An AI visibility audit checks how AI assistants describe your company. You run a fixed set of buyer questions through ChatGPT, Gemini and Perplexity, record which companies each answer names and which pages it cites, mark every wrong fact about you, and trace each one back to the page it came from so you can correct it there.

Picture the founder of a 40-person HR software company. A prospect mentions on a call that ChatGPT called the product "a payroll tool for small retailers". It isn't, and hasn't been for two years. The founder tries the same question and gets a different answer. A colleague tries Perplexity and gets a third. Nobody knows whether this happens once a month or every day, or where the old description is coming from. An audit turns that unease into a list.

What is an AI visibility audit?

An AI visibility audit is a repeatable test of what AI assistants say about your company and your market. It measures three things: whether you are mentioned, whether what is said is correct, and which sources the answers draw on.

It differs from a classic SEO audit in what it inspects. An SEO audit looks at your pages: crawlability, rankings, technical health. An AI visibility audit looks at the answers, and then works back to the pages, yours and other people's, that shaped them. Our guide to getting cited by ChatGPT, Perplexity and Google AI Mode covers the page side; this post covers the checking.

Which questions should go into the prompt set?

The prompt set should hold 20 to 40 questions your buyers actually ask, written in their words, and fixed so you can rerun them. A prompt set is the list of questions you ask every assistant in the same form each time, so changes in the answers reflect changes in the assistants, not in your wording.

The five kinds of question

• Category questions: "What are the best HR tools for a 50-person company?" These show whether you are in the shortlist at all.

• Problem questions: "How do I manage shift schedules across three locations?" These show whether you appear before the buyer knows your category.

• Comparison questions: "Company A vs Company B for onboarding." These show how you are positioned against named competitors.

• Brand questions: "What does [your company] do?", "How much does [your company] cost?", "Who founded [your company]?" These are where wrong facts are most visible.

• Trust questions: "Is [your company] GDPR compliant?", "Does [your company] integrate with Slack?" These are the questions a buyer asks just before a call.

Take the questions from sales calls, support tickets and search queries, not from a brainstorm. If sales hears a question every week, it belongs in the set.

How do you run the audit across ChatGPT, Gemini and Perplexity?

Run every question in every assistant under the same conditions, record the full answer with the date, and log names, sources and errors in one sheet. Consistency matters more than volume, because answers vary from one run to the next.

1. Fix the conditions. Use a fresh session without personal memory or custom instructions, the same country, and the same mode (for example, with web search switched on where the assistant offers it). Write the conditions down.

2. Run each question more than once. Answers change between runs. Three runs per question per assistant gives you a pattern rather than an anecdote.

3. Save the answer, not a summary. Copy the full text and the date. You will need it when you check whether a fix worked.

4. Log every company named. Record each company mentioned and its position in the answer.

5. Log every source. Where an answer shows sources or links, record each URL. These are the pages you will work on later.

6. Mark each claim about you. Label it correct, wrong or out of date, and note the correct fact next to it.

Before you start, check that each assistant can reach your site at all. OpenAI's crawler documentation says OAI-SearchBot is used to surface websites in ChatGPT search, and Perplexity's bot documentation says the same of PerplexityBot. If robots.txt or your firewall blocks them, your own pages can't be among the sources.

How do you measure share of voice in AI answers?

Share of voice in AI answers is the share of all company mentions across your prompt set that are mentions of you. If 30 answers name 120 companies in total and 18 of those mentions are yours, your share of voice is 15%.

Track it alongside these measures, because a mention on its own can hurt you:

• Mention rate: the share of answers that name you at all.

• Accuracy rate: the share of claims about you that are correct.

• Citation rate: the share of answers that cite at least one page you control.

• Share of voice by question type: strong on brand questions and absent from category questions means buyers find you only once they already know your name.

For Google's AI features you also have first-party data. Google's documentation on AI features and your website says sites appearing in AI Overviews and AI Mode "are included in the overall search traffic in Search Console" and are "reported on in the Performance report, within the 'Web' search type." In June 2026 Google introduced Search Generative AI performance reports in Search Console, which show impressions in AI features by page, country, device and date; Google says that as of 31 August 2026 they are rolled out "to all websites worldwide."

How do you correct wrong facts at the source?

You correct a wrong AI answer by correcting the pages it draws on. Assistants don't take edits to their answers directly, so the work is to fix your own site first, then the profiles and third-party pages the answers cite.

Start with what you control

• Your own site. State each key fact once, plainly, on a page that is easy to find: what you do, who it is for, pricing model, locations, founding year, integrations. Retire or redirect old pages that still describe the product you used to sell.

• Organization structured data. Google's Organization structured data guide says the markup "can help Google better understand your organization's administrative details and disambiguate your organization in search results", and recommends placing it on your home page or a single page such as "about us". Its sameAs property lists your other official profiles.

• Google Business Profile. Google's Edit your Business Profile page says you can update "your address, hours, contact info, and photos", and that "we review your changes before we update them live on your profile."

Then the sources you don't control

• Knowledge panels. Google says knowledge panel information "comes from various sources across the web." Its feedback page notes that "anyone can submit feedback", that verified representatives get their feedback prioritised, and that descriptions "can't be edited"; instead, "you can contact the source of the information and ask them to correct it."

• Directories, review sites and articles. Work through the source URLs in your log, most-cited first. Ask each owner to update the outdated fact, and give them the page on your site that states it correctly.

Then rerun the affected questions after the pages have had time to be recrawled, and compare against the answers you saved.

How often should you repeat the audit?

Run the full prompt set monthly, and rerun the affected questions after any fix, launch, rebrand or price change. Monthly is frequent enough to see a trend and to notice when an assistant changes its sources.

Keep the questions fixed between runs. Add new ones in a separate block, so the core set stays comparable over time. If your store depends on product questions, our post on getting a store recommended by AI assistants covers the catalogue side.

How we approach it at Vizio AI

We start with the questions. With the client's sales and support teams we collect what buyers ask before they buy, group the questions by type, and fix the set. Then we run it across the assistants the client's buyers use, under written conditions, and log every company named, every source cited and every claim about the client.

From the log we build two lists: wrong facts, each traced to the page it came from, and missing questions, where competitors are named and the client isn't. We fix the client's own pages and structured data first, then work through third-party sources in order of how often they are cited, and rerun the set each month so the team can see what moved. This is part of our Growth Studio work, alongside our marketing solutions.

Sources

• Google Search Central, AI features and your website (last updated 10 Dec 2025): https://developers.google.com/search/docs/appearance/ai-features

• Google Search Central Blog, Introducing Search Generative AI performance reports in Search Console (3 June 2026): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports

• Google Search Central, Organization structured data: https://developers.google.com/search/docs/appearance/structured-data/organization

• Google Business Profile Help, Edit your Business Profile: https://support.google.com/business/answer/3039617?hl=en

• Knowledge Panel Help, Submit feedback on content about you: https://support.google.com/knowledgepanel/answer/7534842?hl=en

• Knowledge Panel Help, About knowledge panels: https://support.google.com/knowledgepanel/answer/9163198?hl=en

• OpenAI API Docs, OpenAI crawlers (bots): https://developers.openai.com/api/docs/bots

• Perplexity Docs, Perplexity crawlers (bots): https://docs.perplexity.ai/guides/bots

• Vizio AI, Growth Studio (retrieved 1 Nov 2026): https://www.vizio.ai/studios/growth-studio