GuideWhen the buyer asks an AI
AI share of voice
Definition
How often your company is named in AI assistants’ answers about your category, measured against how often your competitors are. It is the assistant-era counterpart of share of search.
How it’s measured
A tool runs a fixed list of prompts against assistants like ChatGPT and Claude on a schedule and counts which companies each answer names; your share is your mentions against the field’s. Two properties decide whether the number means anything. The prompts must be the questions buyers ask, phrased in their words; pull them from sales calls and the support inbox, because a list written in your own product vocabulary measures a conversation nobody is having. And the reading works as a trend: answers vary from run to run, and the models are updated underneath the measurement, so a single week’s level is mostly noise while the slope across months carries the signal.
What to watch for
A share reported without its prompt list attached is unreadable; the number is a function of the questions asked, so ask for the list first. A jump between weeks with nothing shipped on your side is likelier a model update than a marketing result, and if it is presented as a result, your own release calendar is what to hold it against. Check the prompts for your own product names; their presence flatters the score. Read a sample of the answers themselves each month: a mention that names you as the overpriced option still counts in the tally.
The question you ask
“Which prompts produced this number, and how has it moved since we started measuring?”