Skip to content
Back to the blog
AI Visibility

A practical framework for measuring AI share of voice

Define a repeatable AI share-of-voice method using a stable prompt set, explicit mention rules, and careful trend interpretation.

The recrawl team8 min read

AI share of voice is a way to summarize how often a brand appears across a defined set of generated answers. It is useful only when the prompt set, engines, collection schedule, and rules for counting appearances are clear.

Start with the market you want to observe

Build prompts around real stages of discovery, comparison, and evaluation. Include the products, services, locations, and problems that matter to the business. Prompts should be specific enough to produce relevant answers, but broad enough that the set does not simply repeat brand-led queries.

  • Document every prompt and the intent it represents.
  • Choose the answer engines and model surfaces to observe.
  • Define brand aliases and competitor names before collection.
  • Record mentions and linked citations as separate fields.

Choose a counting rule

A simple method divides the number of responses that mention a brand by the total number of collected responses in the prompt set. A competitive method divides a brand's appearances by all tracked-brand appearances. Either can be valid, but they answer different questions and should not be mixed in one trend line.

A percentage needs a denominator

Always show what was counted. The prompt count, engine coverage, collection period, and treatment of repeated mentions are necessary context for interpreting share of voice.

Read direction before magnitude

Generated answers can vary, so small changes may be noise. Look for sustained movement across repeated collections, then inspect the underlying prompts and citations. The summary metric points to where investigation should begin, while the response-level evidence explains what changed.

Keep reading

More from the recrawl team.

Continue with related guides to AI visibility, search measurement, performance, and reporting.