TL;DR
FACT: an AI visibility score is only as good as the questions, models and schedule behind it. A number without those three things is decoration.
INFERENCE: the fastest way to judge a tool is to ask it a question you already know the answer to — your own brand — and check whether the raw answer it shows you matches what the product claims.
Ask what questions it asks
Every visibility tool works the same way underneath: it asks models questions and reads the answers. If the vendor will not show you the exact question set, you cannot know whether the score measures your market or a generic prompt list that flatters everyone.
A credible tool lets you see — and ideally set — the questions. Ours are published on the Experiments page, word for word, because a score you cannot inspect is a score you should not pay for.
Ask which models, and which products
FACT: an answer from an API model is not the same as an answer from the consumer ChatGPT, Perplexity or Gemini app a buyer actually uses. We learned this running our own tests: our October pilots used API models, and we label them as stand-ins precisely because the consumer products can answer differently.
When a tool claims to measure "ChatGPT visibility," ask whether it means the app or an API call with a similar name. The honest answer changes what the score is worth.
Ask to see the raw answers
A score is a summary. The raw, dated, verbatim answers are the evidence. If a tool shows you a trend line but not the answers underneath it, you are trusting the summarizer — and summarizers are exactly the thing this industry is trying to measure.
FACT: every result on our Experiments page links to a dated file containing the full answers, including the runs where we scored zero.
Ask what it cannot tell you
No tool sees what a logged-in user sees with personalization on. No tool guarantees an answer will repeat tomorrow. A vendor who admits this is describing the measurement honestly; a vendor who promises certainty is selling it.
INFERENCE: treat any single run as a sample, not a verdict. The value of a tool is the loop — same questions, same schedule, comparable scores over time — not any one number.
Canonical definitions used in this post
AI visibility and related terms: [What is BackTier](/what-is-backtier) and [GEO, AEO, AIO, HEO and SEO, defined together](/answers/geo-aeo-aio-heo-seo). Other terms: [the glossary](/glossary).