21 Questions to Ask an AI Visibility Audit Provider Before You Hire One

Introduction

You are twenty minutes into a call with an AI visibility audit provider, and everything sounds reassuring.

They test the major AI platforms. Their methodology is comprehensive. Their reporting is actionable. They can benchmark you against competitors and tell you what to improve.

The problem is that almost any provider can say those things. You still do not know what they will test, how they will interpret the answers, what evidence you will receive, or whether you could trust the comparison six months later.

That is what the questions below are designed to uncover. Use this checklist to get past the sales presentation and understand how an AI visibility audit will actually be conducted.

You can send them before a call, ask them while reviewing a proposal, or use them to compare shortlisted providers. You do not need a perfect answer to every question. You need enough specificity to understand what you are buying—and enough evidence to distinguish a working methodology from a polished description of one.

If you want the reasoning behind these evaluation areas, read How to Evaluate a Provider for an AI Visibility Audit. This article is the companion checklist: the questions themselves, what to listen for, and what should make you pause.

---

Key Takeaways

• Make sure the provider can explain what the audit will help you decide, which buyer situations it will test, and what it cannot establish.
• Ask which AI products, model versions, interfaces, and modes will be tested, what conditions will be recorded, and how normal variation between responses will be handled.
• Do not accept a visibility score without definitions. You should know what counts as an appearance, which observations form the denominator, and how answers are classified.
• Ask to see a representative deliverable and trace at least one finding from the original question and response to the recommended action.
• Confirm how future audits will remain comparable, where your data will go, what you will own, and exactly what the quoted engagement includes.
• Compare the providers' evidence, methodological choices, and incentives—not just their prompt counts, platform counts, dashboards, or sales vocabulary.

---

How to Use This Checklist

Ask every shortlisted provider the same core questions. That makes their answers easier to compare.

Whenever possible, ask for an example rather than accepting a description. A sample question set, complete test record, or redacted deliverable tells you more than another slide about a proprietary methodology.

If you only have time for five questions, start with Questions 3, 5, 7, 12, and 15. Together, they cover test design, provider coverage, response variability, evidence, and repeat measurement.

Do not turn the checklist into a score out of 21. Some questions will matter more for your situation than others. Use the answers to understand the tradeoffs and decide whether the provider's process fits the decision you need to make.

---

Questions About the Audit's Scope and Test Design

Why it matters: An audit can be carefully executed and still be unhelpful if it tests the wrong audiences, problems, or competitors. These questions help you determine whether the provider will design the audit around the decisions your team needs to make rather than apply the same generic test to every company.

1. How do you define AI visibility for this engagement, and what decisions will the audit help us make?

A useful answer should include: A plain definition of AI visibility that states whether the audit measures mentions, recommendations, visible citations, competitive share, accuracy, framing, or some combination of them. The provider should connect those measures to specific decisions, such as identifying unprompted discovery gaps, diagnosing inaccurate brand framing, comparing competitor visibility, or prioritizing content and authority work. It should also name the limits of the exercise.

Be cautious if: The answer promises to explain everything about your AI visibility or guarantees that the audit will reveal exactly why a model produced every response.

2. How will you decide which audiences, customer problems, and buying situations to test?

A useful answer should include: A process for learning about your customers and narrowing the test to meaningful combinations of audience, problem, decision stage, and context. The provider may use your customer research, sales material, website, interviews, or an initial discovery workshop.

Be cautious if: The test begins with a generic list of industry keywords and no discussion of who your buyers are or what they are trying to accomplish.

3. How do you create the questions used in the audit, and will we be able to review them before testing begins?

A useful answer should include: Questions written in natural buyer language, coverage of multiple question types, and a review step that checks relevance without turning every question into brand-friendly wording. The set should include questions that name your brand, questions that name a competitor, and questions that name neither.

Be cautious if: The questions are lightly rewritten search keywords, generated without review, or kept completely hidden from you.

4. How will you choose the competitors included in the audit?

A useful answer should include: Your known competitors, alternatives that solve the same customer problem differently, and brands that repeatedly appear during testing even if you did not identify them in advance. Ask how newly observed competitors will enter an exploratory test set during future audits without silently changing the stable benchmark used for comparison.

Be cautious if: The competitive set is limited to the names you provide. That can miss the companies AI assistants are actually placing in front of your buyers.

---

Questions About AI Providers and Testing Conditions

Why it matters: The same question can produce different answers depending on the AI product, interface, settings, and conversation context. These questions help you understand which experiences the audit will measure and whether the results can be interpreted and compared fairly.

5. Which AI providers and interfaces will you test, and will their results be reported separately?

A useful answer should include: The specific products being tested and whether the provider uses consumer interfaces, APIs, or both. Results should remain available at the individual-provider level, even if the report also contains a summary.

Be cautious if: ChatGPT, Gemini, Claude, Perplexity, or other systems are blended into one score without showing where the underlying results came from. Also be cautious if API results are presented as though they necessarily reproduce the consumer product experience.

6. Which model versions, modes, and testing conditions will you record or control?

A useful answer should include: Test date, model version where it is exposed, interface, language and location where relevant, and the steps used to reduce conversation history or account personalization. It should distinguish standard chat from search or browsing, deep-research or agentic modes, and first-response testing from follow-up conversations. The provider should explain which conditions it can control, which it cannot, and how it records silent or partially disclosed model changes.

Be cautious if: The methodology records only the question and answer, combines materially different modes without labeling them, or claims to know an exact model version when the interface does not disclose it. Without the surrounding conditions, later comparisons can become difficult to interpret.

7. AI systems can answer the same question differently. How do you account for that variability?

A useful answer should include: Repeated testing for important scenarios, preservation of complete responses, and a method for distinguishing a recurring pattern from an isolated answer. The provider should be clear about how much confidence the sample supports.

Be cautious if: A conclusion is based on one response to one question, or if ordinary variation is hidden behind a precise-looking percentage.

---

Questions About Definitions and Measurement

Why it matters: A visibility percentage is meaningless until you know what the provider counted and how it classified each answer. These questions help you determine whether the reported numbers represent the outcomes you care about or hide important differences behind a single score.

8. What exactly counts as a brand appearance in your reporting?

A useful answer should include: A written definition that distinguishes being mentioned, listed, recommended, discouraged, compared, or visibly cited. These are different outcomes and should not be treated as interchangeable.

Be cautious if: Any occurrence of the brand name counts as success, regardless of how the brand was described.

9. Do you separate prompted and unprompted appearances?

A useful answer should include: Separate reporting for questions that name your brand and questions where the assistant introduces it without being prompted. Competitor-prompted questions may also deserve their own category.

Be cautious if: All appearances are combined. A brand that appears only after being named has a different visibility pattern from one that enters an unbranded recommendation.

10. What denominator do you use when calculating an appearance rate or visibility percentage?

A useful answer should include: A precise explanation of what is being counted—for example, eligible questions, complete conversations, repeated runs, or provider-question combinations. The denominator should match the claim made about the result.

Be cautious if: The provider can show you a percentage but cannot explain which observations went into it.

11. How do you classify recommendation strength, accuracy, tone, and competitive positioning?

A useful answer should include: Written classification rules, examples of borderline cases, and a quality-review process. If people make subjective judgments, the provider should explain how those judgments are checked for consistency.

Be cautious if: Complex findings are assigned by an unexplained black-box score or by an analyst with no documented criteria.

---

Questions About Evidence and Deliverables

Why it matters: A polished report is only useful if you can verify its findings and use them to decide what to do next. These questions help you establish whether the provider will show its work, preserve the underlying evidence, and connect its conclusions to specific actions.

12. Can we see a complete sample deliverable and the evidence behind at least one finding?

A useful answer should include: A redacted but representative deliverable—not a hand-picked dashboard screenshot. Ask the provider to trace one conclusion from the original question and response through its classification and into the recommendation.

Be cautious if: You can see the presentation format but not the underlying evidence or analytical path.

13. Will we receive the questions, complete responses, test details, classifications, and cited sources from our audit?

A useful answer should include: Access to the records required to verify important findings and revisit them later. When a response displays citations, the records should preserve the cited domains and URLs so you can see which first- and third-party sources repeatedly appear. The format does not need to be elaborate, but you should not be dependent on a summary slide to understand what happened.

Be cautious if: The provider supplies only aggregate scores, selected quotations, or access that disappears when the engagement ends.

14. How do you turn a finding into a prioritized recommendation?

A useful answer should include: A method that separates what the audit directly observed from the provider's explanation of the likely cause. It should connect the evidence to a probable problem, propose a specific action, explain why that action fits, and state what the next test should examine. For example, cited URLs can provide evidence about visible retrieval behavior, while indexation problems, blocked crawlers, weak third-party coverage, or model knowledge may remain hypotheses that require additional checks. Ask to see one real or redacted example.

Be cautious if: Recommendations are generic tasks such as "create authoritative content," "improve structured data," or "build more citations" without identifying what should change and why. Also be cautious if the provider claims to know the model's internal reason for an answer when the available evidence only supports an inference.

---

Questions About Re-Runs and Measuring Change

Why it matters: Repeating an audit is valuable only when you can tell whether the result reflects a meaningful change, a change in the test, or ordinary variation in AI responses. These questions help you find out whether future results will be genuinely comparable.

15. If we repeat the audit, what will stay fixed and what may change?

A useful answer should include: A stable benchmark set for comparison, consistent definitions, and recorded test conditions. The provider may also maintain a separate exploratory set for new customer questions, competitors, or product capabilities. Exposed model versions should be recorded, and silent product or model changes should be treated as context that may affect comparability. Changes to the methodology should be documented rather than silently mixed into the trend.

Be cautious if: The provider promises a trend line while changing the questions, classifications, model mix, or denominator without explaining the effect.

16. How will you distinguish meaningful improvement from response variability or model drift, and when should we test again?

A useful answer should include: Repeated observations, comparison at the question and provider level, and a cadence tied to the changes you make and the decisions you need to take. The provider should explain how known model updates or unexplained system drift will be noted and should avoid treating every movement in a score as a meaningful change.

Be cautious if: The answer prescribes a universal weekly or monthly schedule without considering what changed, or claims that a small score movement proves the work succeeded.

---

Questions About Confidentiality and Ownership

Why it matters: The audit may involve confidential product, customer, competitor, or strategy information, and some of that information may be submitted to third-party AI systems. These questions help you understand where your data will go, how long it will remain there, and what you will still control after the engagement ends.

17. What information will you need from us, where will it be stored, and will any of it be submitted to third-party AI systems?

A useful answer should include: A clear list of requested data, where discovery material will be stored, who can access it, how it will be used, which external systems may receive it, and when it will be deleted. Ask whether any customer material may touch a system that trains or fine-tunes a model. Sensitive customer, product, or strategy information should not enter the testing process by accident.

Be cautious if: The provider asks for confidential material without explaining where it will go or who will be able to access it.

18. Who owns the questions, responses, analysis, and deliverables, and what happens to our data when the engagement ends?

A useful answer should include: Retention and deletion terms, export options, access rules, and a clear statement of what you can continue using. Also ask whether your data may be used to train internal systems or support work for other customers.

Be cautious if: Important evidence remains locked in the provider's platform or the agreement is vague about reuse and retention.

---

Questions About the Engagement Itself

Why it matters: Even a sound methodology can lead to a disappointing engagement when the scope, responsibilities, costs, or post-delivery support are unclear. These questions help you understand exactly what you are buying and who will be responsible for delivering it.

19. What exactly is included in the quoted price, how long will delivery take, and what would cost extra?

A useful answer should include: The number and type of providers tested, scope of the question set, kickoff requirements, expected delivery date, deliverables, meetings, revisions, data access, and implementation support. It should clearly separate a one-time audit from recurring monitoring or paid re-runs. You should be able to connect the price and timeline to a defined scope.

Be cautious if: The proposal promises a comprehensive audit but leaves the test coverage, evidence, or post-delivery support undefined.

20. Who will perform the audit, who will review the findings with us, and what happens after delivery?

A useful answer should include: The people or roles responsible for test design, analysis, quality review, and strategic recommendations. It should also be clear whether the engagement ends with the report or includes help interpreting and acting on it.

Be cautious if: Senior experts lead the sales process but the provider cannot explain who will conduct the work or answer methodological questions once the project starts.

21. Do you also sell the work this audit may recommend, and what would cause you to conclude that we do not need significant follow-on work?

A useful answer should include: A direct explanation of any implementation, content, GEO, AEO, SEO, or monitoring services the provider may later sell. The provider should show how the evidence trail from raw response to classification to recommendation keeps the diagnosis separate from the upsell. A strong answer may include an example in which an audit found no major gap, recommended only limited work, or advised the customer not to buy an additional service.

Be cautious if: Every audit appears to uncover substantial work that maps neatly to the provider's other services, the customer cannot take the recommendations elsewhere, or the provider cannot describe what a satisfactory null result would look like.

---

Pay Attention to the Questions the Provider Asks You

Evaluation should work in both directions. A credible provider needs to understand what you are trying to learn before deciding how to test it.

Expect them to ask about:

• the decision you need the audit to support;
• your priority audiences and customer problems;
• the categories and competitors buyers consider;
• known inaccuracies or positioning concerns;
• recent product, market, or website changes;
• the people who will use the findings; and
• your ability to implement and re-test recommendations.

If a provider can produce a scope and price without learning any of this, ask what the audit is being tailored to.

---

Compare the Answers, Not the Sales Vocabulary

While speaking with providers, capture notes in the same evaluation areas for each one:

• scope and buyer scenarios;
• questions available for review;
• providers and interfaces disclosed;
• how variability is handled;
• whether definitions and the denominator are documented;
• whether raw evidence is available;
• whether recommendations are traceable to findings;
• re-run methodology;
• data handling and ownership; and
• scope, team, and price.

Do not select a provider simply because it tests the most prompts, includes the most AI platforms, or presents the cleanest score. Those things may be useful, but only when they serve a clear question and produce evidence you can interpret.

The strongest answer is often not the most confident one. A provider that can explain its choices, show its work, and state what the audit cannot establish is giving you something important: a method you can evaluate before trusting its conclusions.

---

Copy the 21 Questions

Use the list below in a proposal review, vendor email, or comparison spreadsheet. Each question matches the sections above; the wording is unchanged so you can paste it as is.

Scope and test design

1. How do you define AI visibility for this engagement, and what decisions will the audit help us make?

2. How will you decide which audiences, customer problems, and buying situations to test?

3. How do you create the questions used in the audit, and will we be able to review them before testing begins?

4. How will you choose the competitors included in the audit?

AI providers and testing conditions

5. Which AI providers and interfaces will you test, and will their results be reported separately?

6. Which model versions, modes, and testing conditions will you record or control?

7. AI systems can answer the same question differently. How do you account for that variability?

Definitions and measurement

8. What exactly counts as a brand appearance in your reporting?

9. Do you separate prompted and unprompted appearances?

10. What denominator do you use when calculating an appearance rate or visibility percentage?

11. How do you classify recommendation strength, accuracy, tone, and competitive positioning?

Evidence and deliverables

12. Can we see a complete sample deliverable and the evidence behind at least one finding?

13. Will we receive the questions, complete responses, test details, classifications, and cited sources from our audit?

14. How do you turn a finding into a prioritized recommendation?

Re-runs and measuring change

15. If we repeat the audit, what will stay fixed and what may change?

16. How will you distinguish meaningful improvement from response variability or model drift, and when should we test again?

Confidentiality and ownership

17. What information will you need from us, where will it be stored, and will any of it be submitted to third-party AI systems?

18. Who owns the questions, responses, analysis, and deliverables, and what happens to our data when the engagement ends?

The engagement itself

19. What exactly is included in the quoted price, how long will delivery take, and what would cost extra?

20. Who will perform the audit, who will review the findings with us, and what happens after delivery?

21. Do you also sell the work this audit may recommend, and what would cause you to conclude that we do not need significant follow-on work?

If you only have time for five, copy questions 3, 5, 7, 12, and 15 first.

If you decide to conduct the work internally, use Viziquo's six-step AI visibility audit process to plan the audit. If you are still comparing approaches, return to the complete guide to evaluating an AI visibility audit provider.

Evaluating a provider only makes sense once you know what a rigorous audit should include. See what a complete AI visibility audit actually looks like.

If you'd like to see how these questions play out against Viziquo's methodology, fill out the form below.