Growing unassigned or direct traffic is frequently AI referrals GA4 cannot attribute, because many assistants strip or omit the referrer when a user follows a link. It can also be tagging problems or consent settings. The pattern that points at AI is a steady, accelerating rise with no campaign change behind it.
The short version
- Many AI assistants strip or omit the referrer, so the visit lands as direct or unassigned.
- A steady, accelerating rise with no campaign change behind it points at AI referrals.
- It is not always AI. Tagging errors, consent mode and redirect chains cause it too.
- Treat any AI referral number you do have as a floor, not a count.
- The fix is a channel group plus a source of truth that is not GA4 alone.
Why this happens
When somebody clicks a link inside a generated answer, the visit often arrives without a usable referrer. Some assistants pass one, some pass it inconsistently, some pass nothing at all, and in-app browsers frequently strip it. GA4 has to put the session somewhere, so it lands as direct or unassigned.
The consequence is uncomfortable for anyone trying to prove this channel works: the better your AI visibility gets, the more your unassigned traffic grows, and the less credit AI gets for it. Firms end up looking at a rising unattributed block while concluding that AI search is not sending them anything.
It is not always AI, and jumping to that conclusion is its own mistake. The other common causes:
- Tagging problems. Broken or missing UTMs on campaigns, so paid and email traffic collapses into direct.
- Consent mode. Users declining tracking produce modelled or unattributed sessions.
- Redirect chains. Referrer data is frequently lost through a redirect, particularly after a site migration.
- Cross-domain gaps. Traffic moving between your site and a checkout or booking subdomain without proper configuration.
The diagnostic that separates them is the shape of the curve. Tagging errors appear as a step change on the date something broke. Consent effects track your consent rate. AI referral growth is gradual and compounding, with no matching change in campaign activity, and it tends to correlate with rising branded and question-shaped impressions in Search Console.
How to check it yourself
- Plot unassigned and direct over 12 months. A step change means something broke on a date. A curve means something is growing.
- Cross-check against your commerce or CRM data. If your platform shows orders your analytics cannot source, the gap is attribution rather than reality.
- Build an AI referral segment for the hostnames that do pass a referrer: chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, copilot.microsoft.com, gemini.google.com and you.com. Whatever that captures is a floor.
- Check landing pages. AI-referred sessions frequently land deep on specific answer-shaped pages rather than the homepage.
- Audit your UTMs before blaming AI for anything.
What to do about it
- Create a channel group for the assistant hostnames and review the list quarterly, because new surfaces appear and old ones get retired.
- State the caveat in your reporting: this is a directional floor, not a true count. Better to say it up front than to have it discovered later.
- Fix the mundane causes first. UTMs, redirects, cross-domain. They are cheap and they remove the noise hiding the signal.
- Do not judge AI visibility on referral traffic alone. Track citation presence directly, on a fixed prompt set. It is the only measure that does not depend on somebody else’s referrer policy.
- Give your commerce platform primacy on revenue. Analytics is for understanding behaviour, not for counting money.
Full treatment in how to measure AI search visibility and GEO metrics.