A practical AI visibility measurement methodology

A visibility score looks precise because it is expressed as a percentage. Its meaning depends entirely on the questions included, engines sampled, matching rules, and how variable generated answers are treated.

The decision this method improves

This method helps a marketer distinguish a durable category blind spot from normal answer variation. It also makes period comparisons defensible because the denominator and raw evidence remain visible.

A working method

1. Build prompts from real category, use-case, alternative, comparison, implementation, and trust decisions.
2. Group prompts by intent and keep wording stable during a comparison period.
3. Record engine, model where available, timestamp, full answer, brand position, competitors, sentiment, and returned citations.
4. Define brand aliases and ambiguous-name handling before counting results.
5. Review repeated losses across high-value prompts before choosing an intervention.
6. Rescan the same segment after enough time for public evidence to be discovered and retain both answers.

The evidence to keep

Keep the raw answer behind every metric. Store exact cited URLs when the engine exposes them, classify source type, and retain the page or placement shipped in response. When sources are unavailable, state that limitation instead of reconstructing an assumed research path.

The common failure mode

The most common failure is changing the prompt set while reporting a trend as though the sample remained constant. Another is combining broad awareness questions with bottom-of-funnel comparisons into one number that describes neither clearly.

How to measure the result

Report presence by intent group, repeated competitor wins, accurate descriptions, source recurrence, first-party citations, and later citation changes. Pair AI visibility with qualified referral visits and conversions when available, without claiming one action caused the result.

What to do next

Write the ten questions most likely to shape a buying shortlist. Run them across the engines relevant to your audience and save every answer in a simple table. A free domain check can prepare the private starting context after account creation.