How to track competitor share of voice in AI answers
Share of voice in AI answers tells you how often each brand gets named across the questions that matter. How to define it, set it up so the trend means something, and turn it into a to-do list.
"Are we showing up in ChatGPT?" is a yes-or-no question, and it isn't very useful. The better question is comparative. Across all the prompts your buyers ask, how often does each brand in your category get named, and is that moving? That's share of voice for AI answers.
It's also easy to measure badly. Change the prompt set halfway through, mix engines together, or count a brand twice under two names, and the chart will tell you a story that isn't true. Here's how we set it up.
Decide what you're counting
There are a few reasonable definitions. Pick one as your headline number and keep the others as supporting views.
Mention share. Of all brand mentions across your prompt set, what fraction are yours? Count every time any brand in your category is named, then divide your own mentions by that total. It's the closest thing to the classic share-of-voice metric.
Prompt coverage. In what fraction of prompts are you named at all? This is often the easiest number for leadership to grasp.
Citation share. Of all sources cited across the set, what fraction point to your domain? This tracks the part you influence most directly with your own pages.
Position-weighted share. Mentions near the top of an answer count for more than a passing mention at the end. Useful, but harder to explain, so we keep it as a secondary view.
Write the definition down. When someone asks in six months why the number changed, you'll want to be sure it's the market moving and not the method.
Step 1: fix the competitor list
List the competitors you expect. Then run your prompts and add the ones the engines actually name. These surprise brands are often the most useful finding. They tend to be smaller players that happen to be well covered on the sources engines read, and they show you where your category's real competition is.
Handle aliases. A brand might appear as its full name, a short name, its product name, or a common misspelling. Map them all to one entity, or your share will be split and understated.
Step 2: freeze the prompt set
Share of voice only means something against a stable denominator. Choose a prompt set that covers your main personas, funnel stages, and markets, and keep it fixed for the reporting period. When you need to add prompts, add them as a new segment rather than mixing them into the trend line.
Step 3: run each prompt more than once
AI answers vary between runs. A single run per prompt can make a brand look present or absent by chance. Run on a schedule and look at rates over time, not single snapshots. Small prompt sets are especially noisy, so treat week-to-week wiggles with suspicion and look at the direction over a month or a quarter.
Step 4: segment before you conclude
A single blended number hides most of what's useful. Split it by:
- Engine. You may lead in Perplexity and barely appear in Gemini. Those need different fixes.
- Funnel stage. Strong share on decision prompts and weak share on problem prompts means people who already know you hear good things, but new buyers don't meet you.
- Persona. Engines can recommend different brands to a CFO and an operations lead asking about the same need.
- Market. Country and city answers can differ a lot for anything with a local angle.
Most of our action items come from segments, not the headline.
Step 5: find the reason behind every gap
Share of voice tells you who's winning. It doesn't tell you why. For each segment where a competitor beats you, look at three things.
The cited sources. Which pages and domains did the answers lean on when they named the competitor? Their own comparison page, a publisher roundup, a Reddit thread?
The description. How is each brand described? If a competitor is consistently given a clear reason ("best for agencies"), and you're named with a vague one, that's a positioning problem on your own pages and in the sources.
The trend. Did the competitor gain recently? If so, look for what changed around that date. A new page, a new review, a new listicle.
Step 6: turn gaps into a list
Each gap should end up as a specific task tied to evidence: write the comparison page that's missing, pitch the roundup that names three competitors and not you, clarify the use case on your product page, answer the thread. Rank the list by how many prompts each task touches.
The tooling we use
We track this for clients in Promptwatch. Its competitive benchmarking and share-of-voice reporting runs across the full tracked prompt set and breaks out by engine, so the segmentation in step 4 doesn't mean rebuilding spreadsheets each month. Personas, topics, and tags handle the other splits, and country, state, or city tracking covers the market view (state and city tracking sits on the Business plan and the agency plans). Prompt trends show when a competitor's share moved and what changed in the answers between checks.
For step 5, the same workspace has citation analytics, so the sources behind a competitor's mentions are one click away, and sentiment analysis shows how each brand is being described. That's where the number turns into something you can act on.
Report share of voice monthly, keep the method stable, and always pair the chart with the two or three reasons behind it. A number with no explanation gets ignored. A number with a clear cause gets a budget.