What Should an AI Visibility Audit Include?

What should an AI visibility audit include? At minimum, it should test whether important pages can be discovered and delivered, whether their content is clear enough to retrieve and cite, and whether your brand actually appears across a repeatable set of prompts. It should also preserve the raw evidence behind every score.

This checklist is for website owners, in-house marketers, and agencies evaluating an audit provider or building their own process. Use it to prevent a common failure: buying a polished dashboard that reports one proprietary number but cannot show which pages, prompts, crawlers, or technical conditions produced the result.

What should an AI visibility audit include before testing?

Define the decision the audit must support before running tools. A product team may need to understand why commercial pages are absent from generated recommendations; a publisher may care more about citations; an agency may need a repeatable benchmark for several clients. That decision determines which pages, prompts, competitors, locations, and AI experiences belong in scope.

Choose representative URLs rather than testing only the homepage. Include a core commercial page, a category or service page, a strong knowledge article, a recently published page, and one URL known to have a technical problem. Save the audit date and any major website changes so later comparisons remain meaningful.

• Page inventory: URL, page type, intended audience, target question, canonical URL, status code, and indexability.
• Access evidence: robots.txt rules, relevant crawler user agents, CDN or WAF logs, and anonymous fetch results.
• Content evidence: rendered HTML, headings, answer passages, author or organization details, sources, and structured data.
• Visibility baseline: exact prompts, platform, date, login state, location where relevant, mentions, citations, and cited URLs.
• Business context: priority products, markets, competitors, conversions, and the owner of every recommended fix.

Check 1: discovery and crawler access requirements

The first audit layer asks a narrow question: can the systems in scope reach the pages you want them to use? Review robots.txt separately for each relevant crawler, then compare those rules with server and CDN behavior. A rule that permits access is not proof of a successful fetch; security software may still challenge, rate-limit, or block the request.

Test the final URL and every redirect hop. Record the HTTP status, response headers, canonical target, robots meta directives, and whether the server delivers meaningful HTML to an anonymous client. Distinguish crawling from indexing: Google explains that robots.txt manages crawler access, while a noindex directive controls indexing only when the crawler can access the page. See the official robots.txt guidance for that boundary.

Check 2: technical delivery, rendering, and index controls

A successful response is only the beginning. Confirm that the main answer, headings, internal links, and evidence are present in the delivered or rendered HTML—not inserted only after an interaction that automated systems may never perform. Check mobile and desktop templates, client-side rendering, pagination, lazy loading, and consent layers where they can hide the main content.

For Google’s AI features, eligibility builds on normal Search foundations. Google’s AI features documentation says a supporting page must be indexed and eligible to appear with a snippet; there is no separate technical shortcut. Therefore, the audit should include status codes, canonicalization, internal discovery, sitemap presence, duplicate handling, structured data validity, and page experience alongside AI-specific crawler checks.

• Flag redirect chains, soft 404s, server errors, blocked assets, inconsistent canonicals, accidental noindex, and important orphan pages.
• Compare source HTML with rendered content for the exact answer passages and links users need.
• Confirm structured data matches visible content; treat schema as supporting context, not a substitute for a useful page.
• Retest failures from more than one network or tool before blaming an AI platform.

Check 3: content clarity, entities, and citation signals

When asking what should an AI visibility audit include at the content layer, start with clarity. The audit should evaluate whether a page makes its subject, audience, and answer obvious. Strong pages normally name the entity consistently, answer the central question early, use descriptive headings, define important terms, and support material claims with first-party evidence or reputable sources. The goal is not to repeat a keyword mechanically; it is to reduce ambiguity.

Review organization and author information, product names, locations served, dates, methodology, statistics, and source links. Look for passages that can stand alone without losing their meaning. Generic statements such as “our solution improves performance” provide little evidence. A specific explanation of what was measured, how it was calculated, and what limitation applies is easier for both people and retrieval systems to evaluate.

Check 4: platform tests and pass/fail recording

Platform testing must use a documented prompt panel, not a few hand-picked screenshots. Separate branded prompts from non-branded discovery, comparison, problem, and purchase prompts. Keep the wording stable for the baseline, run the same panel across the AI experiences relevant to the audience, and record the complete output where permitted.

For each response, capture whether the brand was mentioned, whether the domain received a clickable citation, which URL was cited, the prominence and accuracy of the mention, and which competitors appeared under the same conditions. OpenAI’s publisher guidance notes that sites allowing OAI-SearchBot can track ChatGPT referral traffic with the utm_source=chatgpt.com parameter. Include that attributable traffic, but do not treat referral sessions as the only visibility signal.

Because generated answers can vary with wording, platform, date, geography, account state, retrieval freshness, and model changes, define a pass condition before testing. A pass might mean access succeeds on all priority URLs, or that citation rate improves for a fixed commercial prompt group over three repeated runs.

Prioritize findings by risk, impact, and proof

A long issue list is not a strategy. Every finding should state the affected URLs, evidence, likely business impact, confidence level, owner, effort, and retest method. Fix access and delivery failures before polishing content that systems cannot reliably fetch. Then improve high-value pages with weak answer fit or evidence before expanding lower-priority coverage.

Priority	Use when	Example action	Pass condition
Critical	Discovery or delivery is blocked	Remove an unintended crawler block or server challenge	Priority URL returns usable content consistently
Important	Page is accessible but unclear or unsupported	Add a direct answer, methodology, evidence, and descriptive internal links	Required passage and evidence appear in rendered HTML
Improvement	Baseline works but coverage or measurement is incomplete	Expand the prompt panel or publish a missing comparison page	New scope is documented and repeatable

Evidence and screenshots the report should include

• The exact robots.txt rules and fetch result for every crawler or system tested.
• Status codes, redirect hops, canonical tags, robots directives, and rendered HTML for representative URLs.
• The complete prompt panel or a versioned sample, with platform, date, location, and account state where relevant.
• Raw answers showing mentions, citations, cited URLs, competitors, and any inaccurate statements.
• Calculations for mention rate and citation rate, including the denominator and eligibility rules.
• Analytics evidence for attributable AI referrals, plus the limits of that data.
• Before-and-after evidence for each fix and the unchanged test used for verification.
• A prioritized backlog with owner, effort, confidence, expected impact, and retest date.

The most common interpretation mistake

Frequently asked questions

How many prompts should an AI visibility audit test?

There is no universal number. Start with a focused panel that represents meaningful customer journeys and can be repeated consistently. Twenty to forty well-chosen prompts may produce a more useful baseline than hundreds of loosely related queries. Expand by intent and market once the method is stable.

Which AI platforms should be included?

Include the products your customers actually use and that you can test responsibly. The mix may include ChatGPT search, Google AI features, Microsoft Copilot, Perplexity, Gemini, or others. Record the product and date because availability, models, citations, and behavior change.

Can an audit guarantee AI citations?

No. An audit can verify access, reveal weak evidence, measure observed mentions and citations, and recommend controllable improvements. External systems decide what to retrieve and cite. A credible provider states that limitation clearly.

How often should the audit be repeated?

Repeat after important technical or content changes using the same baseline inputs. Monthly testing suits many active programs; weekly tests may support controlled experiments, while quarterly checks may be enough for stable sites. Consistency matters more than frequency.

Should an AI visibility audit replace an SEO audit?

No. It should extend technical and content SEO with crawler-specific access checks, prompt-based visibility measurement, citation evidence, and platform comparison. Conventional discoverability remains part of the foundation.

Next step: turn the checklist into a repeatable baseline

The practical answer to what should an AI visibility audit include is evidence at every stage: access, delivery, content, retrieval outcomes, and measurement. Use the checklist to define scope, record raw findings, prioritize fixes, and retest with unchanged inputs. For a complementary comparison of measurement and optimization work, read AI visibility audit vs GEO audit.