AI Traffic Tracking

Separate assistant referrals to the site from visibility in Google's generative search features. Neither dataset alone proves a specific assistant recommended the brand in an answer.

GA4 introduced a default AI Assistant channel in May 2026 for recognized assistant referrers. Check that channel and session source/medium before creating a custom grouping. The definition excludes Google's AI Overviews and AI Mode. A custom group is useful for verified uncovered sources; record its rules and effective date.

In acquisition reports or an Exploration, examine sessions, landing pages, countries and devices, then engagement, key events or revenue only with a clear metric scope. Compare raw source, channel, UTM and available referrer. A utm_source=chatgpt.com value records the link's tag and may survive copying into another app. Missing referrer does not prove fake traffic. Avoid an evergreen list of presumed AI source domains; verify observed patterns.

Use GSC's Generative AI performance report for supported AI Overview and AI Mode impressions. Record page, country, device and preliminary-data status. Do not use question-like queries as a measured AI-feature segment or equate impressions with assistant visits.

For an investigation, record the date window, acquisition scope, concentration by country/device, first visible URL, UTM, referrer evidence, downstream outcomes and known distribution. Label observed, calculated, inferred and hypothetical claims. Use traffic-analysis for broader anomaly diagnosis; neither workflow identifies a specific assistant answer from a source label alone.

For a recurring AI traffic report, read weekly-report. It defines the three measurement layers, comparable periods, source-classification limits and conditional report structure. Collect GA4 evidence through an available Analytics MCP session, a user export, an authorized browser session or the optional API collector; the current runtime does not write the narrative report automatically.

If using the optional project-configured GA4 API collector, run scripts/ with OPEN_ANALYTICS_PROJECT_ROOT set to the client AI report directory. The project supplies ai-source-registry.yaml (including ga4Hostname and candidate sourceRegex), landing-page-rules.yaml, and private scripts/.env; Open Analytics owns the reusable executable code. npm run fetch-all writes a two-period bundle to the project data/ folder. This collector treats source labels as candidate sessions; page-referrer rows remain separate diagnostic evidence, never added to source sessions. Reject legacy API raw files with the old dual-date parser. MCP and export evidence does not require a project .env or this JSON bundle; record its property, hostname, period, filters and completeness before analysis.