AI Visibility Audit Template: 30 Prompts and a Scoring Sheet

An AI visibility audit is a fixed list of prompts run across AI assistants, with each answer scored for whether the company is named, how it is described, which sources are cited and whether the facts are correct. This template gives 30 prompts in five groups and a scoring sheet, so a team can run a baseline audit in a few hours and repeat it every month.

Updated November 11, 2026

What Are the Key Takeaways From This Template?

The template turns a vague question, "what does AI say about us?", into a repeatable measurement. The points below summarize how to use it.

• Use 30 prompts in five groups: category, brand, comparison, problem and people or news.
• Score every answer on six fields: mention, position, accuracy, sources, error type and action.
• Run each prompt more than once, because assistants do not return identical answers every time.
• Fix factual errors at the source before chasing new coverage, since assistants repeat conflicts they find.
• Repeat the same prompts monthly and compare against the first baseline.

In This Guide

• What an AI visibility audit is
• How to set up the audit
• The 30 prompts
• The scoring sheet
• How to sort errors
• Summary measures
• What to do with the results
• Frequently asked questions

What Is an AI Visibility Audit?

An AI visibility audit is a structured check of how AI assistants describe a company, run with the same prompts, the same settings and the same scoring each time. The structure matters because ad hoc questions produce anecdotes, while a fixed prompt set produces numbers a team can compare month to month.

The audit is the first step of the six-step program in Virgo PR's guide to GEO for PR. It tells a team what is missing, what is wrong and which sources the assistants lean on, before any new coverage is planned.

How Do You Set Up the Audit?

Set up the audit by choosing the assistants, fixing the settings and agreeing the scoring rules before the first prompt is run.

• Choose the assistants your buyers, investors or journalists use. A common set is ChatGPT, Claude, Gemini and Perplexity.
• Turn web access on for each one, so the answer can cite current sources.
• Use a clean session for each prompt, without prior conversation, and keep the wording exactly as written.
• Record the assistant, the model or mode and the date on every row.
• Run each prompt at least twice, because assistants can return different answers to the same question.
• Replace the capitalized words in the prompts: CATEGORY, COMPANY, COMPETITOR, USE CASE, PROBLEM, INDUSTRY, EXECUTIVE and TOPIC.

What Are the 30 Prompts?

The 30 prompts fall into five groups, each testing a different way a buyer or reporter might ask about a company.

Group 1: category prompts (10)

Category prompts do not name the company. They test whether the company appears when someone asks a general question.

• What are the best CATEGORY companies?
• Who are the leading providers of CATEGORY for mid-sized companies?
• How do I choose a CATEGORY vendor?
• What should I look for in a CATEGORY provider?
• Which CATEGORY companies are growing fastest?
• What are the risks of buying CATEGORY from a startup?
• Which CATEGORY vendors do large enterprises use?
• What does CATEGORY usually cost?
• Who are the main players in CATEGORY in the United States?
• What are the alternatives to the biggest CATEGORY vendors?

Group 2: brand prompts (6)

Brand prompts name the company. They test whether the basic facts are right.

• What is COMPANY?
• What does COMPANY do?
• Who founded COMPANY and who runs it?
• Where is COMPANY based and how large is it?
• Is COMPANY a trustworthy provider?
• What has COMPANY announced recently?

Group 3: comparison prompts (6)

Comparison prompts name a competitor. They test how the company is positioned against the alternatives.

• COMPANY vs COMPETITOR: which is better?
• Is COMPANY better than COMPETITOR for USE CASE?
• What are the alternatives to COMPETITOR?
• Which costs less, COMPANY or COMPETITOR?
• Should a small team choose COMPANY or COMPETITOR?
• Why do customers switch from COMPETITOR to COMPANY?

Group 4: problem prompts (4)

Problem prompts describe the need, not the company. They test whether the company is part of the answer to the problem it solves.

• How do I solve PROBLEM?
• What tools help with PROBLEM?
• What is the best way to approach PROBLEM in INDUSTRY?
• What does a good solution to PROBLEM look like?

Group 5: people and news prompts (4)

People and news prompts test whether executives and recent announcements are described correctly.

• Who is EXECUTIVE?
• What has EXECUTIVE said about TOPIC?
• What funding has COMPANY raised?
• What is the latest news about COMPANY?

What Does the Scoring Sheet Look Like?

The scoring sheet has one row per prompt, per assistant, per run, with six fields that turn an answer into data.

Scoring sheet columnsColumnWhat to recordPrompt, assistant, datePrompt number, assistant name and mode, and the date of the runMentionedYes or no: is the company named in the answerPositionFirst named, named among others, or not namedDescriptionAccurate, partly accurate or inaccurate, with the wrong statement copied outSources citedEach source named or linked, and whether it is the company's own page, independent coverage or another typeError typeMissing, wrong, outdated, wrong company or unsupported (see below)ActionThe fix, the owner and the date it is due

How Do You Sort the Errors?

Sort errors into five types, because each type needs a different fix.

Error types and fixesError typeWhat it looks likeUsual fixMissing factThe answer leaves out something important, such as a product line or a locationPublish the fact on a clear owned page and in a profile the assistants citeWrong factA date, figure or name is incorrectFind the source of the error, correct it and update every page that repeats itOutdated factThe answer describes an old product, leader or addressRefresh owned pages and profiles, and issue current news that supersedes the oldWrong companyThe assistant confuses the company with another with a similar nameStrengthen pages that separate the two, with full name, location and founding factsUnsupported claimThe answer makes a claim with no cited sourcePublish a sourced page that states the verified version of the claim

Which Summary Measures Should You Track?

Four summary measures turn the scoring sheet into a scorecard, and each is a simple ratio.

• Mention rate: prompts where the company is named, divided by prompts run.
• Accuracy rate: statements that match verified facts, divided by statements checked.
• Citation share: answers that cite the company's own pages or coverage, divided by answers that cite any source.
• Top cited sources: the outlets and page types that appear most often in the answers, in order.

Report these as trends, not promises. Assistants decide what to cite, and no team controls the answer, which is why the measures describe what happened and not what will happen. See the measurement section of GEO for PR for how to pair them with referral traffic and search data.

What Should You Do With the Results?

Use the results to rank the fixes, starting with facts the company controls and ending with coverage it has to earn.

• Correct wrong and outdated facts on the company's own pages, filings and public profiles.
• Publish missing facts in plain, answer-first pages with dates and named people.
• Look at the top cited sources and plan coverage in the outlets and source types that assistants already use.
• Structure key pages so a system can lift a clear answer, which is the separate discipline described in AEO for PR.
• Repeat the same 30 prompts next month and compare.

Public companies can start with the filings, as explained in the guide to structuring a 10-K for AI comprehension. For a broader view of how assistants describe companies, see Virgo PR's AI research library and the guide to reputation management inside the AI engines.

Virgo PR runs this audit as the first step of its GEO work, alongside media relations. For what a full engagement includes, read the guide to a GEO agency, and for budgeting see GEO agency cost in 2027. You can also see the public relations services page, or contact Virgo PR to discuss a program.

Frequently asked questions

What is an AI visibility audit?

An AI visibility audit is a repeatable test in which a fixed set of prompts is run across AI assistants and each answer is scored for mentions, accuracy and sources.

How many prompts should an audit use?

Thirty is a practical starting point because it covers category, brand, comparison, problem and people questions without taking more than a few hours. Larger companies add prompts for each product line and market.

Which AI assistants should be tested?

Test the assistants your audience uses. A common set is ChatGPT, Claude, Gemini and Perplexity, each with web access turned on.

Why run each prompt more than once?

Assistants can return different answers to the same prompt, so a single run can mislead. Running each prompt at least twice shows which results are stable.

How often should the audit be repeated?

Monthly is a sensible start, with extra checks after a launch, a funding announcement or a major news event.

Can the audit guarantee a company will appear in AI answers?

No. The audit measures what assistants say today and shows what to fix. It does not control what they say tomorrow.

Who should own the audit?

PR, communications and web teams should share it, because the fixes span corrected facts, owned pages and earned coverage.