How to measure AI Share-of-Voice
The weekly prompt battery behind a credible AI visibility report.
What is AI Share-of-Voice?
AI Share-of-Voice measures how often answer engines mention your firm, your competitors and your sources across the prompts buyers use to form a shortlist. It turns AI visibility from a vague feeling into a repeatable baseline.
The point is not to claim perfect attribution. Answer engines vary by model, freshness, location, session and prompt wording. The point is to run the same prompt set on a regular cadence so you can see whether your firm is absent, appearing, cited or being framed correctly.
Which prompts should you measure?
Start with prompts that match buyer intent, not vanity terms. A consulting firm might test "best firms for post-merger integration in healthcare", "who can help our leadership team adopt AI responsibly" or "which advisory firms specialize in B2B pricing strategy".
Group prompts by problem, industry, service and buying stage. Include comparison prompts, risk prompts and "what should we look for" prompts. These are often closer to how buyers actually use answer engines than a short keyword phrase.
What should the report capture?
A useful report records the engine, date, prompt, answer text, named firms, source URLs, position in the answer, sentiment and whether your firm was described accurately. It should also record the competitors the engine named, because those firms reveal the current answer set.
Screenshots can help with auditability, but the structured log matters more. You want a dataset that shows patterns over time: which prompts never mention you, which sources get cited repeatedly and which content gaps block the engine from understanding your fit.
How often should you run it?
Weekly is a practical cadence for most teams. Daily runs create noise. Quarterly runs hide movement for too long. A weekly prompt battery gives the team enough signal to connect publishing, technical fixes and third-party mentions to changes in the answer set.
Do not overreact to one answer. Look for repeated patterns across engines and prompt groups. If your firm is never named for a service you claim to own, that is a strategy problem. If one engine misses you once, that is a data point.
How do you use the findings?
Use the baseline to prioritize work. If competitors are named because they have clearer service pages, build better service pages. If answer engines cite directories, improve those profiles. If they quote educational content, publish stronger answer-first explainers. If your firm appears but is described poorly, fix entity clarity and internal links.
Measurement should guide the method, not replace it. The value of AI Share-of-Voice is that it shows where authority is missing and whether your visibility work is becoming easier for engines to verify. When the baseline shows engines answering buyer questions without ever naming you, the diagnosis in why does ChatGPT never mention our firm explains which evidence gap to close first.
What mistakes make the number useless?
The most common mistake is treating one prompt as the market. A single answer can change with wording, model behavior or context. Another mistake is measuring only brand mentions and ignoring whether the answer is accurate, sourced and connected to the right service.
Avoid vanity dashboards. A credible report should tell the team what to fix next: which source is missing, which page is weak, which competitor owns the answer and which claim needs public proof. If the report does not lead to action, it is only a screenshot collection.
Related reading
