Conversion Attribution: How to Measure in the Zero-Click Era
A user sees your answer in AI, doesn't click through to your site, and three days later searches your brand directly and converts — who gets credit for that order? Zero-click makes attribution harder, but it's not unsolvable.
Bottom line first: zero-click doesn't break attribution — it only breaks the "last click" model. You need to look at four signals together: branded search volume, direct visits, referral sources, and the source customers state themselves when they inquire.
1. Why last click misleads
A user sees your name in AI but doesn't click at the time; two days later they search your brand name directly on their phone and convert. Under last-click attribution, that order counts as "branded organic search," and the AI exposure gets zero credit.
The data isn't wrong — the model doesn't fit. The decision process has moved upstream, but the recording point is still at the end.
2. Four usable signals
| Signal | Where to get it | What it tells you |
|---|---|---|
| Branded search volume | Search console, webmaster tools | Downstream response after answer exposure — the most sensitive |
| Direct visits / sessions with no source | Site analytics | Someone remembered the name and came back |
| AI platform referrals | Conversational domains in the Referer | The portion that actually clicked through |
| Self-reported inquiry source | "How did you find us" on the form | The simplest and also the most accurate |
3. Make "How did you find us" a field
This is the lowest-cost, highest signal-to-noise move. Include options like "AI assistant recommendation," "search engine," "friend referral," and "trade show."
Three months later you'll have a table you can't get anywhere else: exactly what share AI brought you. A small sample is still better than none.
4. Three things not to do
- Don't treat GEO as a pure ROI project: it behaves more like an asset, decays slowly when you stop investing, and measuring it by single-month ROI will always get it killed.
- Don't use total traffic as your only metric: AI traffic is small but high-intent, and it gets diluted in the total.
- Don't draw conclusions from a single Q&A: model answers fluctuate each time — watch the four-week trend, not a single day.
5. A minimal dashboard you can actually implement
Four numbers: mention rate for core questions, branded search volume, direct visits with no source, and the share of inquiries that self-report "AI/assistant." Record once a week; in eight weeks you'll see the direction.
Related reading: Four metrics that reflect change earlier than traffic, Why AI doesn't give you clicks.
