How do I measure my visibility in AI answers (AI share of voice)?
AI share of voice measures how often AI engines like ChatGPT, Perplexity, and Gemini name your brand when people ask buying questions in your category. You measure it by running a fixed set of real prompts across each engine on a schedule, recording whether you appear, where, and against whom, then tracking that percentage over time. There's no single authoritative tool yet, so the method is part prompt panel, part discipline.
The short version
Rankings tell you where you sit on a page of blue links. AI share of voice tells you something blunter: when a buyer asks an AI assistant to recommend someone like you, do you get named, or does a competitor?
That's the whole point now. Your customers are asking ChatGPT for "the best-in-class company," and Perplexity for "who should I hire for what you do," and Google is answering with an AI Overview before anyone scrolls. If your brand isn't in those answers, you're not on page two. You're not in the room. AI share of voice is the number that tells you whether you're in the room and how much of it you own.
It's the AEO equivalent of "what's my ranking." And like rankings, it only matters if you measure it instead of guessing.
What AI share of voice is
Share of voice is an old idea: out of all the attention in your category, what slice is yours. Traditional SOV measured ad impressions or branded search. AI share of voice moves that question into the answer engines.
Concretely, it's the percentage of relevant AI answers that mention your brand, ideally weighted by how prominently you show up. Mentioned first as the top recommendation counts for more than buried in a list of seven. Cited as a linked source counts differently from named in passing. A good measurement separates those, because "we got mentioned" and "we got recommended first" are very different business outcomes. (If you want the formal definition to point a colleague at, we keep one in the AI share of voice glossary entry.)
Three things make it a distinct metric, not just SEO with a new coat of paint:
- The answer is the destination. In classic search, the result links you somewhere. In an AI answer, the model often resolves the whole question in place. Being the cited brand IS the win, because there may be no click to fight over afterward.
- It's probabilistic, not fixed. Ask the same question twice and you can get two different answers. AI share of voice is a rate across many runs, not a single rank you can screenshot once and file away.
- It's cross-engine. ChatGPT, Perplexity, Gemini, and Google AI Overviews each have their own retrieval logic, their own training cutoffs, and their own source preferences. You can own one and be invisible in another. A real number spans all of them.
If you want the discipline underneath the metric, that's answer engine optimization. AI share of voice is how you keep score. (And if you're still untangling AEO from SEO and GEO, here's how the three fit together.)
Why this is the number that matters now
Here's the uncomfortable part for anyone who's been coasting on Google rankings. You can hold position one for your money keyword and still be losing, because a growing slice of your buyers never sees the SERP. They ask the assistant, take the recommendation, and move on. If that recommendation isn't you, your ranking is a trophy from a race fewer people are running. The traffic doesn't vanish all at once either; it leaks, which is exactly why organic traffic can slide while your rankings hold steady.
AI share of voice catches that gap early. It tells you:
- Whether AI engines even know your brand exists as an entity.
- Whether they know what you do well enough to recommend you for the right queries.
- Who they recommend instead, which is the fastest competitive intel you'll ever get. The names that keep showing up are your real competitive set as the model sees it, which is often not the set you'd have written down yourself.
- Whether the AEO work you're investing in is moving the needle, or just feeling productive.
That last one is the point. Without a number, "AI visibility" is a vibe. With one, it's a metric you can move.
How to measure AI share of voice (the actual method)
No tool will hand you a clean dashboard for this the way ahrefs hands you keyword rankings. The category is too young and the engines don't expose the data. So the method is deliberately manual and repeatable. Done right, it's rigorous. Done lazily, it's noise.
1. Build a fixed prompt panel
Pick 15 to 40 prompts that mirror how real buyers ask AI for someone like you. Cover the spread:
- Recommendation prompts: "Who are the best-in-class companies?" / "Recommend a your service provider for a your ideal customer."
- Comparison prompts: "a competitor vs alternatives" / "Best tool or service for your use case."
- Definitional and adjacent prompts where you'd hope to be cited as a source, not just a vendor.
Two rules keep the panel honest. First, write the prompts the way a buyer types them, not the way you wish they'd phrase it: messy, specific, often with a location or a constraint baked in. Second, write them down and freeze them. The panel only works if it stays constant. Change the prompts every time and you're measuring your prompts, not your visibility.
2. Run the panel across every engine that matters
At minimum: ChatGPT, Perplexity, Google AI Overviews, and Gemini. Run each prompt and record the result. Because answers vary run to run, run the important prompts more than once (three runs is a sensible floor for your top recommendation prompts) and treat the appearance rate as the data point, not a single answer.
A few things that quietly distort the picture if you let them:
- Log out, or use a clean session. Personalization and memory will feed you a flattering picture that no first-time buyer ever sees.
- Separate organic mentions from ads. On ChatGPT, sponsored cards now appear alongside or below the answer, labeled and distinct from the organic response. Those are paid placement, not share of voice; count them separately or you'll flatter yourself. Perplexity exited advertising in early 2026, so a mention there is purely organic citation. (More on what is and isn't possible on Perplexity here.)
- Note whether retrieval was live. Some answers pull from the live web, some from training data. When you can tell, write it down, because it explains a lot of the drift you'll see later.
3. Score each result, don't just check a box
For every prompt, capture more than yes/no:
- Mentioned at all? (yes / no)
- Position: named first, in the list, or an afterthought.
- Form: recommended as a brand, cited as a linked source, or just referenced.
- Who else showed up: your real competitive set, as the AI sees it.
- Source, if visible: which page or third-party mention the engine pointed to. This is gold, because it tells you which of your assets are doing the work and which third-party sites the engine trusts in your space.
4. Roll it into a share-of-voice percentage
The simplest version: out of all your panel runs, what percentage named you. The better version weights by prominence so a top recommendation outscores a footnote mention. A workable weighting most people can live with: named first counts full, named in the body counts half, referenced in passing counts a quarter. Apply it consistently and the exact weights matter less than the fact that you never change them. Either way, you now have one number per engine and one blended number overall. That's your baseline.
5. Track it on a cadence and watch the trend
Re-run the same panel monthly, or every two weeks if you're actively pushing AEO. The single reading is interesting. The trend is the answer. Pair it with Google Search Console for classic search so you're seeing both maps at once, because the ground under search split into Google and AI and you need to watch both. If you want a faster, lighter starting point before you stand up the full panel, our AI visibility checker gives you a quick read on where you currently land.
The honest limitations (because someone needs to say it)
Anyone selling you a tidy, fully automated AI share of voice score is selling further than the data goes. The real constraints:
- There's no authoritative source of truth. A wave of monitoring tools has appeared, and some are genuinely useful for scale. But they sample, they approximate, and they disagree with each other. Treat any single tool's number as a directional signal, not gospel.
- Answers drift. Model updates, retraining, and live web retrieval mean your number can move for reasons that have nothing to do with anything you did. That's why the method, not the snapshot, is what you trust. A small wobble between two readings is usually noise; a sustained slide across several is signal.
- It's a rate, not a ranking. You will never get a clean "you are number 3." You get a probability of appearing. Make peace with that and the metric becomes useful instead of frustrating.
- No one can guarantee a placement. Not us, not anyone. You influence the inputs (entity clarity, citable content, third-party corroboration) and the inputs move the rate. Anybody promising a guaranteed citation is bluffing.
None of that makes the metric worthless. It makes it real. A measured estimate you re-run consistently beats a confident guess every single time.
What to do once you have the number
Measuring is step one. The point is to move it. If your AI share of voice is low or flat, the levers are the AEO fundamentals, and they tend to fall in this order of impact:
- Be a clear entity. Make your brand something the engines can resolve without guessing: consistent name, category, and location across the web, reinforced with structured data. If the model isn't sure who you are, it won't risk recommending you. This is usually where a low score comes from.
- Publish content built to be extracted. Direct answers, clean structure, the question answered before the pitch. Pages written to be quoted get quoted.
- Get corroborated by sources the engines already trust. Third-party mentions, reviews, and citations are what move you from "exists" to "recommended." You rarely cite yourself into an answer.
- Keep cornerstone pages fresh. Stale entities fade out of live retrieval.
If a specific engine is the problem (ChatGPT names a competitor and ignores you, or you're missing from AI Overviews), the fix gets more targeted. We get into the engine-by-engine playbook in why your brand isn't cited by ChatGPT and how to get cited in Google AI Overviews.
That's the work. The metric just tells you whether the work is landing. Want the full method behind moving it, start with what AEO is and how the AEO/GEO service runs it for you, with AI share-of-voice tracking built into the reporting so you're never guessing.
Find out where your brand stands in AI search
Most companies have no idea whether ChatGPT recommends them or their competitor. We'll tell you. We run the panel, hand you the number, show you who's beating you in the answers, and lay out exactly what it takes to close the gap. No guaranteed-placement fairy tales, just a real baseline and a real plan.
Want to see your AI share of voice before you commit to anything? Book 30 minutes or email admin@moonsauceagency.com. No quote-form games, no pressure, just the number and what to do with it.
