Definition and formula

For a prompt set P of n buyer-intent questions and an engine E: Visibility(E) = (prompts where the brand is named / n) × 100, with partial credit (0.5) for indirect mentions — e.g. the engine describes your product category accurately and names a list you appear in without naming your brand. A prompt set is graded when it mixes head questions ("best project management software"), mid-tail ("CRM for logistics companies under 50 staff") and long-tail situational prompts ("what should I use to track customs paperwork for EU shipments"). We recommend n ≥ 50 per category, 30 minimum.

Why per-engine scores are mandatory

Visibility diverges sharply by engine because retrieval stacks differ. A brand strong in Google's index may be absent from Perplexity's live retrieval if its pages are slow or JS-rendered; a brand with an active Reddit footprint may surface in ChatGPT and Perplexity while remaining invisible to Claude, which leans on primary documents. A single "AI visibility" number hides exactly the information you need to act.

What moves Visibility

  • Retrievability fixes — unblock AI crawlers, get indexed by Bing (feeds ChatGPT/Copilot), ship server-rendered content. Highest leverage per hour of work.
  • Answer-shaped coverage — publish pages that directly answer prompt-set questions with liftable answer blocks.
  • Source-graph presence — get listed, reviewed and discussed where the engine already looks: directories, review platforms, Reddit, trade press.
  • Entity clarity — one name, one description, everywhere; fragmented entities split your appearance credit.

Common measurement mistakes

Three ways brands mismeasure visibility: (1) testing only branded prompts — "what is [my brand]" — which measures training-data presence, not category visibility; (2) asking once — answers are probabilistic, so each prompt needs repeated samples (we use 3 runs and take majority); (3) ignoring language — for outbound brands, test prompts in the buyer's language, not yours.

Frequently asked questions

What is a good Visibility score?
Category-dependent, but as a working calibration from our benchmark runs: under 20 is invisible (buyers will not encounter you), 20–50 is present-but-losable, 50–75 is competitive, above 75 is category-defining. The absolute number matters less than the trend and the gap to your named competitors.
How many prompts do I need for a reliable score?
Fifty per category is our standard, thirty the floor. Below that, a single probabilistic miss swings the score by points. Split prompts across head/mid/long tail roughly 20/40/40.
Does visibility in one engine predict others?
Weakly. Shared infrastructure (Bing's index feeding ChatGPT and Copilot) creates correlation clusters, but cross-engine correlation in our benchmark runs is far lower than within-engine week-to-week stability. Measure each engine separately.