What an AI visibility audit includes

An AI visibility audit should help you decide what to change, who should change it and how to check the result. A collection of screenshots may show that your brand appeared somewhere. It does not, on its own, explain the business problem or give your team a workable next step.

Before commissioning an audit, ask to see its proposed scope and evidence format. The following framework describes the components worth agreeing in advance. It includes a clearly labelled worked example and a public example from Zenith Cite’s own website. It does not present an anonymised client report or promise an outcome for a particular business.

1. A defined commercial question

The first page of an audit should explain what the business wants to learn. A SaaS company might want to understand whether small agencies encounter its product during a comparison. A local business might need to check whether answers describe its service area correctly. Those questions require different prompts and different reference information.

Agree the products or services, intended customers, markets, languages and competitors in scope. Name the interfaces to be checked, such as ChatGPT search, Perplexity or particular Google search experiences. Avoid a vague promise to cover “all AI”. List exclusions, access limitations and any account settings that cannot be controlled.

Also define the handover. Does the fee cover research only, implementation, a follow-up check, or some combination? Specify the deliverables before work starts. A short scope that both parties understand is more valuable than a long report whose practical boundaries emerge only after delivery.

2. A reproducible answer baseline

The audit needs a question register that connects each prompt to a customer need. Separate discovery questions from branded accuracy checks. Record the exact wording, platform, date, language, relevant settings and repeated runs. Keep the full responses and source links in a format the client can inspect.

Define mentions, recommendations and citations independently. Explain how ambiguous answers are classified and how failed runs are handled. If the brand appears in two of twelve completed checks, the report should show those twelve checks. It should not turn that limited observation into a claim about total consumer exposure.

A sensible baseline is small enough to review carefully and broad enough to cover the agreed buying questions. The number of prompts is a scope decision. Ask why those prompts were chosen, rather than assuming a larger automated count means a better audit. Our manual ChatGPT checking guide explains a starter method you can inspect yourself.

3. A technical access review

Review the pages that matter to the agreed questions. Check their HTTP responses, canonical URLs, indexing directives, sitemap entries and internal links. Look at what a visitor and a crawler can actually retrieve, including whether important product information is present in the page text.

For ChatGPT search, inspect the access controls relevant to OAI-SearchBot. For Perplexity, its crawler documentation recommends allowing PerplexityBot and its published IP ranges. Review hosting or firewall restrictions as well as robots.txt. Record a confirmed block separately from a suspected access issue.

The deliverable should identify affected URLs and practical consequences. “Technical SEO needs work” is not actionable. “The service page redirects to an unrelated page, and its intended content cannot be retrieved at the linked URL” gives a developer something specific to reproduce. A local fix should remain marked as awaiting deployment until the public response has been checked.

4. A factual identity and content review

Compare answers with a reference sheet approved by the business. Include the correct name, website, services, prices where public, contact information and confirmed geography. Do not fill missing fields by guessing. The reference sheet gives reviewers a consistent basis for deciding whether an answer is accurate.

Then inspect the pages that should support those facts. A service page needs enough detail for a buyer to understand who it is for, what is included and what happens next. Clear authorship and contact details help a reader check who is responsible. Structured data should describe those same visible facts.

As a public example, Zenith Cite’s About page identifies Taksh Dange, links his professional profile and gives a direct email address. The audit page supplies the price and scope information. An identity review can check consistency between these pages without claiming that the presence of an About page caused an AI recommendation.

5. Source and competitor comparisons

For each meaningful answer, list the linked sources and inspect what they actually support. Distinguish your own pages from independent reviews, directories, publisher articles and competitors’ websites. A citation beside a statement does not remove the need to check that statement against the source.

Compare competitors using the same customer questions and scoring definitions. Look for observable differences: a detailed integration guide, current public pricing, a relevant comparison page, or a clear explanation of the intended customer. Avoid claiming access to a platform’s private ranking logic.

The report should label hypotheses. “The answer cites a competitor’s integration guide, and our site lacks equivalent documentation” is an observable gap plus a possible explanation. “The competitor won because it has more authority” skips the evidence unless authority has been defined and measured in a way relevant to the finding.

6. An issue register your team can use

The resulting task might be to publish accurate recurring-booking documentation, link it from the product page and assign a product owner to verify the feature description. The completion check is whether that information is available and correct. A later answer check can observe visibility, but the audit should not promise that publication will produce a recommendation.

Each issue needs an evidence link, affected page, recommended action, owner, estimated effort and priority rationale. Keep confidence separate from priority. A confirmed access failure may deserve immediate attention; a plausible content opportunity may need research before implementation. This makes the handover useful to a small team with limited time.

7. A follow-up measurement plan

Preserve the original question set and keep dated change notes. Decide when to repeat the checks and what would count as progress. Examples include corrected descriptions, fewer identity errors, more accurate cited pages or improved results for an agreed subset of buying questions.

Use available platform reporting with its boundaries intact. Bing’s AI Performance documentation describes citation reporting across supported Microsoft and partner experiences. That is a useful additional source of evidence, not a complete record of every AI answer on the internet.

Keep website enquiries and sales outcomes in the same review conversation. If answer coverage rises but the enquiries are irrelevant, revisit the question selection and landing-page offer. The audit should support a business decision, not create a score that improves while the customer fit gets worse.

What to ask before you buy

Ask for a deliverables list, an example of the evidence structure and an explanation of the limits. Confirm who reviews factual claims, who implements changes and whether the follow-up is included. Check that the report can be shared with your developer or content team without needing the original analyst to interpret every line.

At Zenith Cite, the published one-time AI visibility audit starts at €750, including taxes. The exact scope is agreed before work begins. The free snapshot is an initial conversation about your starting point, not a substitute for a scoped audit. Ongoing implementation is described separately on our GEO services page.

Leave the conversation with a clear question to answer and a defined set of deliverables. That gives you a basis for judging the work by its evidence, usefulness and follow-through.

Start with your business.

Bring your website and the customer questions that matter to you. A free AI visibility snapshot is a practical starting point for discussing your current presence and useful next steps.

Book your free snapshot ↗