How to Track AI Search Traffic in GA4 & GSC (2026)

What changed in AI traffic measurement in 2026?

Two first-party reporting changes made many older guides obsolete:

• Google Analytics introduced an AI Assistant channel. On 13 May 2026, Google added automatic classification for recognised assistant referrals. Matching sessions receive the medium ai-assistant, the default channel AI Assistant and the campaign label (ai-assistant).
• Search Console introduced dedicated Generative AI reports. The Search report shows impressions for URLs displayed in AI Overviews and AI Mode, grouped by page, country, device and date.

The datasets answer different questions. GA4 records identifiable visits and on-site behaviour. Search Console records Google AI search visibility, including appearances that do not generate a click. Neither is a universal record of brand mentions inside ChatGPT, Gemini Apps, Perplexity or other assistants.

The four-layer AI measurement model

Step 1: Find AI Assistant traffic in GA4

1. Open Reports → Acquisition → Traffic acquisition.
2. Set the primary dimension to Session default channel group.
3. Find the AI Assistant row. If volume is low, extend the date range before concluding there is no traffic.
4. Add or switch to Session source / medium to see which identifiable assistant sent the visit.
5. Break the segment down by Landing page, country and device.
6. Add business metrics: key events, session key-event rate, total revenue, ecommerce purchases or qualified form submissions.

The default channel group is Google’s maintained classification and cannot be edited. Keep it as a stable reference. A session appears only when Analytics receives enough source information; an AI-influenced visit with no usable referrer may remain Direct or otherwise unidentifiable.

Step 2: Validate coverage with source/medium

Use an Exploration or the Traffic acquisition report to inspect recognised domains directly. Useful fields include:

• Session source / medium
• Session campaign
• Landing page + query string
• Country and device category
• Engaged sessions and engagement rate
• Key events, lead-quality status and revenue

Expect sources and classification to change. Maintain a dated list of sources actually observed in your property instead of copying an enormous regex from an old blog post.

When should you create a custom AI channel group?

Create one when you need a broader internal taxonomy, want to test sources not covered by the native channel, or need a client-specific grouping. Google documents custom channel groups and provides an AI-assistant regex example.

1. Open Admin → Data display → Channel groups.
2. Create a new group; do not replace the default reporting reference.
3. Add a channel such as AI referrals—custom.
4. Use Session source matches regex and start with confirmed sources from your own data.
5. Place the rule above generic Referral so the first matching rule assigns the session correctly.
6. Compare the custom result with the native AI Assistant row and document why they differ.

A controlled starter pattern could be:

^(chatgpt.com|chat.openai.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com|meta.ai|grok.com|deepseek.com|you.com)$

Test it against real Session source values. Do not match every domain ending in .ai; that can mix unrelated referrals into the channel. A custom grouping can reclassify identifiable historical sources in reports, but it cannot recover a source that was never sent.

Step 3: Measure Google AI Overviews and AI Mode in GSC

Open Search Console → Performance → Generative AI. The Search report currently covers impressions from AI Overviews and AI Mode. Use these dimensions:

• Pages: which canonical URLs appeared in a supported Google AI feature.
• Countries: where the impression originated; keep UK and US baselines separate.
• Devices: desktop, mobile and tablet distribution.
• Dates: hourly, daily, weekly or monthly trends where available.

The dedicated report currently provides impressions. It does not provide a query dimension, clicks, CTR or average position. AI report data is also included in the normal Web performance report, so do not add the two impression totals together.

How to connect the GSC AI report to GA4

There is no one-to-one join key between an AI impression and a GA4 session. Use a page-and-market measurement window:

1. Export GSC Generative AI impressions by page, country and date.
2. For the same canonical pages and date range, export normal GSC Web clicks and impressions.
3. In GA4, review Google organic sessions, AI Assistant sessions, landing pages and conversions for the same market and period.
4. Annotate major content, indexing, product and reporting changes.
5. Report correlation as correlation. Do not claim that an AI impression caused a specific session or lead.

Clicks from Google’s AI search features are part of the wider Google Search journey; do not assume they will appear in GA4’s AI Assistant row. Use the GSC AI report for exposure and the normal Web/GA4 views for visits and outcomes.

What about dark AI and Direct traffic?

Mobile apps, desktop clients, privacy controls and copied links can remove referral information. Those visits may arrive as Direct, but Direct is not an “AI traffic” bucket. It also contains bookmarks, typed URLs, untagged campaigns and other unattributed visits.

Use supporting evidence without rewriting attribution history:

• Review deep landing pages that receive new Direct demand, but label the pattern as a hypothesis.
• Add a “How did you hear about us?” field to qualified lead or checkout flows.
• Preserve landing page, first-touch and last-touch fields in the CRM.
• Compare branded-search demand and sales-call language around major visibility changes.
• Use controlled UTMs only on links you publish or distribute; you cannot force public assistants to retain them.

Server logs: useful, but not an AI traffic report

Server or CDN logs can show that a declared crawler requested a URL. They are useful for checking access policy, response codes, crawl frequency and which sections bots reach. They do not show whether a model indexed the content, cited it, recommended the business or sent a human visitor.

Keep crawler analysis separate from Analytics. Before changing robots rules, use the AI crawler and robots.txt guide and verify current official user-agent documentation.

Build an AI search revenue dashboard

Segment every KPI by market and service. A US informational visit and a UK buyer asking for implementation should not be averaged into one global AI number.

Common AI analytics mistakes

• Using an old regex as the whole strategy: start with the native AI Assistant channel and use custom rules as validation or extension.
• Adding GSC AI impressions to normal Web impressions: the dedicated report is a subset of the wider performance data.
• Calling every Direct visit AI-driven: the missing source cannot be reconstructed confidently.
• Treating bot requests as leads: crawler access and human sessions are different datasets.
• Reporting visits without business outcomes: include qualified leads, purchases, revenue per session or assisted pipeline.
• Combining countries and languages: report UK, US and local-language acquisition separately.

Turn measurement into an AI SEO action plan

The dashboard should finish with decisions: which commercial page earns Google AI impressions but no useful visits, which assistant sends qualified users, which market has no coverage, which landing page needs a stronger CTA, and which source or content gap blocks discovery. Maksut.net’s AI SEO and search visibility service connects those findings to content, technical access, entity clarity, authority and conversion work.