# How to run an AI visibility audit: prompts, models, runs and scoring > A practical guide to running an AI visibility audit: which buyer prompts to ask, which models to query, how many runs, how to score the answers, and what a credible audit does not claim. Canonical page: https://askedthrice.com/blog/what-is-an-ai-visibility-audit Published: 2026-06-24 Updated: 2026-08-25 An AI visibility audit measures whether assistants like ChatGPT, Gemini, and Perplexity recommend your brand when buyers ask about your category. The short definition, and what an audit should and should not contain, is on the learn page: [what an AI visibility audit is](/learn/ai-visibility-audit). This post is the practical side: how to run one yourself, step by step, so the result is comparable across runs and against competitors. If your buyers are asking AI assistants "what's the best tool for X" and your brand never comes up, an audit is how you find out, and where the gap is. ## What an AI visibility audit measures The audit answers one question: when someone asks an AI assistant about your category, does your brand get named? To do that, it looks at three things. **1. Presence.** Does your brand appear at all in the answer, or is it absent while competitors get recommended? This is the headline signal. **2. Consistency.** LLM outputs are non-deterministic, so a single answer is noise. A brand that shows up in most runs is a far stronger signal than one that appears once by chance. **3. Competitive gap.** Which competitors get named where you don't, and which external sources those recommendations lean on. This is where the audit turns into an action list. ## Why AI visibility is different from search rankings In traditional search you can rank fourth on Google and still get clicks. In an AI answer there is no page two. The assistant names two or three tools and you are either in that set or you are invisible. That changes what you optimize for. Search rewards ranking. AI answers reward being recommendable: having enough independent proof, comparison content, and clear positioning that a model is confident naming you. What that re-weighting does to an existing SEO playbook is in [GEO vs SEO for B2B SaaS](/blog/geo-vs-seo-for-b2b-saas). ## How an AI visibility audit works A rigorous audit follows the same steps every time so results are comparable across runs and across competitors. 1. **Define buyer-intent prompts.** These are the real questions buyers ask, like "best [category] for [use case]" or "alternatives to [competitor]." They [fan out into four families](/blog/query-fan-out-buyer-questions-ai-visibility), and only the open ones count as a recommendation. 2. **Query multiple models.** Different assistants [pick sources differently](/blog/chatgpt-vs-perplexity-vs-google-ai-overviews), so testing only one hides half the picture. 3. **Run each prompt several times.** Repetition separates a durable recommendation from a one-off fluke. 4. **Record and count.** Log which brands appear, how consistently, and which sources are cited, then report counts with their total ("named in 3 of 12 open questions") and a gap list. A count can be repeated and checked; a score out of 100 promises a precision that three models with different opinions do not have. For the exact models we query and how we separate estimated from observed signals, see our [methodology](/methodology). ## What a good audit does not claim Trust is the whole point of an audit, so it should be honest about its limits. A credible audit does not claim to read the live consumer ChatGPT or Gemini apps, which are closed surfaces that change constantly. It does not present a single run as ground truth. And it does not invent metrics it cannot observe. ## Who needs an AI visibility audit Any brand whose buyers research with AI assistants before they buy. That increasingly means most B2B SaaS, but also consumer categories where people ask AI for recommendations. If your category is one people ask AI about, you want to know what those assistants say. ## How often to run one AI models and the web they draw on change constantly, so a single audit is a snapshot, not a permanent verdict. Re-running on a regular cadence shows whether your visibility is improving as you build comparison content and third-party proof. Every few weeks to monthly is enough for most brands, and always after you publish something meant to move the answer. --- *Want to see where your brand stands in AI search today? [Run the free audit at llmaudit.app](https://llmaudit.app). It asks live models open buyer questions about your category and reports how many name your brand, with the total.* --- Machine-readable index: https://askedthrice.com/llms.txt · Full context: https://askedthrice.com/llms-full.txt · All URLs: https://askedthrice.com/sitemap.xml · Developer and agent resources: https://askedthrice.com/developers.md