How to Create an AI Visibility Audit for Prospects

An AI visibility audit for a prospect is a documented investigation of how a business appears in selected AI answers, followed by a careful review of what those observations can support. It should define the question set, surfaces, business identity, timing and interpretation method. It should not pretend that a small sample measures every AI platform or proves the commercial value of an absent mention.

Start with the business decision, then choose a research method that can inform it. Preserve the answers, review identity matches and distinguish observed facts from explanations you have not established. This guide covers an audit deliverable. The AI visibility checker guide explains the measurement category, while the AI lead-magnet guide covers campaign design.

What question should the audit answer?

An AI visibility audit should answer a specific discovery or information question relevant to the prospect. For example, it might investigate whether a business appears in questions about a genuine service and whether descriptions of that business are accurate. It should not begin with a predetermined claim that the prospect needs an expensive optimisation programme.

Ask which services matter, who buys them and what questions customers might reasonably ask before knowing the business name. The owner's understanding of the service can prevent a technically tidy but commercially irrelevant audit. A question about a service the business does not provide cannot establish a meaningful visibility gap for that business.

Record exclusions. If the audit does not examine product shopping, conversational follow-up, every country or authenticated consumer sessions, say so. An explicit exclusion helps the prospect understand the report and gives the agency a defensible boundary. It also prevents a later proposal from treating unperformed research as if it were already complete.

Build a small, deliberate question set

Choose questions by buyer task rather than simply generating many paraphrases. Distinguish discovering a provider, comparing known options and checking information about a named business. These tasks can all be useful, but success in one should not be presented as success in every stage of customer discovery.

For a local prospect, include location context only when it reflects the intended task and the method supports it. Record the exact wording. Do not add adjectives such as “best” or “trusted” merely to provoke flattering output. If a comparison question is relevant, explain why and review whether the answer actually supports the comparison being discussed.

These are research patterns, not a claim that Fusion44 offers an arbitrary prompt editor or runs every category automatically. When using a product, work within its documented controls. When conducting a separate manual audit, disclose that method rather than presenting the manual findings as output from a different tool.

Record the measurement conditions

An audit record should identify the actual surface or provider-mediated method, question, date, business and any location context. Where available, retain relevant configuration information and the original response. This is the minimum context a reviewer needs to understand what was sampled and to recognise a later change in method.

Avoid describing a software-mediated answer as an exact reproduction of every person's consumer account. Sessions may involve different context, and product methods may change. Google's AI feature documentation describes Google experiences; OpenAI's ChatGPT search documentation describes another product. Neither should be used as a universal specification for all AI discovery.

If repeated observations are part of the audit, define the repetition policy before interpreting results. Keep unsuccessful and unhelpful observations as part of the record. A report assembled by repeatedly asking until the preferred result appears is selective evidence. More runs do not repair an unclear question or a poorly defined business identity.

Review the business match manually

Confirm that any mention actually identifies the prospect. Check the name, website, location and relevant service. A common trading name can refer to several organisations; a matching string alone is not enough. For a multi-location business, distinguish the brand from the specific branch being investigated.

Next, classify the role of the mention. The business might be recommended, mentioned as background information, linked as a source or included in a comparison. A named entity is not automatically an endorsement. If a summary reports visibility without describing the role, inspect the underlying answer before making a prospect-facing claim.

Keep unsupported answer content separate from business facts. If an answer attributes a service the prospect does not offer, record the discrepancy and verify the real service information with the owner. Do not repeat the invented service in the audit's executive summary as though the AI response established it. The audit should reduce confusion rather than propagate it.

Separate absence, failure and uncertainty

An absent business mention means the intended business was not found in the usable checked answer under the recorded conditions. A failed request means the measurement was unavailable. An ambiguous identity means interpretation remains uncertain. These outcomes should not share a single alarming label, because they require different responses.

For a failed request, investigate the technical or operational issue before drawing a visibility conclusion. For an ambiguous identity, improve the working record or manually review the match. For a valid absence, examine whether the question is relevant and what additional research would help. None of these outcomes alone proves the business is losing a particular number of enquiries.

This distinction also matters when using percentages. If you report appearances across a defined set, state the denominator and explain how failed or ambiguous observations were handled. Do not call the result market share or universal share of AI recommendations. A transparent sample summary is useful without being promoted into a larger statistic it cannot justify.

Turn observations into a proportionate evidence review

Review the information relevant to the observed question. That may include the prospect's service pages, contact details and the sources linked in the response. The purpose is to find concrete information problems or useful questions for further work. It is not to reverse-engineer a guaranteed recipe from one answer.

For example, the business may genuinely offer an important service but explain it poorly on its website. Improving that explanation can be a sensible communication task. The audit should distinguish the verified information gap from the unproven hypothesis that correcting it will change a particular answer. The recommended action can be worthwhile without a guaranteed placement claim.

Do not recommend fabricated endorsements, invented case studies or additional schema solely because an AI answer seems to prefer a competitor. Review actual business facts and relevant official guidance. If technical access needs investigation, specify that separately from content accuracy. A crawl-access check and a recommendation outcome are different questions.

Assemble the prospect deliverable

A useful deliverable begins with the scope and direct findings, then provides enough evidence for the prospect to understand and challenge them. Keep the executive explanation compact, but retain a clear evidence appendix or working record. The report should remain meaningful if forwarded to another stakeholder without the agency presenting it live.

1. State the business question, service and intended audience.
2. Explain the sampled surfaces, questions, timing and exclusions.
3. Show representative observations without concealing contradictory results.
4. Identify correct mentions, absences, uncertain matches and failed measurements separately.
5. List verified information issues and distinguish them from hypotheses.
6. Propose a bounded next engagement, or explain why no immediate work is justified.

For every recommendation, identify the evidence, the proposed change and how completion would be verified. “Improve AI visibility” is too vague to define a deliverable. “Review the accuracy of these service descriptions and correct confirmed inconsistencies” is something the agency and owner can assess. Future answer observations remain a separate measurement, not the only proof that work was completed.

An illustrative audit scenario

Imagine a fictional specialist consultancy that wants to understand discovery for one genuine advisory service. The audit records a small question set and finds a mixture of correct mentions, absences and one ambiguous match. The owner confirms that the service is commercially important, but the website uses different names for it across several pages.

The agency reports the mixed sample honestly and proposes a focused information review. It does not claim that the inconsistent wording caused every absence. It also does not count the ambiguous match as a recommendation. The first engagement has a clear boundary: establish accurate service language, identify the pages that need correction and document a repeatable baseline for later review.

This is an example of a possible process, not customer evidence. A different prospect might have accurate information and no immediate reason to invest further. The ability to explain that outcome is part of a credible service. See getting AI visibility clients for turning a suitable need into an accountable offer.

Where Fusion44 fits

Fusion44 for Agencies can provide a branded AI check and capture context as the first interaction. Its Agency guide documents Quick and Full options, capture placement and publishing. The agency can then decide whether a broader human-reviewed audit is appropriate, with additional research clearly identified as agency work.

Do not assume the initial diagnostic includes every element of the audit described here. Confirm the actual product scope, and do not infer specific provider or model coverage from a campaign headline. If the prospect needs a custom research programme beyond the available controls, use an appropriate separate method and explain the distinction in the deliverable.

Frequently asked questions

How many questions does an audit need?

Enough to address the defined business question responsibly, rather than an arbitrary large number. Explain how the questions were selected and what they exclude. A small coherent sample is easier to interpret than many repetitive prompts presented without a research rationale.

Can I report a percentage of answers mentioning the business?

Yes, if it is clearly described as a summary of the specified usable sample. State the denominator and treatment of failed or ambiguous observations. Do not present that percentage as universal recommendation share, market share or a measure of customer demand.

Should I include competitor names?

Include relevant businesses actually observed when doing so helps explain the sample. Verify their identity and avoid claims about quality, revenue or causation that the answer does not establish. A competitor's presence is a research observation, not a complete commercial assessment.

Does the audit need private account access?

The initial answer sample may not, but additional technical or performance work can require authorised access. Request only what the defined engagement needs. Clearly state which systems were not inspected so the prospect does not mistake a public-information review for a comprehensive account audit.

Can an audit guarantee future recommendations?

No. It can document observations and support defined research or information improvements. The agency controls its work and reporting method, not every future answer produced by an external system. Describe deliverables accordingly and avoid contracting around an outcome you cannot ensure.

What happens after the audit?

Choose a next step that follows from verified needs: clarify information, investigate access, commission deeper research or take no action. Assign ownership and define completion. A retainer is one possible commercial arrangement, not the automatic conclusion of every visibility sample.

Make the evidence understandable

An effective AI visibility audit helps the prospect understand a bounded observation and decide what to investigate next. Keep the method visible, preserve uncertainty and connect recommendations to real information needs. That produces a more useful foundation for a client relationship than an impressive-looking score with an unclear denominator.

The broader AI agency client acquisition guide explains how to select a service and buyer before offering this audit. A visibility deliverable should follow from that fit.