How to measure whether assistants recommend you
The method we use to turn "are we showing up in AI?" from an anxious guess into a number you can track. Run it yourself if you'd rather not hire anyone.
"Are we showing up in AI?" is the question every merchant asks and almost nobody can answer, because answering it honestly means running the queries and reading the output rather than reasoning about it. This is the method we use. It is not proprietary and there is nothing stopping you running it yourself.
Step one: build an intent set
Write roughly fifty questions a real buyer would ask in your category. Spread them across five kinds: category entry ("best X for Y"), constraint resolution ("X that works with Z"), comparison ("X vs a named competitor"), objection ("is X worth it"), and post-purchase ("X return policy").
The constraint questions matter most, because they are where an assistant has to verify a fact about your product rather than repeat a general impression.
Step two: run them properly
Each intent goes to each assistant you care about, several times, from clean sessions with no history. This last part is where most informal attempts go wrong: asking from your own logged-in account, on a brand you have been researching all week, tells you nothing about what a stranger sees.
Step three: score three things, not one
- Presence. Were you named at all? Who was named instead?
- Accuracy. If you were named, was what it said about you true and current?
- Reachability. Was a way to buy offered, and did it point at you rather than a marketplace listing or a reseller?
Collapsing these into a single score hides the diagnosis. A brand absent from every answer has a very different problem from one that is recommended enthusiastically and then linked to a reseller.
Step four: keep the intent set fixed
The number in isolation means little, models change, and so does the answer. What is worth having is the same intent set, scored the same way, re-run on a schedule, so you can see direction. That is the difference between a scary anecdote and something you can put in a plan.
What we deliberately do not publish
You will not find a leaderboard here of which brands are winning in which category. We could produce one, but a ranking published today is substantially wrong within weeks, and presenting it as durable would be dishonest. The method is the durable part, so the method is what we publish.