Research

What the assistants say, measured.

We ask the questions buyers ask, in any category, on ChatGPT, Perplexity and Gemini, and count what comes back: who is named, how often, and from which sources.

Reading an answer · How a category is measured · Case files

Anatomy of an AI answer

Five things every answer tells you.

One answer, read the way we read all of them. The example is illustrative: the category is real, the firms are not.

  1. 01 · Named

    Who was named.

    Three firms make the answer. For a buyer, that is the shortlist.

  2. 02 · Order

    In what order.

    The first name is read first. Order changes between runs, so we record it every time.

  3. 03 · Described as

    How each was described.

    A few words decide how a buyer sees each firm. We record them word for word.

  4. 04 · Sources read

    Which sources were read.

    Most answers point to the pages the assistant read. Some show none. Your own pages and profiles among them are where the work happens.

  5. 05 · Missing

    Who was missing.

    A firm not in the answer is not on that shortlist. That is the line most companies never see.

How a category is measured

Six questions. Three assistants. Three runs.

The questions are written the way buyers in that category ask, in their country, with no company named. Each one runs three times on each assistant, in fresh sessions. Fifty-four answers make one baseline.

The full method

Tell us your category and your buyers' country.

We run the same test on your buyers' questions, three times per assistant, with every answer on record.