How to Track AI Traffic in GA4 and Search Console

AI engines are sending you traffic and telling you almost nothing about it. The platforms report exposure without clicks, the assistants often strip the referrer, and the most valuable outcome, being named in an answer nobody clicked, produces no analytics event at all. Here is how to build the best picture currently available, and where to stop pretending precision exists.

What each platform actually gives you

Source	What you get	What is missing
Search Console generative AI report	Impressions, pages, countries, devices, dates	Clicks, prompts, position
Bing Webmaster Tools AI report	Comparable exposure metrics, available globally	Clicks
GA4	Sessions, engagement, conversions from AI referrers	Impressions, mentions, unlinked citations
Prompt monitoring	Presence, sentiment, competitor share	Traffic

Scroll the diagram sideways to see all of it.

No single source covers the funnel. You need at least three of the four, and you need to be clear which question each one answers.

Search Console’s generative AI performance report

Google began rolling this out in June 2026, initially to a subset of UK site owners, with expansion promised. It covers appearances in AI Mode and AI Overviews across Search and Discover, and reports:

• Impressions: how often your URLs appeared in generative AI features.
• Pages: which URLs appeared.
• Countries and Devices.
• Dates, with hourly through monthly granularity.

There is no click data. Asked directly, Google said it would “introduce additional metrics over time” without committing to clicks. Plan on exposure-only reporting from this surface indefinitely.

Bing shipped its equivalent earlier and made it global from the start. If you have not verified your site in Bing Webmaster Tools, do it now: it is the only free AI exposure report most sites can actually access today, and Bing’s index disproportionately shapes ChatGPT recommendations, as we cover in retrieval vs citation.

Building AI referral reporting in GA4

AI assistants show up as plain referral traffic. GA4 will not group them for you, so you build the grouping yourself.

1. Know the hostnames to watch

The recurring ones are chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai, and you.com. Treat this as a living list and re-check it quarterly; new surfaces appear and old hostnames get retired.

2. Create a reusable segment

In an Exploration, build a segment where Session source matches a regex covering those hostnames. Save it. Every subsequent question, which pages, which conversions, which countries, becomes a matter of applying that one segment.

3. Compare against your baseline, not against zero

The useful question is not “how much AI traffic do we get” but “how does it behave differently.” Put your AI segment beside organic search and look at engagement rate, pages per session, and conversion rate. AI-referred sessions frequently arrive later in the buying cycle because the assistant already did the comparison work, so raw volume understates their value.

4. Mark the date you started

Because the hostname list changes and GA4 does not backfill segment definitions retroactively in reports you build later, note when each source was added. Otherwise a jump in “AI traffic” six months from now may just be you adding a hostname.

5. Watch direct traffic as a proxy signal

Some assistants omit or strip the referrer, so a share of AI-driven visits lands in direct. A rising direct segment on deep interior pages, the kind nobody types from memory, is a reasonable indicator of uncredited AI referrals. It is a signal, not a measurement.

What none of this captures

The most important AI outcome is invisible to all of it: being named in an answer where the user never clicked. With AI Overviews driving high zero-click rates, the brand impression often is the whole result. Someone who reads “Ambeltek, a studio specialising in AI SEO” inside an answer and searches your name a week later shows up as branded organic, not AI traffic.

This is why traffic reporting alone will systematically undervalue AI work, and why you need a parallel measurement track that does not depend on clicks:

• Presence rate across a fixed prompt set, covered in measuring brand visibility in ChatGPT.
• Share of model against named competitors, defined in share of model.
• Branded search volume as a downstream proxy for answer exposure.

Our rundown of the AI visibility metrics that actually matter covers which of these are worth the reporting overhead.

A reporting cadence that holds up

• Monthly: GA4 AI referral segment, branded search trend, Search Console and Bing AI impressions.
• Monthly: prompt-set run for presence, sentiment and share of model.
• Quarterly: refresh the hostname list, re-baseline the prompt set, re-check competitor movement.

Keep the two tracks side by side in one report. Exposure without traffic is normal in this channel, and a stakeholder who only sees the traffic number will draw the wrong conclusion.

Where this leaves you

You can measure AI exposure and AI-referred behaviour reasonably well. You cannot yet measure AI-attributed revenue cleanly, and no platform has signalled that clicks are coming. Build the reporting you can, label its limits explicitly, and pair it with prompt-level monitoring so the zero-click majority does not go unrepresented.

If you want this set up as a running scorecard rather than a one-off spreadsheet, that reporting layer is part of our AI visibility audit.