How to measure whether AI assistants name your store
Most attempts at measuring this go wrong the same way: the merchant asks an assistant about their own brand, sees a flattering answer and concludes they are visible. That question was always going to name them. It measures nothing.
In short
- Branded questions prove nothing. Only unbranded questions measure whether an assistant would introduce you to a stranger.
- Freeze the question set. If the questions change between runs, the trend line is noise.
- Record which brands were named instead of you; that list moving is a better signal than your own number.
Separate branded from unbranded
Is Acme good for running shoes will name Acme. What are the best running shoes might not. Only the second tells you whether an assistant would introduce you to someone who has never heard of you, and that is the entire question.
Freeze the question set
Assistants vary between runs. If your questions also change, the resulting line is noise. Write the set once, derived from your actual product types and price bands, and reuse it every time.
Record who was named instead
Your own absence is half the story. The brands named in your place tell you who the assistant currently considers the obvious answer, and watching that list change is a far better signal than watching your own number alone.
Expect it to move slowly
Clearing exclusions can change things quickly, because a product goes from absent to present. Being preferred over an established brand is slower, and no tool that promises otherwise is being straight with you.
Doing it without a tool, and what changes with one
None of this requires software. Ten questions, a spreadsheet, once a month, and the discipline not to rewrite the questions is a genuinely sound programme, and it is better than a tool used carelessly.
What automation buys is consistency and the record. Endcap derives the question set from your actual product types and price bands, freezes it per store, separates branded from unbranded, and keeps the competitor list from each run. It also reports Not measured rather than 0% when a catalog is too sparse to build an unbranded question, because reporting zero there would accuse a store of being invisible on a check that never ran. If you build your own version of this, that is the trap worth avoiding.
Common questions
Why not just ask if my brand is good?
Because that question was always going to name you. It measures the assistant's willingness to repeat a brand name back, not whether it would recommend you to someone who has never heard of you.
How many questions do I need?
Enough to survive variance, which in practice means at least six and preferably more. What matters more than the count is that the same questions are reused every time.
What if my catalog is too small or too varied to build questions from?
Then say so rather than reporting zero. A store with no clear category cannot have an unbranded question built from it, and reporting 0% there accuses it of being invisible on a check that never ran.
Check your own store
Enter your Shopify URL and see whether the public products we can inspect show these problems. No install required for the first scan.