Measuring Your AI Share of Voice Without Guessing

What is AI share of voice, exactly?

AI share of voice is your slice of the answers. Take a fixed set of buyer questions, run each one through the AI engines your customers actually use, and look at who gets named in the response. Your share of voice is your cited mentions divided by the total cited mentions across that set. If you run 40 questions, your brand shows up in 12 answers, and three competitors split the other 28, your share of voice is 12 of 40 cited slots, or roughly 30%, measured against named rivals.

The term borrows from the old advertising idea of share of voice, your portion of total category spend or impressions. The AI version swaps paid impressions for earned citations. You are no longer buying a share of attention; you are earning a share of the synthesized answer. That is a healthier metric, because it tracks whether the model trusts you enough to name you when a real buyer asks a real question.

It is also the metric most brands quietly skip. They will tell you they "want to show up in ChatGPT" but cannot tell you, this week, how often they do versus the competitor down the street. Measuring share of voice turns a vague ambition into a number you can move.

Why share of voice beats a single citation check

Plenty of teams check whether ChatGPT mentions them once, screenshot it, and call it a win. That is a vanity moment, not a measurement. A single answer is a coin flip: the same prompt can name you on Monday and your competitor on Thursday as sources refresh and the model reweights. One screenshot tells you nothing about whether you are winning the category.

Share of voice fixes this in two ways. First, it is comparative. You are not asking "did I appear?" but "what fraction of the answers went to me versus everyone else?" That reframes the work around winning, not just existing. Second, it is longitudinal. Because you hold the prompt set and the engines constant, week-over-week movement reflects real change in how the models see you, not the question you happened to type. If you have not built the underlying citation tracking yet, start with our walkthrough on how to track ChatGPT mentions weekly, then layer the competitive count on top.

How do I actually measure it, step by step?

The method is simple enough to run by hand and stable enough to automate. Here is the loop we use across more than 10 brands on a single visibility engine.

• Fix a prompt set. Pick 30 to 50 buyer questions that matter, the ones with purchase intent: "best non-toxic all-purpose cleaner for pet owners," "X versus Y for small teams," "is X worth it." Keep this set frozen so you compare like with like.
• Pick your engines. ChatGPT, Perplexity, Gemini and Google AI Mode cover most of the market. Run every prompt through each one.
• Name your competitors. Decide upfront which three to five brands you are competing with. Anything cited outside that set is "other," so your denominator stays clean.
• Log each answer. For every prompt on every engine, record who got cited, in what position, and with what link. A spreadsheet works to start; a scheduled job scales it.
• Compute share. Your cited mentions divided by total cited mentions, per engine and overall. Repeat on the same cadence and chart the trend.

That is the whole mechanic. The discipline is in keeping the set and the schedule constant so the trend means something.

How often should I measure, and what cadence works?

Weekly is the practical floor for an active program, with a monthly rollup for reporting. AI answers move week to week as models refresh and the sources behind them change, so any single check is noisy by nature. A weekly run smooths that noise into a line you can trust, and a monthly summary is what you actually show a stakeholder.

Resist the urge to re-tune the prompt set every week to chase a better number. The instant you change the questions, you have broken the comparison. Add new prompts to a separate, second cohort if you must, but keep the core set frozen for at least a quarter. Expect a meaningful trend to take a few months. This is the same patience curve we describe for an AI visibility audit, where the baseline is the start of the work, not the end of it.

Turning the number into action

A share-of-voice number is only useful if it points to the next move. The most valuable cut is not the headline percentage; it is the breakdown of which prompts you lose and to whom. When a competitor consistently takes the answer for a specific question, that is a content and credibility gap you can close.

This is where share of voice connects to a deeper diagnosis. Pair it with the citation gap audit, which maps the exact questions where rivals get cited and you do not, then traces why: an extractable answer they have and you lack, a third-party source grounding their claim, a consistency problem in your facts. Share of voice tells you the score; the gap audit tells you which possession to run next. Together they turn a dashboard into a roadmap.

What we will not claim about your share of voice

Here is the honest line. No one can guarantee a fixed share of voice, because citation selection is not fully controllable and the models reweight their sources on their own schedule. Anyone selling "guaranteed 40% AI share of voice by Q3" is selling a number they cannot hold. What is real is the method: a frozen prompt set, the same engines, a steady cadence, an honest count, and a trend you watch move over months.

We also will not flatter the number. If your share is 8% and three competitors are eating the category, the measurement should say so plainly. The point of tracking share of voice is not to feel good about a screenshot; it is to know, in a number you can defend, whether the work is moving you up the answer. Measure it honestly, act on the gaps, and let the trend line make the argument.

Italo Campilii

Founder, Acromatico · runs AI visibility & brand systems across 10+ live brands