ChecklistUpdated 2026-09-21

Seven questions to ask an AI visibility vendor

Ask these before buying AI visibility monitoring or optimisation. Each one separates a measurement from a claim: which step of the evidence ladder a number comes from, which models and dates it covers, and whether anything it reports was verified.

  1. When you report AI crawler traffic, are the requests verified? A good answer names the check — published IP ranges or reverse DNS — and says what happens to requests that fail it. "We detect AI bots by user agent" means unverified.
  1. Which exact models, and on which dates? "ChatGPT" is not a model. A usable report names the provider, the model and the date of every answer, because behaviour changes under a stable name.
  1. Do you count mentions, citations or links to the product page? These differ by orders of magnitude. Ask which one the headline number is.
  1. How do you separate a citation from use of a fact? A model can link a page and state something the page does not say. Ask whether they check the value, not only the link.
  1. What would your report look like if nothing changed? Answers drift on their own. A vendor who cannot show a baseline and its noise cannot tell you whether their work moved anything.
  1. Can you show a controlled result — one change, one comparison? If every result is a before-and-after screenshot, none of them is evidence of cause.
  1. What do you not claim? A serious vendor has a list: no ranking guarantees, no causation from a crawl, no revenue figures from a mention. If there is no such list, the report will not have one either.

Why each item is on the list

Related

← All checklists