# What AI visibility is, and how to measure it > AI visibility is how often, and how favorably, an AI assistant names your brand when someone asks a question you should win. It is measured by running the same prompts on a schedule and recording mentions, share of voice, and the sources each answer cites. - Published: 2026-09-05 - Author: Selfwin team - Canonical: https://selfwin.co/blog/what-is-ai-visibility - Tags: ai-visibility, measurement, chatgpt, perplexity, gemini ## Key takeaways - AI visibility is measured per prompt: the same question, asked on a schedule, on each engine you care about. - The three numbers that matter are mention rate, share of voice against named competitors, and which sources the answers cite. - One day of data is a snapshot. A trend needs at least ten prompts and about two weeks of daily runs. - Citations tell you where to act: the pages and communities the engines already trust for your topic. When someone asks ChatGPT which tool to use, the answer names a handful of brands and stops. There is no page two. Either you are in the answer or you are invisible for that question. AI visibility is the measurement of that fact, over time, across the questions that matter to your business. ## How AI visibility is measured The method is simple and borrowed from rank tracking. You write down the prompts your customers ask, you run them on a schedule on each engine, and you record what came back. For every run, three things are worth recording: - **Mention.** Did the answer name your brand at all? Across many runs this becomes a mention rate, the core visibility number. - **Share of voice.** Of all the brands named in the answer, how many mentions were yours versus each competitor you track? This is the number that tells you whether you are winning a category or just present in it. - **Citations.** Which pages did the engine cite or draw from? Perplexity shows sources inline; ChatGPT and Gemini increasingly do too. The cited domains are the engines' trusted list for your topic. Engines also run their own searches behind an answer. Gemini exposes the search queries it issued while grounding; those sub-queries, sometimes called fan-outs, show you the phrasing the engine thinks the question is really about. ## Why one run is not a measurement AI answers are not deterministic. The same prompt asked twice can name different brands, in a different order, citing different pages. That is why a single screenshot of ChatGPT recommending a competitor proves very little, and why a single screenshot of it recommending you proves just as little. Measurement means repetition. Run every prompt daily, keep every answer, and look at the rate. With ten prompts and two weeks of runs you have around a hundred and forty observations per engine, which is enough to see whether a change moved the number. ## What a good prompt set looks like The measurement is only as good as the questions. A prompt set that consists of "best [category] tool" ten different ways will move together and tell you one thing. A useful set covers: - the category head term, the short phrase a buyer types first; - the differentiators, the specific capabilities that make your product the right answer for a subset of buyers; - the audience's job to be done, in the words that audience uses; - a few comparison and alternative questions that name competitors. Mix the forms too. People type keyword fragments, full questions, imperative requests ("recommend a…"), and constraint-laden long-tail asks. Engines answer each form a little differently. ## Reading the numbers Mention rate answers "are we in the conversation." Share of voice answers "are we winning it." Citations answer "where does the engine get its opinion," and that last one is the lever. If every answer to your category question cites the same two review sites and one Reddit thread, that is where the work is, not on your own homepage. Track the numbers per engine. It is normal to be visible on Perplexity, which leans on live web sources, and absent on ChatGPT, which leans more on what it has already learned. The gap tells you what kind of content to produce. ## What to do with it The point of measuring is to act. Every prompt you lose is a diagnosis: the engine needed a page, a comparison, a community answer, or a mention on a trusted source, and did not find yours. Fix that, keep measuring, and the rate moves. That loop, measure, diagnose, draft, approve, measure again, is the whole job, and it is what Selfwin runs for you. ## FAQ ### What is AI visibility? AI visibility is the share of relevant AI-assistant answers that mention or recommend your brand. It is tracked per prompt and per engine, the way search rankings are tracked per keyword. ### How is AI visibility different from SEO rankings? A search ranking is a position on a results page. An AI answer either includes you or it does not, and it may name several brands at once, so visibility is measured as a rate over many runs rather than a position. ### Which engines should a brand track? Start with the assistants your customers actually use: ChatGPT, Gemini, and Perplexity cover most consumer and prosumer traffic today. Add others only once you can act on the data. ### How many prompts do you need before the numbers mean anything? Ten active prompts run daily gives a readable trend in about two weeks. Fewer prompts or fewer days mostly show noise, because answers vary run to run.