The short version
- Measure AI visibility by sampling: run a fixed set of real buyer prompts through each engine on a schedule and score every answer.
- Track mention rate, share of voice, average position, citation share and net sentiment, ideally rolled into one headline score.
- Results differ by engine and vary between runs, so break them out per engine and judge trends, not single answers.
- Report the trend, the prompts you won and lost, the sources that matter and what you shipped, not screenshots.
To measure AI visibility, run a fixed set of real buyer questions through each answer engine on a schedule, store every answer with its citations, and score how often your brand is mentioned, how early it appears, whether you're cited and how you're described. Then track the trend, because single answers vary too much to trust.
Why can't you just ask ChatGPT about your brand?
Because one answer is an anecdote. Ask the same question twice and you may get different brands in a different order. Answers also shift with the model version, whether the engine searched the web, the user's location and personal settings like saved memories.
And “What do you know about Acme?” is the wrong question anyway. Buyers rarely ask about you by name. They ask “What's the best tool for X?” and see who comes up. Measurement has to start from their questions, not yours.
How do you build a prompt set?
A prompt set is the fixed list of questions you track. Everything else rests on it, so build it from evidence, not brainstorming:
- Sales calls and demos. The questions prospects ask before they buy.
- Support tickets and onboarding. The problems people are trying to solve with products like yours.
- Search data. Queries you rank for, rewritten as the full questions people now type into assistants.
- Community threads. How people describe the problem in forums and social posts, in their own words.
- Competitor comparisons. “X vs Y” and “alternatives to Y” for each competitor that matters.
Then balance the set across intents, because each tells you something different:
| Prompt type | Example | What it tells you |
|---|---|---|
| Category discovery | “What's the best payroll software for a 20-person startup?” | Whether you make the shortlist at all |
| Problem-led | “How do I run payroll for contractors in three countries?” | Whether engines connect your product to the job |
| Comparison | “[Your brand] vs [rival] for small teams” | How you're positioned against a named competitor |
| Alternatives | “Cheaper alternatives to [rival]” | Whether you're the fallback when a rival disappoints |
| Brand and pricing | “Is [your brand] good for international teams?” | Whether engines get your facts right |
A few rules keep the set useful. Weight prompts by importance, since a high-intent comparison matters more than a trivia question. Keep the wording stable so trends mean something; when you must change a prompt, treat it as a new one. And include prompts where you're currently invisible. Tracking only the questions you already win is flattering, not useful.
Finally, group prompts into topics, such as product lines, use cases or markets, so you can report by topic as well as overall. A sharp drop in one topic is easy to miss inside a blended number.
Which AI visibility metrics should you track?
Each answer gets checked for whether your brand appears, where, whether you're cited and how you're described. Across many answers, that rolls up into six metrics. These are the definitions Elite Answers uses:
| Metric | Definition | Why it matters |
|---|---|---|
| Mention rate | The share of answers that mention your brand | The simplest read on whether you're in the conversation |
| Share of voice | Your mentioned answers as a share of all tracked brands' mentioned answers | Puts your visibility in competitive context |
| Average position | The rank of your first mention among tracked brands, when you're mentioned (1 means named first) | Named first and named fifth are not the same result |
| Citation share | The share of cited sources that link to your domain | Shows whether engines use your own pages as evidence |
| Net sentiment | Percent of positive mentions minus percent of negative mentions | Being called “expensive and clunky” is not a win |
| AI Visibility Score | A 0–100 roll-up of position and citations across prompts and engines | One trend line for stakeholders |
The AI Visibility Score is built from a per-answer AnswerScore. When the brand is mentioned, AnswerScore = 100 × (0.7 × position weight + 0.3 × cited). The position weight is higher the earlier you're named, and “cited” is 1 when the answer cites your domain. The score is the importance-weighted average of AnswerScore across the latest answer for every prompt and engine in the last seven days, and answers that don't mention you count as zero. The full math is in our methodology.
Why do results differ so much between engines?
Because they're different products built on different data. Expect your numbers to vary by engine, sometimes a lot:
- Different retrieval. Perplexity crawls the web with its own bot. Google's AI Overviews and AI Mode draw on Google Search. ChatGPT combines its own crawler with third-party search providers. Each sees a different slice of the web.
- Different habits. Some engines search on most questions; others answer more often from model memory. Memory-heavy answers favor brands with a long, broad footprint.
- Different citation styles. Some engines show many sources per answer, others few. Citation share isn't comparable across engines without that context.
- Different update cycles. Model releases and retrieval changes land on each vendor's schedule, so one engine can shift while the others hold steady.
So report per engine before you average. A blended score that hides a collapse on one engine behind a gain on another is worse than no score at all.
Why track AI visibility over time?
Single runs are noisy. The same prompt on the same engine can name you on Monday and skip you on Tuesday with nothing changed on your side. Repeated sampling turns that noise into a signal you can act on.
- Run on a schedule. Weekly is enough for most prompts. Daily makes sense for a small set of priority prompts where you need to see changes fast.
- Store the full answer. Scores tell you something moved. The stored text and citations tell you why.
- Annotate changes. Mark the dates you shipped content, fixed crawler access or earned coverage, and the dates engines released new models. Otherwise you'll credit or blame the wrong thing.
- Compare like with like. Use rolling windows, such as the last seven days, and compare them period over period.
What are the most common measurement mistakes?
- Only branded prompts. Engines usually get your name and basics right. The risk sits in category questions where you're not named.
- Changing prompts every month. You can't trend a moving target. Version the prompt set and add new prompts alongside the old ones.
- Counting mentions without reading them. A mention in a list of tools to avoid still counts toward mention rate. That's what sentiment is for.
- Too few competitors. Share of voice against one rival flatters you. Include the brands engines actually name, even ones you don't think of as competitors.
- Ignoring the sources. Scores say where you stand. The cited sources say how to move.
What should you do with the data?
Measurement only pays off when it changes what you ship. A simple loop works:
- Find losing prompts. High-importance prompts where you're absent or named late.
- Read the answers. Which brands win, what the engine says about them and which sources it cites.
- Diagnose. Is it access (bots blocked), content (no page answers the question), facts (engines describe you wrongly) or authority (the cited sources don't mention you)?
- Fix and re-measure. Ship the fix, note the date and watch the same prompts.
Elite Answers' tracking runs this across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and AI Mode, storing every answer in full with its citations and listing the domains engines cite in answers that leave you out.
How do you report AI visibility to stakeholders?
Executives want three things: are we showing up, is it getting better, and what are we doing about it. A one-page report that answers those beats a forty-tab dashboard:
- Headline trend. One score or share-of-voice line over time, with the competitor set beside it.
- Engine breakdown. Where you're strong, where you're weak, where it moved.
- Wins and losses. The specific prompts you gained or lost, with a quote from the actual answer.
- Sources. The domains engines cite most for your category, and whether they mention you.
- Work and next steps. What shipped, what's next and which metric each item should move.
A trend line you can explain beats a score you can't.
Be careful connecting visibility to revenue. Referral traffic from chatgpt.com, perplexity.ai and similar domains is measurable but undercounts influence, since many people read the answer and never click. Pair it with softer evidence: branded search trends, and an “AI assistant” option in your “How did you hear about us?” field. Agencies running this across many clients can borrow a cadence from AEO for agencies.
Frequently asked questions
How many prompts should I track?
Enough to cover your main products, use cases and buying stages, plus a few prompts per important competitor comparison. For a single brand, a few dozen well-chosen prompts is a solid start, and you can expand once the reporting habit sticks. Forty prompts drawn from real buyer questions will tell you more than four hundred generated from a keyword list.
How often should I measure AI visibility?
Weekly works for most prompts: often enough to catch real shifts, not so often that you chase noise. Run your highest-priority prompts daily if you need to spot drops quickly, for example around a launch or a pricing change. Whatever the cadence, compare rolling windows period over period rather than reacting to a single day's answers.
What's a good AI Visibility Score?
It's relative. Scores depend heavily on your category, your competitors and how demanding your prompt set is. A set full of prompts where you're the incumbent will score higher than one built around categories you're trying to break into. Judge the score against competitors on the same prompts, and against your own trend, rather than an absolute benchmark.
What's the difference between mention rate and share of voice?
Mention rate is about you alone: the share of answers that name your brand. Share of voice is competitive: your mentioned answers as a share of all tracked brands' mentioned answers. Your mention rate can rise while your share of voice falls, if competitors gain mentions faster than you do. Tracking both shows whether you're growing or the whole category is.
Does personalization make AI visibility tracking pointless?
No. Personalization means no tracker sees exactly what every user sees, but a consistent, neutral sample run the same way each time still shows the baseline engines give someone with no history. That baseline moves with your content, coverage and crawler access. Treat the numbers as a well-controlled sample, not a census, and focus on direction.
Can I measure AI visibility manually?
Yes, for a small set. Run each prompt in each engine, paste the answers into a spreadsheet and record mentions, position, citations and tone. It gets heavy quickly: prompts times engines times repeated runs adds up, and inconsistent manual scoring makes trends unreliable. Plenty of teams start by hand to learn the method, then automate once they know which prompts matter.