Most SaaS teams already receive trials from ChatGPT and Perplexity. Almost none can prove it, because the referral shows up in analytics as direct traffic or vanishes into a generic referrer.
That measurement hole has a cost. The channel that influenced the signup gets no credit, so it gets no budget, and the team keeps funding the channels analytics can see.
Fixing attribution here is unglamorous plumbing, four pieces of it. None requires new tooling, and together they turn "we think AI sends us trials" into a number you can defend in a budget meeting.
Why does AI traffic hide in your analytics?
Because the answer layer breaks the referrer chain in several ways at once. Understanding which is which tells you what is recoverable.
Where does the referrer survive?
Some AI surfaces pass a readable referrer. Perplexity and some ChatGPT link clicks arrive labelled, so a referral report already shows a slice of the truth.
That slice is the floor, not the total. Treat any labelled AI referral count as a minimum bound on the real number.
The floor still earns its keep. It moves in the right direction when visibility moves, and a floor that triples is an argument no one can dismiss as survey bias.
Where does it break?
Three places. Users read the answer and type your domain later, which lands as direct, and assistants open links in ways that strip the referrer. And the answer resolves the question with no click at all, influencing a signup that arrives days later through search or direct.
The last one is the big one. Zero-click behaviour means the influence and the visit are different events, and no referrer will ever connect them.
Assistant browsing modes muddy it further. An agent that fetches your page on the buyer's behalf shows up as bot traffic, or nothing, while the human it briefed arrives later with no trail at all.
What are the four pieces of the fix?
Layered, cheapest first. Each catches traffic the previous one misses, and the four together bound the truth from both sides.
1. Clean up referrer capture
Group every AI referrer you can see into one channel in your analytics. Each surface arrives under its own origin, and a channel group turns the scattered rows into a trend line:
- ChatGPT referrals arrive from its chat domains
- Perplexity passes its own domain as referrer
- Copilot shows under its Microsoft origin
- Gemini surfaces under its Google origin
- Everything unrecognised stays out of the group, so the floor stays honest
This is an hour of configuration. It is also the step most teams have not done, which is why their AI channel reads as zero.
Check the channel group quarterly, because assistants change how they open links. A referrer string that identified a surface last quarter can vanish in a product update, and the channel quietly shrinks for reasons that have nothing to do with your visibility.
2. Ask at signup
Add "How did you hear about us?" with an explicit AI assistant option to the trial flow. Self-reported attribution is imprecise and directionally excellent, and it is the only method that catches the zero-click case.
Keep the field optional and the options short. The AI option's share, tracked monthly, is the single most convincing number this whole exercise produces.
One design detail matters: put the AI option in a randomised position, not first. Order bias in these fields is real, and you want the number defensible when someone challenges it.
3. Watch branded search and direct as a proxy
AI influence shows up as people arriving already knowing your name. A rising branded-search and direct trend that tracks your visibility gains on the prompt set is circumstantial evidence, and circumstantial evidence across three methods becomes persuasive.
Correlation windows matter here. Compare 8 to 12 week periods, not weeks, because both series are noisy.
Segment the branded lift by geography if you sell in multiple markets. Visibility gains concentrate where the content shipped, so a lift that matches your target market is stronger evidence than a global blur.
4. Reconcile against your prompt-set scans
The scan data says which prompts you appear on; the funnel data says what signed up. When citations on commercial prompts rise and the AI-attributed trial share rises a few weeks later, you have the closest thing this channel offers to causal evidence.
This is the layer that turns visibility measurement into pipeline conversation. Neither dataset persuades a CFO alone; the pair does.
How do you report it without overclaiming?
State the floor and the estimate separately, and label the method on both. Credibility here is worth more than a big number.
| Metric | Source | What it honestly claims |
|---|---|---|
| Labelled AI referrals | Analytics channel group | Minimum bound, direct clicks only |
| Self-reported AI share | Signup survey | Directional, catches zero-click |
| Branded and direct lift | Analytics trend | Circumstantial corroboration |
| Prompt-citation correlation | Scan data plus funnel | The causal story, stated carefully |
The overclaim to avoid is assigning every unexplained direct signup to AI. The channel is real; inflating it discredits the real part. Show the visibility score movement alongside the funnel numbers and let the two series make the argument together.
What does this change in practice?
Budget allocation, mostly. Once AI-attributed trials are a tracked series, the content and listing work that moves citations competes for budget on evidence instead of faith.
The series also settles arguments nothing else can. When someone claims the AI channel is hype, or conversely that it explains a good quarter, the four-layer number is the only grounded answer in the room.
It also changes which prompts you prioritise. Commercial prompts that correlate with trial movement get resourced first, and purely informational prompts stop absorbing production effort they cannot repay. That prioritisation is the whole point of running answer engine optimization as a measured program rather than a belief.
Expect the honest number to start small. A low single-digit share of trials, growing, is the normal early reading, and it is enough to justify the channel because the trend and the trajectory are what a budget decision actually needs.
Where Unveilr fits
Unveilr supplies the scan half of the reconciliation: per-prompt presence and citation over time, so your funnel data has something to correlate against. Agents scan, detect which commercial prompts moved, feed fixes into content, and re-scan, while your analytics tracks what arrived.
Clients who wire the signup survey in week one thank themselves in month three. It is the cheapest instrument in the stack and the hardest evidence it produces.
In one D2C case study, the brand moved from the 9th most-cited domain in its category to number 1, with ChatGPT visibility rising from 3.3% to 44.7%.