The channel only sees sessions with a source. It cannot pull Gemini app visits back out of Direct.
  • infoOrder. GA4 places a session in the first channel it matches, so recheck the order whenever you add a channel.

UTM rules for your own AI ad links

Organic AI referrals and paid AI clicks are two different things, and they need to stay apart in your reports. If you run ChatGPT Ads, every destination URL should carry your own UTMs, set by you and not by the assistant.

  • arrow_forwardPick one source value and keep it. For example utm_source=chatgpt_ads. Avoid chatgpt.com, which is the value ChatGPT itself uses on organic citations.
  • arrow_forwardUse a paid medium. utm_medium=cpc places the click in a paid channel. Combined with the chatgpt_ads source, which the regex above does not match, it stays out of the AI assistants channel you just built.
  • arrow_forwardName campaigns the same way as elsewhere. Follow your existing UTM naming conventions so ChatGPT campaigns sit next to Google and Meta campaigns in one report.
  • arrow_forwardWrite values in lowercase. GA4 treats ChatGPT_Ads and chatgpt_ads as two sources.

Check that your UTMs survive URL rewrites

With creative enhancement on by default since 30 October, the platform may change the destination URL of an ad. So far there is no public word on whether UTM parameters survive that. Until there is, test it yourself:

  • checkDecide on the feature first. If you cannot verify tracking yet, turn creative enhancement off for your test campaigns.
  • checkCompare clicks to sessions. Each day, put platform clicks next to GA4 sessions with your chatgpt_ads source. A gap that suddenly widens points to lost tags.
  • checkWatch your ad landing pages. If the pages you only promote in ChatGPT Ads start getting chatgpt.com / referral or Direct sessions, your tags are being dropped somewhere.
  • checkUse a dedicated landing page or query value. A URL only used in ads makes this check much easier, because any non-paid session on it is suspicious.

The same logic applies to Google, Meta and LinkedIn. Our GA4 UTM tracking guide covers redirects, auto-tagging and the other places tags go missing.

A weekly check on direct traffic in Google Analytics

Since part of your AI search traffic will always land in Direct, the useful question is whether Direct moves in ways normal visitor behaviour does not explain. This check takes about ten minutes a week.

1. Track Direct share, not Direct sessions

Direct as a percentage of all sessions is more stable than the raw count. A jump of a few points in one week without a campaign, PR mention or offline push is worth a look.

2. Look at Direct by landing page

People type your homepage. Few people type a long blog URL or a deep product page. Direct growth on pages that AI answers tend to cite is the clearest sign that assistant visits are hiding there.

3. Split by device

App visits without a referrer show up as mobile Direct. If mobile Direct rises while desktop Direct stays flat, apps such as Gemini on iOS and Android are a likely cause.

4. Read Direct and AI together

When the AI channel drops and Direct rises in the same week on the same pages, suspect tracking before you suspect demand. That was exactly the pattern during the ChatGPT gap in October.

5. Annotate known gaps

Add GA4 annotations for 8 to 15 October 2026 (ChatGPT) and 9 to 28 September 2026 (Gemini tag active). Anyone comparing periods later will know why the lines bend.

If you have access to server logs or a server-side tracking setup, add one more step. User agents can identify some app traffic GA4 cannot, such as the GeminiiOS user agent that Turrado found in his logs. Compare those counts with your GA4 Direct numbers once a month.

How to report AI search traffic to stakeholders

Leadership will ask how much traffic and revenue AI brings. The honest answer has two parts, and I report them separately.

  • arrow_forwardMeasured. Sessions, conversions and revenue in the AI assistants channel. This is a minimum, and you should label it that way.
  • arrow_forwardEstimated. The likely share hiding in Direct, based on the landing page and device checks above. Give a range and say how you got it.
  • arrow_forwardPaid, kept apart. ChatGPT Ads results belong in your paid reporting with spend and cost per conversion, next to your other channels.

AI answers also shape demand without a click. Someone reads an answer, remembers your brand and searches for it a week later. That visit shows up as branded search, and no channel group will move it. If branded search and Direct both grow while the AI channel stays small, that is worth saying in your report too. The same applies to AI Overviews in Google Search, where answers absorb clicks that used to reach your site.

Where aubado fits

We are building two apps for this kind of work. UTM Organizer lists every UTM in use and flags duplicates, misspellings and tags that stop driving traffic. Analytics Brief turns GA4 data into a short daily read, so a sudden shift between channels gets noticed in days instead of at month end. Both are in development. If you want early access, join the waitlist.

Frequently Asked Questions

Catch tracking gaps in days, not at month end

aubado is building UTM Organizer and Analytics Brief to flag broken tags and sudden channel shifts in your GA4 data. Join the waitlist for early access.

Explore features

Related Resources

Share this article

send