How to measure your AI share of voice

AI share of voice is the percentage of AI answers that name your brand, out of all the answers where you or a competitor could have been named, for a defined set of buyer prompts. It's the AI-era version of the share-of-voice metric marketers already use — but measured inside ChatGPT, Perplexity, Gemini, and Google AI Overviews instead of ad impressions or search rankings. It answers one question: when buyers ask the engines about your category, how often is it you they hear about?

Measuring it takes a deliberate method, because none of this shows up in your analytics and the same prompt returns a different answer on different days. Here's how to sample prompts, what to record, how to turn it into comparable numbers, and how often to re-run it.

What is AI share of voice?

AI share of voice is your brand's proportion of AI-answer mentions within a category, relative to the competitors that appear alongside you. If you sample 100 buyer prompts across the engines and your brand is named in 30 answers while your top rival is named in 45, your rival has the larger share of voice — and the gap is the thing you work to close.

The metric only means something against a fixed, representative set of prompts and a defined competitive set. Change the prompts every cycle and you're measuring noise; keep them stable and you're measuring movement. Share of voice is the headline number, but it sits on top of finer-grained metrics — mention rate, citation rate, and position — that tell you why the share is what it is.

How do you sample prompts to measure it?

Start from real buyer intent, not your brand name. Nobody discovers you by typing "Acme" into ChatGPT — they describe a need, and the engine names options. So build a prompt set from the situations that happen before a buyer knows you exist: category and location ("best [category] in [city]"), comparisons ("[competitor] vs [alternative]"), alternatives ("alternatives to [big competitor]"), and jobs to be done ("how do I [the problem you solve]").

Aim for 20–50 prompts that cover your real buying situations, and freeze them as your test set. Because answers are non-deterministic, run each prompt more than once per engine — three runs is a sane minimum — and treat "named 2 of 3 times" as a real, recordable signal rather than a yes/no. Use fresh sessions so prior chat history doesn't bias the result. This sampling discipline is the same one behind a full AI-visibility audit; share of voice is what you compute from its output.

Mention rate vs citation rate vs position: what's the difference?

These three metrics answer different questions, and conflating them hides where you're actually losing. Keep them separate:

• Mention rate — the share of runs where your brand is named in the answer at all. This is the broadest measure of presence.
• Citation rate — the share of runs where the engine links to you as a source, not just names you. A link is a stronger signal and often drives a real visit.
• Position — where you appear when named: first in the list, buried third, or mentioned only in passing. Being named first carries far more weight than being named last.

A brand can have a high mention rate but a low citation rate (everyone talks about you, nobody links you), or a decent mention rate but weak position (always named, never first). Tracking all three tells you whether the fix is more corroboration, better answer-shaped content, or stronger sources — the levers behind how ChatGPT recommends brands.

How do you benchmark against competitors?

Share of voice is inherently comparative, so run the same prompt set for your competitors at the same time. For every prompt and engine, record which rivals are named alongside you, how often, and in what position. Over a full cycle this produces a leaderboard: who owns your category's answers, by how much, and on which specific prompts.

The gaps are the gold. When a competitor is named on prompts where you're absent, look at the sources the engine cited to build that answer — those are usually the pages that mention them and not you. That turns a vague "we're behind" into a concrete list: the roundups to pitch, the review platforms to claim, the threads to show up in. Benchmarking also keeps you honest about progress, because a rising tide can lift everyone; what matters is your share relative to the set, not your raw mention count in isolation.

How often should you re-measure?

Monthly is the sensible default for most businesses. GEO changes don't propagate overnight — corroboration takes time to spread across the web and into the models — so measuring too often just samples day-to-day randomness, while measuring too rarely lets problems fester. A monthly cadence, using the same prompts and engines each time, lets you separate "the fix worked" from "the shortlist reshuffled this week."

Re-measure sooner around specific events: after you earn a batch of new third-party mentions, after a competitor launches, or after a major model update that could reshuffle how engines retrieve and rank sources. The one rule that makes any cadence work is consistency — same prompts, same engines, same number of runs — so the trend line reflects your work, not your method. Doing that by hand across four engines and dozens of prompts gets heavy fast, which is exactly the loop PromptHawk automates; you can see the first slice of it with the free AI-visibility check or read the full methodology.

Frequently asked questions

What is AI share of voice?

It's the percentage of AI answers that name your brand, out of all answers where you or a competitor could appear, for a fixed set of buyer prompts across engines like ChatGPT, Perplexity, Gemini, and Google AI Overviews. It measures how often the engines recommend you versus your rivals.

What's the difference between mention rate and citation rate?

Mention rate is how often you're named in an answer; citation rate is how often the engine actually links to you as a source. Citation is the stronger signal because it often drives a real visit, but a name-drop still shapes the buyer's consideration set.

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

Aim for 20–50 prompts that reflect real buyer intent — category, comparison, alternative, and job-to-be-done queries — and run each several times per engine because answers vary. Freeze the set so every cycle measures the same thing.

How often should I measure AI share of voice?

Monthly for most businesses, using the same prompts and engines each time. Measure sooner after earning new third-party mentions, a competitor launch, or a major model update. Consistency of method matters more than frequency.

Can I measure AI share of voice manually?

Yes, for a small prompt set: run your prompts across each engine in fresh sessions, record mentions, citations, position, and competitors in a sheet, and repeat monthly. It gets heavy across many prompts, engines, and runs — which is the work tools like PromptHawk automate — but the method is the same by hand or automated.