How it works

From a product description to a measured answer, in five steps

  1. 1

    Describe what you sell

    Your product in plain language, your website, the category you compete in, and any other names people use for you. This is what prompt generation works from, so a vague description produces vague prompts.

    Alternate spellings matter more than they sound. A model writing "Acme Inc" when your project says "Acme" is still a mention, and the parser needs to know that.

  2. 2

    Name your competitors and topics

    The brands you lose deals to, and the problems a buyer is actually trying to solve. Topics drive the questions; competitors give every number something to be compared against.

    Without competitors we can tell you whether you appear, but not who is taking the space you are not.

  3. 3

    Review the prompt set

    We generate the questions a buyer would really type — best tools for a use case, alternatives to a named rival, what people regret buying. You edit, approve or reject every one.

    Nothing is sent to any AI provider until you approve it. Generating prompts is free; asking is what costs money, so the approval step is real rather than ceremonial.

  4. 4

    Run the scan

    Every approved prompt is asked of every provider you enabled, once each, and the full answer is stored exactly as returned.

    A provider that times out or has no key configured produces a recorded gap, not a zero. Scan health is shown next to every figure so you always know how much of the scan actually completed.

  5. 5

    Read the result, then repeat

    Mention rate, citation rate, share of voice and coverage — each with the counts behind it — plus the specific prompts where competitors appeared and you did not.

    Then schedule it. A single scan is one sample of a probabilistic system; the direction across several is the part worth acting on.

How a number is built
A worked example, so nothing about the arithmetic is mysterious.

Say you approve 10 prompts and enable 4 providers. That is 40 runs. Three time out, so 37 answered.

Your brand appears in 11 of those 37 answers, so your mention rate is 30% — and the product shows “11 of 37 runs” beside it, never just the percentage. The three that failed are excluded, not counted as misses, because we cannot tell the difference between not being recommended and not being asked.

Across the same 37 answers your competitors were named 73 times in total. Your share of voice is 11 of 84, or 13% — appearing in an answer alongside three rivals is a quarter of that answer, not all of it.

Only two of the four providers return sources, covering 18 of the answers. Three of those linked to a page you own, so your citation rate is 3 of 18, or 17%. The other two providers are outside that denominator entirely — counting them would make the number measure your provider mix rather than your attribution.

Run your first scan free