How to Measure AI Share of Voice: A Step-by-Step Method

AI share of voice is your brand's mentions divided by all tracked-brand mentions in the same AI answers, shown as a percentage, per AI SEO Tracker. To measure it, fix your prompts, competitor set and engines, repeat each prompt, and report the metric, the margin of error and the method together.

• Fix prompts, competitors and engines before the first run, and do not change them afterwards.
• Repeat each prompt on each engine. A single run tells you little.
• Name the metric in every report, because mention rate, mention share and citation share can rank you differently.
• Publish the margin next to each figure. Changes inside the margin are noise.

What do you need before you start?

You need four things: a fixed list of prompts, a fixed list of brands, the AI engines you will query and a log of every answer. Write them down before the first run. A list that changes midway makes the trend unreadable.

Keep the log in a spreadsheet with one row per answer: date, engine, prompt ID, brands named and citations shown. It is usable when any row can be traced back to a prompt and an engine.

Be clear about what the number means. Share of voice is relative presence in the answers you monitor. The LLM Pulse glossary states that it does not establish market share or predict purchases.

Step 1: How do you fix your prompts and competitor set?

Pick the prompts. Choose the questions a buyer would ask an AI tool about your category, such as which providers to shortlist for a specific need. Keep the wording identical in every run. A changed word can move the numbers for reasons unrelated to your brand.

How to know it is right: every log row shows the same prompt text for the same prompt ID. Typical mistake: rewriting prompts each week to test new ideas, then comparing results with the old ones.

Sources disagree on how many prompts you need. LLM Pulse says 50 is the minimum for a defensible measurement and 100 to 200 is the sweet spot. Patrick Stox puts the minimum at 100 to 200 and serious tracking at 250 to 500. Pick a count you can afford to repeat every cycle, and state it in the report.

Pick the competitor set. Share of voice compares you against a named set of rivals. Grow by Data separates this from mention rate, which divides only by AI responses.

Name the rivals that compete for the same buyer, and keep the list short and fixed. Write your own rule, such as "companies that sell to the same buyer at a similar price," and apply it every time. A typical mistake is mixing direct rivals with very large brands that appear in almost every answer, which lowers your share from the start.

Step 2: Why run each prompt more than once?

LLM answers vary between runs. LLM Pulse recommends repeating each prompt and averaging the results rather than trusting one run.

How to know it is right: the report shows the number of runs and the spread between them. Typical mistake: running each prompt once a week and reading every change as a trend.

Sample size sets how many runs you need. AI Growth Agent estimates roughly 100 runs per engine per prompt for a ±10-point margin at 95% confidence, for a brand that appears in about 40% of answers.

That adds up fast. Set your budget first, then choose the prompt and run counts that fit it, and state the trade-off.

Step 3: Which metric should you use?

Most reports use one of three metrics. Grow by Data defines them by what they divide by:

Metric	Divides by	Compares against a named competitor set
Mention rate	AI responses	No
Mention share	Brand mentions in the competitor set	Yes
Citation share	Citations	The source does not say

Only mention share compares you against a named competitor set, so put the metric name in the report title.

Count a brand once per answer that names it, however often the answer repeats the name. AI SEO Tracker states this rule and says Ahrefs uses it too.

A worked example: LLM Pulse shows that 38 brand mentions against a 120-mention comparison set give a share of voice of 31.7%. The formula is your mentions divided by the total, times 100.

Mention rate alone can mislead because it ignores rivals. Grow by Data puts it this way:

"You can rise to 40% while a rival rises to 70%." — Grow by Data

Your rate rises and your share can still fall. The metric also changes the ranking: in one Digital Applied example, the same brand sits third on mention-based share and first on another measure.

How to know it is right: every report states the metric and its divisor. Typical mistake: comparing one month's mention rate with another month's mention share.

Step 4: Should you split results by engine?

Yes. Do not blend all engines into one number.

Track each engine separately and read the blended figure only as a summary. You can start with fewer engines, but name the list in the report. Citable measures whether ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI cite or recommend a site, so each platform can sit on its own line.

How to know it is right: each engine has its own row and its own margin. Typical mistake: picking the engine that is easiest to query and treating it as the whole market.

Step 5: How do you read the margin before acting?

Small samples are imprecise.

Single prompts are unstable too. A preprint on arXiv, not peer-reviewed, reports that share of AI voice on a single tracking prompt has a confidence interval wider than typical week-over-week movement.

Set a rule: if a change is smaller than the margin, treat it as noise and add prompts or runs before you act. Typical mistake: rewriting content after a weekly dip that falls inside the margin.

Step 6: What should the report include?

Put the method next to the figure. Foglift recommends publishing the prompt count, engine count, repeat-run count, observed variance and competitor set. Add the metric name. Without these, nobody can repeat your number, including you next quarter.

Say what the figure does not show: relative presence in monitored answers, not sales or market share.

Then measure again after you change a page. For background, see how to check if ChatGPT recommends your brand and how to measure your brand's visibility in ChatGPT.

Next steps

• List 50 or more buyer prompts, give each an ID and freeze the wording.
• Write the competitor-set rule and name the rivals in a spreadsheet.
• Run each prompt several times per engine and log every answer with date, engine and brands named.
• Calculate mention share per engine and write down the margin next to each figure.
• If you want this measured weekly across ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI, Citable does it as described in how it works, starting at $24/month.

Frequently asked questions

How many prompts do I need to track for AI share of voice?

Sources disagree. LLM Pulse says 50 is a defensible minimum, while Patrick Stox says 100 to 200. Choose a number you can repeat every cycle and publish it.

How often should I run AI visibility prompts to get a stable number?

Repeat each prompt several times per cycle and average the results, as LLM Pulse recommends. For a ±10-point margin at 95% confidence, AI Growth Agent estimates roughly 100 runs per engine per prompt.

What is the difference between mention share and citation share in AI search?

Grow by Data says mention share divides by brand mentions in a named competitor set, and citation share divides by citations. Only mention share compares you against named competitors.

Does a higher AI share of voice lead to more traffic or sales?

The metric does not establish market share or predict purchases, according to the LLM Pulse glossary. Check traffic separately, for example with GA4 tracking for AI assistants.

Sources

• AI SEO Tracker: How to Measure Share of Voice in AI Search
• LLM Pulse: Share of Voice glossary
• LLM Pulse: How to Measure AI Share of Voice
• LLM Pulse: Share of Voice in AI Search
• Patrick Stox: AI Share of Voice
• Grow by Data: Share of Voice in AI Search
• AI Growth Agent: How To Measure AI Share Of Voice
• Digital Applied: AI Share of Voice framework
• arXiv preprint: Paraphrase Brittleness in Production Retrieval-Augmented Commercial Recommendation
• Foglift: AI Share of Voice Tools