Every AI visibility report ends the same way: a mention rate, a citation count, a share-of-voice chart against two competitors, and a trend line pointing in a direction. It is useful. It is also the point where someone in the room asks the only question that matters: so what did it earn? Most reports cannot answer, because the tools that produce them were never in a position to know. A report should end somewhere else, and the number it ends on needs careful reading.
Four stages, and where most reports stop
The AI channel has a funnel like any other, and it is worth naming the stages plainly:
- Mention. An assistant names you in an answer. This is what visibility tools measure, by asking the engines the same questions on a schedule.
- Citation. The answer links to one of your pages. Better than a mention, because a link is something a person can act on.
- Referral. Someone actually clicks and lands on your site. Now you have a visitor, usually a deep-landing one who already asked the shallow questions elsewhere.
- Conversion. They buy, book, enquire or sign up. The only stage a finance person recognises.
Most reports stop at stage two because of where the tools stand. The tools that measure mentions and citations do it from outside your site, by prompting models and reading answers. They see what the engines say. They cannot see what happens on your checkout page, because they are not there.
Why outside-in tools stop at two
Give the visibility platforms their due. Polling the engines is genuinely hard to do well, and a clean competitor benchmark is worth paying for. But the vantage point sets the ceiling. A tool that lives in a dashboard on someone else's servers can tell you that Perplexity cited your pricing page four times last week. To tell you that one of those clicks turned into a booking, it would need to be present when the booking happened: inside the site, wired into the booking plugin, holding a memory of who sent that visitor.
The usual workaround is Google Analytics: segment sessions by referrer, look at conversions for the segment, paste the number into the report. It works, roughly. It also means your AI channel's revenue number lives in a different tool from everything else in the report, depends on a tag that many visitors block, and inherits every quirk of GA4's channel grouping. The number you want is the one from your own ledger: the orders table, the form submissions, the enrolments, with the assistant that sent each visitor written next to it.
That is the design behind AI Conversions in Viz: a plugin already sitting where the conversions happen credits each purchase, booking, lead and sign-up to the assistant that sent the visitor, with the value, on the server, and shows the result per assistant. Whatever tool you use, the last stage of the funnel can only be measured from inside the site.
How to read a per-assistant conversion table honestly
Once you have the number, the temptation is to headline it. Resist that for a quarter and read it properly first.
- It is a floor. Several assistant apps open links without a referrer, so a share of AI-sent visitors arrives looking like direct traffic and can never be credited. Your attributed count is lower than the truth by an amount you cannot know. Say so in the report, every time.
- Read the rate before the count. Twelve conversions means little on its own. Twelve conversions from three hundred AI-referred visits, against a site average of one in sixty, means the channel is pre-qualifying people and the true volume is higher than the floor you measured.
- Small numbers move a lot. One assistant showing a 9% conversion rate from eleven visits tells you nothing yet. Set a minimum before you compare assistants to each other, and compare trailing periods, not weeks.
- Value beats count. Five bookings credited to one assistant and twenty leads credited to another are not comparable until you attach value. If your conversions carry a currency amount, report that; if they are leads, report your own close rate next to them.
- An experiment needs both arms. If you show AI-referred visitors a next step and want to claim it lifted conversion, half of them have to not see it. A lift measured against last month is a correlation. A lift measured against the other half of the same visitors is a result.
The agent question, without the hype
Every report now wants a line about AI agents buying things. Assistants fetching pages on a person's behalf is already routine: pricing pages, booking pages, contact pages, checked by an agent that will summarise what it finds. An agent then completing a request on those pages is still rare. Both facts belong in the report.
What you can do today is make sure your site can answer an agent's questions and act on its request when it comes. Two checks are worth running. First, where do your prices and availability actually live? If they are inside an embedded widget from a third-party booking or checkout service, an agent reading your page sees a script and leaves. Second, is there anything on your site an agent could submit? A machine-readable contact or quote form, with every field named and typed, turns an anonymous agent visit into a lead with a person behind it. Viz publishes one at /ai-contact/ and points agents at it from your money pages; whether you use that or build your own, the principle is the same. Since Viz 2.12 the same site also registers WebMCP tools on every page, so an agent running in the visitor's browser can describe the business, send the request or, on a store, fill the cart, and each call shows up in the report as an agent request. The channel is early. The sites that are readable and reachable when it stops being early will get the first orders.
What the last section of the report should say
A workable template, five lines, in this order:
- AI-referred visits this period, with the referrer caveat stated once.
- Conversions and value credited to AI assistants, as a share of all conversions on the site.
- Conversion rate for AI-referred visitors against the site average.
- The assistants ranked by conversions and by rate, with a minimum visit count before ranking.
- What agents came to do on the site this period, which tools they called, and whether the site could answer them.
Then the mention rate, the citations and the share of voice, as the leading indicators they are: visibility explains the conversions, and the conversions justify the visibility work.