How to run an AI visibility audit for your brand

An AI visibility audit is a structured check of how AI assistants answer the questions your buyers ask, and whether those answers name, describe, and cite your brand. It covers the main answer engines people use to research purchases: ChatGPT, Gemini, Perplexity, and Claude. For each question, you record whether your brand appears, where it appears in the answer, which competitors appear alongside it or instead of it, and which web pages the answer cites as sources.

A good audit gives you three things. The first is a baseline: a set of numbers you can measure again later, such as the share of answers that name you. The second is a diagnosis that explains why competitors show up where you don't. The third is a short list of changes ranked by how much each is likely to move those numbers.

You can run an audit by hand. Write a list of questions a buyer might ask, such as "What's the best accounting software for a five-person agency?" Then ask each one in every engine, ideally in a logged-out or private session so your own history doesn't shape the answer. Record the results in a spreadsheet, with a row for each question and engine and columns for whether you were named, your position, how you were described, which competitors were named, and which sources were cited. Then ask the whole set again on another day, because the same question often gets a different answer the second time.

That manual route works for a first look, but it breaks down quickly. Four engines and thirty questions make 120 answers per round, and the answers that matter most, the ones that cite sources, take the longest to record. AI visibility tracking tools automate the asking and the counting, and they're good at showing your numbers over time. What they leave to you is the next part of the job: opening the cited pages, working out why a competitor keeps appearing, and deciding what to change.

That next part is where an audit earns its keep. The scoreboard tells you where you stand, but the cited sources tell you what to do. A competitor that appears in most answers is usually being carried by a handful of pages the engines trust, such as a category roundup, a comparison article, or a well-structured page on its own site. Find those pages and you've found your plan.

Kite runs the whole audit, from choosing the questions to building the plan, and then it can do the work the plan calls for.

If you want a quick first look before a full audit, try Kite's free AI visibility checker. It asks ChatGPT, Gemini, Perplexity and Claude the questions your customers ask, and in about 45 seconds it shows whether they recommend your business and who they recommend instead. It's free and needs no account.

For the full audit, ask Kite in plain language:

"Run an AI visibility audit for our site and tell me why competitors show up in ChatGPT when we don't."

Kite is an AI marketer. It researches your business, builds a growth strategy, and ships the work that brings you new customers. By the time you ask, it already knows your products, positioning, customers, and competitors from that research, so it can write the audit questions and choose the competitors to measure without a long brief. You can add questions of your own, and you can name competitors it should include.

To track your own brand over time, Kite ties the audit to your website in your Kite workspace, either a site Kite built or one you imported, and the domain you want measured. You don't need accounts with the AI engines or a separate tracking tool, because Kite asks the engines through its own built-in connections. Each round of questions counts toward your monthly spending cap, and Kite shows what new questions will add before you track them.

Measuring and analyzing run on their own. Kite asks for your approval before it publishes a change to your website or sends outreach to another site.

The questions you choose decide what the audit can tell you, so Kite starts with questions your buyers would actually type, not questions built around your brand name. A question like "Which AI tools help a small team write ad copy?" shows whether an engine recommends you to someone who hasn't heard of you yet. That's where most of the value is, and a branded question can't measure it.

Kite writes dozens of questions from what it knows about your business and sorts them by buying stage:

• Problem. The buyer describes the problem before they know a category of product exists.
• Finding options. The buyer asks what tools or services solve the problem.
• Comparing. The buyer weighs two or more named options against each other.
• Pricing. The buyer asks what things cost and what they get for the money.
• Setting up. The buyer asks how to get started with a product like yours.

It also groups the questions by topic, such as your core category, each product line, and any feature or score you've given a name. Measuring a feature by what it does, rather than by its name, shows whether answers credit your version or a competitor's. Kite adds a smaller set of branded questions, such as "What does your company do?" or "How much does it cost?", and keeps them labeled so you can view your numbers with or without them.

You see the list before the first run. Edit any question, drop the ones that don't fit, or paste in exact wording you want measured. Kite asks your pasted questions exactly as written in every run. Choosing the questions well is a job of its own, and How to choose which questions to track for AI visibility Coming soon covers it in more depth.

Kite asks every question on ChatGPT, Gemini, Perplexity, and Claude, and it keeps asking on a schedule so a single unusual answer doesn't decide your result. You choose whether the questions run daily, weekly, or monthly. Each answer is saved with the engine and model that produced it.

Before it counts anything, Kite needs to recognize your brand in an answer. You give it your brand name, any other names people use for you, and any extra domains you own, along with the same details for each competitor. That way an answer that calls you by a nickname or links to your help center still counts as naming you.

For every answer, Kite records the details an auditor would copy into a spreadsheet, plus a few that are hard to see by hand:

• Whether the answer names your brand, and its position among the brands named.
• Which competitors the answer names.
• The full text of the answer, with each brand mention highlighted.
• Every source the answer cites, with the page address.
• Pages the engine retrieved while answering but chose not to cite.
• The web searches the engine ran to build its answer, on the engines that report them.

The last two are especially useful. The searches an engine runs tell you which search results it leans on for that question, so they show you which searches your pages need to rank for. The pages it retrieved but didn't cite show where you were considered and passed over, which is often a sign that the page answers the question less directly than the one that was cited. You can open any answer in the Kite web app and read it exactly as the engine returned it.

Kite turns the answers into three numbers you can measure again later. Each one is calculated across the questions you track:

• Visibility is the share of answers that name your brand.
• Share of voice is the share of all brand mentions that belong to you, counting you and each competitor once per answer.
• Cited rate is the share of answers that link to your website.

Mentions and citations are tracked separately because they mean different things. An answer can name you without citing your site, which usually means a third-party page put you there. It can also cite your site for a fact without recommending you.

Kite breaks the numbers down by engine, by topic, by buying stage, and by question, so you can see exactly where you drop out. A common pattern is a brand that appears when buyers compare named options but disappears at the problem and finding-options stages, where most buyers start. Kite ranks you against every competitor the engines name, including ones you didn't know you were competing with. Because answers vary, each visibility figure comes with a likely range, and Kite labels a change as up or down only when it is bigger than the normal day-to-day variation.

Kite checks whether the answers describe you the way you want to be described, which matters as much as whether they name you at all. The first time it presents findings, Kite states how it understands your positioning, such as who you serve and what you're best at, and asks you to confirm or correct it. It then compares the answers against that positioning.

The answers to your branded questions show what each engine believes about you: your category, your prices, your customers, and your main features. Kite flags descriptions that are out of date or wrong, such as an old product name or a price you've since changed. It also flags answers that describe your category without naming you, and answers that put you in the wrong category entirely. Each finding points back to the cited pages the engine drew on, because fixing a wrong description usually means fixing the source that taught it.

Kite reads the pages the engines cite and works out which ones are carrying your competitors into answers. The scoreboard shows that a competitor is ahead, and this step shows why.

Kite starts with the most-cited pages across your questions. For each one, it records what kind of page it is, such as a roundup, a comparison, documentation, a review site, or a community thread, and which brands it names. It then lists the citation gaps: sources that are cited in answers naming a competitor but not you. A roundup that's cited for six of your questions, names three of your competitors, and leaves you out is a clear opportunity.

For every competitor ranked at or above you, Kite writes a short evidence summary with names and numbers: the cited pages that feature it, the topics it wins, and how often it's mentioned. When two or more competitors share the same pattern, such as all appearing on the same review site, Kite calls it out, because a shared pattern usually shows what it takes to get into those answers. It also brings in search data, including the keywords you rank for, their monthly search volume, and whether Google shows an AI Overview for them. The strongest opportunities are often searches where you already rank well but the AI answers on the same topic leave you out.

Kite scans your website to check whether AI systems can find and understand your pages, because an engine can't cite a page it can't read. The scan runs the 24 checks a marketing website controls, across five areas:

• Crawl access. Your crawler rules and sitemap let AI crawlers reach the right pages.
• Content. Your important text loads without depending on JavaScript.
• Structured data. Your pages carry markup that describes your company, products, and content.
• AI guidance. Files such as llms.txt point AI systems to what matters on your site.
• Brand presence. Your brand shows up consistently in the places AI systems check.

The scan gives your site a score out of 100 and shows each check's result and the points it earned. Every check that doesn't fully pass comes with a recommended fix, and the fixes are ranked by how many points each would recover.

Kite turns the audit into ten ranked actions, each tied to a gap it found, plus a few alternates in reserve. It ranks them by expected impact first and effort second. An action aimed at a question with real search demand, an AI Overview, and no mention of you ranks above one with no measurable demand.

Each action says which kind of visibility it targets: AI answers, Google search, or both. It also says who does the work:

• Work Kite does on your website. This includes new comparison or category pages, clearer answers on existing pages, structured data, llms.txt, crawler rules, and internal links. Kite builds the change on a site it hosts or publishes it through a supported content system such as Webflow, Contentful, Sanity, or Storyblok.
• Work Kite prepares for you to send. For a roundup or review site that leaves you out, Kite researches the person responsible for the page and drafts the pitch, so you send it from your own account.
• Work only you can do. Partnerships, podcast appearances, and customer reviews need you, so Kite says who to approach, what to say, and which number should move.

Every content action comes as a complete brief: the working headline, the claims the page must make, and the elements it must include, such as a comparison table with named competitors or a block of common questions. When a site problem stops engines from reading a page, Kite ranks the fix above any new content for that page, since content there can't win answers until the fix is live. Tell Kite which actions to start, and it does the work and brings the changes back for your approval.

Kite runs the same questions again after the work ships, so you can see whether it changed the answers. Because the question set stays the same, every round is comparable with the one before it. Kite shows which questions moved, which competitors gained or lost ground, and which pages are newly cited, and it connects each change to the work that shipped. Changes smaller than the normal variation between runs are marked as within noise, so you don't chase movement that isn't real.

When a round shows a new gap, Kite adds it to the plan. When it shows a question the engines keep circling that your list doesn't cover, Kite suggests tracking it.

When the audit is done, you have a measured baseline, an explanation of where you stand, and a plan Kite can start on.

You can read all of this in the Kite web app, download the results as a CSV file, or save the AI visibility report as a PDF to share with your team.