AI referral reporting is a classification problem with missing data. A session with no referrer is not proof that a person arrived from an AI assistant.
Use three attribution buckets
- Identified referrer: the analytics system records a known assistant domain or app referral. Preserve the raw source and landing page.
- Tagged campaign: a tracked link carries campaign parameters. Report it as tagged traffic, even if the user later switches apps.
- Unattributed or unknown: direct, stripped, private-share, app, and missing-referrer visits that cannot be reliably assigned. Keep them unknown.
Maintain an allowlist of observed referrers with the date and reason each was added. Verify actual traffic before adding a new pattern. Do not label every “direct” session or branded search as AI traffic.
Keep the measurement chain honest
Separate sessions, engaged visits, contact clicks, messages sent, qualified inquiries, and confirmed outcomes. If a user copies an answer into a private chat and then types the site URL, the origin may be invisible; no regex can restore evidence that was never transmitted. A model mention in a manual prompt check is also not an analytics referral.
Review source/medium and landing pages, test channel rules on known examples, and keep a change log when adding patterns. Compare identified AI referrals separately from unknown traffic. Explain coverage limits in every report and avoid presenting a regex as a complete measurement solution.
Optional AI-assisted work brief — Referral classification rules
Create transparent rules for identifying likely AI referrals from available referrer data. Define exact domains, unknown or stripped referrers, campaign-tag precedence, bot traffic handling, and a manual review category. Keep “known AI referral” separate from “AI-assisted influence.” Document that attribution can be lost and avoid renaming direct traffic as AI.
Worked example — Do not mislabel direct visits
If an AI answer influenced someone but the browser strips the referrer, analytics may show direct or unknown traffic. Keep that category intact. Report only identifiable referrer domains as observed AI referrals and describe unobserved influence as unmeasurable with current data. Do not inflate AI traffic by relabeling unattributed sessions.
AI disclosure: AI tools assisted with drafting. No referral volume or attribution accuracy is claimed.