On this page
You know how your site ranks in Google. You probably have no idea whether ChatGPT names your product when a buyer asks for the best tool in your category. AI brand visibility is the measure of that gap, and unlike rankings it has no built-in dashboard from the engines themselves, so you have to build the measurement yourself. This post defines the term, lists the metrics worth tracking, and shows a setup that a two-person team can run.
What AI brand visibility means
Searchable defines AI brand visibility as how often, how accurately, and how favorably AI answer engines name a brand. The Rank Masters gives nearly the same definition for AI search visibility: how often, how prominently, and how accurately a brand appears in AI-generated answers. Two independent sources agree, so the definition is safe to use.
The wording has three parts, and each one can fail on its own:
- How often. Out of the questions your buyers ask, in what share does the answer include your brand at all?
- How accurately. When it does, are the price, features, and use case correct, or is the model describing a product you retired two years ago?
- How favorably. Is the mention a recommendation, a neutral list entry, or a warning?
A brand can score well on the first and badly on the other two. Being named often with a wrong description is worse than a small, accurate presence, because the buyer arrives with the wrong expectations.
If you want the broader concept first, What Is AI Visibility? covers it. This post stays on the brand side: what to count and how.
The signals to measure
Adobe’s materials describe brand visibility through three signals: mentions, citations, and recommendations. Adobe also points out that a brand mention can happen without any link to the brand’s site. That detail matters, because a click-based view of the world (analytics, referral traffic) will miss every mention that never produced a link.
Across Searchable, Semrush, and The Rank Masters, the core measurement approaches line up on the same short list:
| Signal | What it answers |
|---|---|
| Mentions | In how many answers is the brand named? |
| Citations | Which of your pages do the engines cite as sources? |
| Share of voice | What share of all brand mentions in your category is yours? |
| Sentiment | Is the tone positive, neutral, or negative? |
| Accuracy | Does the answer state true things about the product? |
Semrush defines mentions as the total number of prompts in which a brand is included in AI responses, and describes its AI Visibility metric as a 0 to 100 benchmark score showing how often a brand appears compared with competitors (source: Semrush knowledge base). That is one vendor’s definition, not an industry standard, so treat the 0 to 100 scale as a convention of that tool.
The Rank Masters extends the list with distribution by LLM, prompt coverage, recommendation rank, AI Overview presence, and AI referral conversion. These are useful extras, but they come from a single publisher, so read them as one proposed KPI set rather than agreed practice.
Why one run tells you almost nothing
Language models do not return the same answer twice. Ask ChatGPT for the best tool in your category on Monday and again on Tuesday, and the list can change order, drop an entry, or add one. If you check once and see your brand, you cannot say whether you are reliably there or got lucky.
Searchable recommends measuring by repeating the same prompts across multiple AI engines. That advice has two dimensions:
- Repeat the prompt. Run each question several times and count the share of runs that name you. This is multi-sampling, and it turns a coin flip into a rate.
- Repeat across engines. An answer from one engine says little about another. Your brand can be common in Perplexity answers because it cites fresh web pages, and rare in a model that leans on older training data.
A mention rate of “6 out of 10 runs on ChatGPT, 1 out of 10 on Gemini” is a finding you can act on. “I saw us once” is an anecdote.
Choosing the prompts you measure
The measurement is only as good as the questions. Your list should look like what a real buyer types, not a list of your own product names. A workable split for a SaaS product:
- Category questions: “best invoicing software for freelancers”, “tools for tracking customer churn”
- Comparison questions: “X vs Y”, “alternatives to a named competitor”
- Problem questions: “how do I stop losing leads after a demo”
- Brand questions: “what is your product”, “is your product good for small teams”
Brand questions measure accuracy and sentiment, since the model is asked about you directly. The other three measure whether you appear unprompted, which is the harder and more valuable test.
How many do you need? A small set is fine to start and a bigger one reduces topic bias. The sizing trade-offs are in How Many Prompts Should You Track?. For a first baseline, 10 to 25 prompts that match real purchase intent beat 200 generic ones.
Competitors are the denominator
A mention rate of 40% means nothing without context. If the category leader sits at 90% and the next three tools at 60%, you are behind. If everyone sits at 20%, you are ahead. That is why the definitions from Adobe and Semrush both measure visibility against competitors, not in isolation. Adobe says its Brand Visibility product benchmarks against up to five direct competitors, a limit set by that product and not a rule of the metric.
Pick three to five competitors you actually lose deals to and track them with the same prompts. Then compute share of voice: your mentions divided by all tracked-brand mentions across the same runs. It is the number that tells you whether your position is improving relative to the market, even when the models themselves are changing underneath you.
A measurement setup you can run this week
You can do the first pass in a spreadsheet and move to a tool when the manual work gets annoying.
- Write 15 prompts across the four types above. Freeze the wording. Changing a prompt resets your history.
- List 3 to 5 competitors and the aliases for your own brand (product name, domain, common misspellings).
- Run each prompt several times on at least two engines. Record whether you are named, in what position, with what tone, and whether the facts are right.
- Compute three numbers: mention rate per engine, share of voice against your competitor list, and the count of inaccurate statements.
- Repeat on a fixed schedule and read the trend, not any single day.
Step 3 is where hand-measurement breaks down. Fifteen prompts, two engines, five samples each is 150 answers to read per round. A tracker automates it. AskAiRank runs your prompts on a schedule, detects brand mentions, and shows a visibility score with competitor comparison. It can poll 8 engines: ChatGPT, Claude, Perplexity, Gemini, Grok, Llama, Mistral, and DeepSeek. The Free plan covers 1 brand, 10 prompts, and 2 engines at once. Pro at $39/mo raises that to 5 brands, 75 prompts, and 8 engines.
If you only care about ChatGPT first, How to Track Your Brand in ChatGPT walks through that single-engine case.
Reading the results without fooling yourself
A few traps come up repeatedly:
- Comparing across tools. Each tool picks its own prompts, engines, and weights, so a score of 40 in one product and 40 in another are not the same quantity. Compare within one setup over time.
- Chasing daily movement. Small swings are sampling noise. Act on multi-week trends.
- Ignoring engines. An average across engines hides a zero. If one engine never names you, that is a separate problem to investigate.
- Counting mentions as success. A mention in a “tools to avoid” answer is a mention. Pair the count with sentiment.
Next step
Open a document and write down 15 buyer questions for your category, plus the four competitors you most often meet in sales conversations. Run each question by hand in two engines, three times. You will have a rough baseline in about an hour, and you will know which part of the definition (how often, how accurately, how favorably) is your weakest. Then decide whether to keep doing it by hand or put it on a schedule.
Frequently asked questions
In practice the terms overlap. AI brand visibility puts the emphasis on your brand as an entity: whether the models name it, describe it correctly, and place it next to the right competitors. AI visibility is the broader umbrella that can also cover your pages and citations. For a founder, the measurement setup is the same.
Daily polling is enough for most teams, because answers shift slowly and a single run is noisy anyway. Read the trend weekly, not the daily number. If you only check once a quarter you will miss the moment a model update or a competitor launch changes your position.
You can run a spot check by pasting 10 buyer questions into each engine and noting who gets named. It works for a first baseline. It does not work as a metric, because answers vary between runs and you cannot compare last month to this month without repeating the exact same prompts on a schedule.
Yes. A model can recommend your product by name without linking to it, and the buyer may search for you directly afterwards. Mentions and citations are separate signals, so track both. A mention shows the model knows you, a citation shows which of your pages it trusts.
Start with ChatGPT and Perplexity, since buyers tend to ask them for tool recommendations, then add Claude and Gemini. The free plan of AskAiRank lets you enable 2 engines at once, which is enough to get a first baseline before you widen the set.
