A buyer asks ChatGPT for the best tool in your category and gets a shortlist of three brands. They pick one, land on the site and convert a week later. In GA4 that visit shows up as direct traffic or a branded search. The moment that shaped the decision, the AI naming a shortlist, is nowhere in your reporting.
Short answer: you cannot measure AI search in Google Analytics alone, because the influence happens inside the answer, before any click. Use GA4 to capture the clicks that do arrive, and a dedicated AI visibility tool to measure presence inside the answers.
The measurement gap is structural
Traditional analytics rests on the click: query, result, click, session. AI search breaks that chain in two places.
Most AI answers are zero-click. The model resolves the question in the chat, so the user gets what they need without visiting anyone.
When there is a click, the referral data is often generic or stripped, so you cannot tell that ChatGPT or Perplexity sent it.
So the part of the journey that builds the shortlist never reaches your funnel. You see the visit after the recommendation, never the recommendation itself.
What GA4 can and cannot see
Signal | Visible in GA4? | Why |
|---|---|---|
Click from a cited link with a clear referrer | Partly | Shows as referral if the referrer survives |
Click with stripped referrer | No | Lands in direct traffic |
Visit after seeing a brand in an answer, typed later | No | Looks like direct or branded search |
Being named without a link | No | No session is created at all |
Which competitors the AI named instead | No | Happens outside your site |
Which sources the AI used | No | Needs answer-level data |
Even where tracking works, the volume is small and the value is high. HubSpot's research, as summarised in its AI visibility tools roundup, reports that AI-referred ecommerce visitors convert at 11.4% globally versus 5.3% for organic search, and Ahrefs found AI-referred visitors converting 23 times better than traditional organic traffic, from a small base. Small and valuable is exactly the profile that gets ignored when you only watch totals.
Why this is dangerous, not just annoying
You can lose ground in AI answers for months while your dashboards look calm. The drop only surfaces later, when pipeline thins out and nobody can explain why. By then the cause is invisible and weeks old.
There is a second cost. You optimise what you measure. If AI visibility goes unmeasured, it goes unmanaged, and the competitors who do track it quietly compound an advantage. It is the same gap early movers opened in the first years of SEO. Worth understanding what AI visibility actually is before it widens.
What to track instead
Stop counting only the clicks after the answer and start measuring your presence inside the answer. Four signals do the work:
Visibility Score: how often and how prominently you appear across the prompts your buyers actually ask.
Share of Voice: how your mentions stack up against competitors in the same answers.
Sentiment: whether the model describes you positively, neutrally or with doubt.
Mentions vs citations: whether you are named, or actually used as a source. The difference matters.
You need this per platform, because each model pulls from a different source pool and answers differently. Being strong in ChatGPT tells you little about Perplexity or Gemini. There are 11 platforms worth tracking in 2026.
How to set up the GA4 side
GA4 still earns its place for the part of the picture it can see.
Create a custom channel group called "AI assistants" with a source rule that matches the referrers you see in practice, such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Check your own referral report first: the exact domains change over time.
Add a segment or exploration for that channel and compare sessions, engagement and conversions with organic search.
Watch branded search and direct traffic as a lagging indicator. A steady rise in both, with no campaign behind it, is often AI influence showing up late.
Add a "How did you hear about us?" field to demo and signup forms with an option for AI assistants. Self-reported attribution fills part of the gap that referrers cannot.
How to set up the answer-level side
Define the prompts that matter: category questions, comparison questions and "best tool for [use case]" variations.
Run them across the major AI platforms on a fixed cadence, not as a one-off snapshot.
Track movement over time, because models refresh and answers shift week to week.
Review the cited sources, so you know which pages the models actually rely on.
Running that manually across platforms every day is not realistic, which is why Trackbase runs the prompts every 24 hours and turns the answers into metrics you can act on. If you want a first look before committing, the free AI visibility checker shows how you appear today, and the pricing page lists the plans. For the practical GA4 workflow, see how AI visibility tools fit SEO workflows and GA4.
FAQ
Can GA4 track traffic from ChatGPT?
Partly. When a user clicks a link in an answer and the referrer is passed, GA4 can show it as referral traffic. Many visits arrive without a usable referrer and are counted as direct, so GA4 always understates the real influence.
Why is my direct traffic rising?
There are many possible causes, but AI recommendations are one. A buyer who sees your brand in an answer often types your name or domain later. Combine direct and branded search trends with answer-level visibility data to see whether the timing matches.
What is the minimum setup to measure AI search?
A GA4 custom channel group for known AI referrers, a self-reported attribution field on forms, and a tracked set of prompts across the main AI models with a Visibility Score and cited sources. That covers both the clicks you can see and the answers you cannot.
Ready to win in AI search?
Join the brands already tracking their visibility across ChatGPT, Gemini, Claude and beyond. Start your free trial today.




