How to Measure Your AI Share of Voice
AI share of voice is how often your brand appears in AI assistants' answers compared with competitors. Learn a practical, repeatable way to measure it, the pitfalls that make results unreliable, and what to do with the numbers.
Short answer: choose a fixed set of questions your buyers ask, run them through the main AI assistants on a regular schedule, and record whether your brand and each competitor is mentioned. Your share of voice is your mentions divided by all brand mentions across those answers. Because answers vary from run to run, repeat each prompt and track the trend, not a single result.
Table of Contents
- What It Is
- Step 1: Build a Prompt Set
- Step 2: Run It Consistently
- Step 3: Record the Right Things
- Step 4: Calculate
- Pitfalls
- What to Do With the Result
- Key Takeaways
What It Is
In advertising, share of voice is your portion of total exposure in a market. Applied to AI assistants, it is your portion of brand mentions in answers to the questions that matter to your business.
It is a rough measure. Assistants do not publish what they say to whom, so you are sampling. Done consistently, the sample is still informative.
Step 1: Build a Prompt Set
Choose fifteen to thirty questions that a buyer would genuinely ask. Mix the types.
| Type | Example |
|---|---|
| Category | "What are the best session replay tools?" |
| Use case | "How can I see why users abandon checkout?" |
| Comparison | "What are alternatives to a well-known competitor?" |
| Constraint | "What is an affordable heatmap tool for a small store?" |
| Problem | "Why is my mobile conversion rate low?" |
| Regional | "Which analytics tools have pricing in rupees?" |
Rules:
- Write them as a customer would, not as you would.
- Leave your brand out of the prompt. You are testing whether you come up unprompted.
- Include problem questions, where no tool is requested. Being suggested there is the strongest signal.
- Fix the wording and reuse it every time.
Sources for realistic questions: sales calls, support tickets, search queries from Search Console, and community forums.
Step 2: Run It Consistently
- Which assistants: the ones your buyers use, typically ChatGPT, Perplexity, Gemini, Copilot, and Claude.
- How often: monthly is enough for most teams.
- How many times each: at least three runs per prompt per assistant, because answers vary.
- Clean conditions: a fresh conversation each time, logged out or with memory and personalisation off where possible, since history changes answers.
- Note the settings: whether web search was on, the date, and the region.
Step 3: Record the Right Things
For each answer, log:
| Field | Why |
|---|---|
| Brands mentioned, in order | For share and position |
| Whether you were mentioned | The core measure |
| Position if listed | First mention carries more weight |
| Whether a link to you was cited | Affects traffic |
| Which page was cited | Shows what content is working |
| How you were described | Accuracy check |
| Sentiment | Recommended, neutral, or cautioned against |
| Sources cited overall | Shows which third-party sites shape answers |
The description column is easy to skip and often the most useful. Assistants repeat outdated prices, retired features, or a competitor's positioning attached to your name.
Step 4: Calculate
Mention rate = answers that mention you ÷ total answers.
Share of voice = your mentions ÷ all brand mentions across all answers.
Illustrative example: across 90 answers, brands are mentioned 400 times in total and you account for 36 of them. Your share of voice is 9%. You appeared in 30 of the 90 answers, a mention rate of 33%.
Break both down by assistant and by prompt type. You may be well represented for comparisons and absent for problem questions.
Pitfalls
- One run is not a measurement. Answers differ between identical prompts.
- Personalisation. Your own account history skews results toward your brand.
- Leading prompts. Naming yourself in the question guarantees a mention.
- Changing the prompt set. Altering questions makes months incomparable. Add new ones as a separate group.
- Reading small changes as real. With ninety answers, a few mentions either way is noise.
- Ignoring accuracy. A mention that describes you wrongly is not a win.
- Model updates. A new model version can shift results overnight. Note release dates.
Track cited sources as closely as your own mentions
The third-party pages assistants cite in your category are your target list. Being included in those comparisons and reviews is often the fastest way to appear in answers.
What to Do With the Result
| Finding | Action |
|---|---|
| Never mentioned anywhere | Check crawler access first (guide), then positioning and third-party presence |
| Mentioned, described wrongly | Correct and clarify facts on your own site and wherever the wrong version appears |
| Mentioned for comparisons, not problems | Publish answer-first content for problem questions (how) |
| Competitors cited via the same review sites | Get included in those sources |
| Mentioned, no link | Make sure the cited facts live on a clear, crawlable page |
Automating it. A spreadsheet works for a small prompt set. DeepSync's AI visibility feature, on Business and Enterprise, tracks where your brand is mentioned in AI answers for topics you choose, alongside which AI crawlers read your site and the referral traffic that answer engines send.
Pair share of voice with outcomes: referral sessions from assistants and what those visitors do (tracking AI traffic).
Key Takeaways
- Use a fixed prompt set written in your customers' words.
- Repeat each prompt and track trends.
- Record accuracy and cited sources, not only mentions.
- Act on the gaps: access, content, and third-party presence.
Conclusion
AI share of voice is an imperfect measure of something that matters. Measure it the same way each month and it will tell you whether your visibility is moving in the right direction.
Frequently Asked Questions
Related articles
Stay in the loop
Get the latest insights on product analytics and user behavior delivered to your inbox.
