An AI visibility audit should answer three questions: can people and relevant crawlers access the information, is the information accurate and well structured, and can the team repeat the same observations over time? This checklist covers those questions without treating any item as a guaranteed ranking or citation factor.
Google's AI features guidance says existing SEO foundations still apply and there are no extra technical requirements, special schema types, or new AI text files required for AI Overviews or AI Mode. Meeting normal requirements still does not guarantee crawling, indexing, serving, or inclusion.
Content review
1. Does each important page answer its main question clearly?
Put the useful answer where a reader expects it. A concise opening can improve usability, but there is no universal word count or citation multiplier.
Check: Ask a reviewer to identify the page's subject, audience, and main answer without reading the navigation or footer.
2. Do FAQs answer real user questions?
Use FAQ sections when they remove uncertainty for readers. Do not add repetitive questions merely to create markup.
Check: Compare the questions with support conversations, sales objections, search queries, or documented user research.
3. Do headings accurately describe their sections?
Descriptive headings help readers scan and make page structure easier to understand. They do not need to be forced into question form.
Check: Read only the headings and confirm they form an accurate outline.
4. Are material claims traceable to evidence?
Numbers, comparisons, outcomes, and third-party claims should link to a primary source where possible. A citation improves traceability; it does not create a guaranteed AI citation probability.
Check: Record the source, date, sample, context, and limitation for every material external claim. Remove claims that cannot be verified.
5. Are comparison pages fair and current?
Comparison content should distinguish documented facts, your interpretation, and unknowns. Verify competitors' current official pages before publication.
Check: Add a review date and a correction path. Remove invented feature gaps, prices, or superlatives.
6. Does the content cover the user's actual journey?
Map pages to discovery, evaluation, setup, troubleshooting, and renewal questions. Coverage is a planning aid, not proof of visibility.
Check: Identify important user questions that have no accurate public answer.
Technical review
7. Do robots rules reflect the site's real crawler policy?
Review robots.txt intentionally. Different crawler controls serve different purposes, and allowing a crawler does not require it to crawl or use the content.
Check: Test the live file, document why each rule exists, and confirm sensitive pages are protected by authentication rather than robots rules alone.
8. Is llms.txt described and used accurately?
llms.txt is an optional community proposal for a Markdown summary and curated links. It is not a Google requirement or guaranteed AI search signal.
Check: If published, verify every link and statement. No AI system is required to read, index, or cite it.
9. Is agent.json clearly experimental?
agent.json is SkillAEO's experimental format for organizing public metadata, not an industry standard or authorization protocol.
Check: Publish no secrets, use current public URLs, keep experimental: true, and make no discovery or integration promise.
10. Does structured data match visible content?
Use documented Schema.org types and follow each search feature's eligibility rules. Google says valid structured data does not guarantee display.
Check: Validate syntax, required properties, page eligibility, and consistency with visible content. Remove fabricated ratings, prices, or FAQ answers.
11. Is performance acceptable for users?
Measure real user and lab performance, then fix the largest bottlenecks. Avoid presenting one universal response-time threshold as an AI citation rule.
Check: Review Core Web Vitals, server response, caching, image size, script cost, and regional behavior.
12. Do public URLs return the intended status?
Fix broken internal links, accidental error responses, redirect loops, and misleading soft 404s.
Check: Crawl the site and manually inspect high-value pages. A clean response does not by itself prove inclusion in an AI answer.
13. Are images understandable and accessible?
Use informative alt text for meaningful images and empty alt text for decorative ones. Do not stuff keywords or describe invisible claims.
Check: Navigate with images disabled or a screen reader and confirm the content still makes sense.
14. Is the sitemap accurate?
Include canonical, indexable URLs and exclude redirects, errors, and private pages. Submit it to the search tools your team uses.
Check: Compare the live sitemap with canonical tags, robots directives, status codes, and internal links.
Evidence and authority review
15. Are important third-party references genuine?
Relevant independent references can help people verify a brand. Do not purchase deceptive endorsements or treat a link metric as a universal quality score.
Check: Review who published each reference, what it actually supports, and whether the relationship is disclosed.
16. Are roundup and directory listings accurate?
If your brand appears in a list or directory, verify the description, category, price, and link. Inclusion does not guarantee recommendation by an AI system.
Check: Keep a list of material profiles and correction contacts.
17. Are authors and reviewers represented truthfully?
Use real names, roles, relevant experience, and review dates where appropriate. Structured data must match visible author information.
Check: Remove fabricated biographies, credentials, profile links, and anonymous authority claims.
18. Are reviews genuine and policy-compliant?
Publish or quote only real reviews with permission and necessary disclosure. Do not invent people, companies, outcomes, or review counts.
Check: Keep evidence of origin and make incentives clear.
19. Is core brand information consistent?
Names, descriptions, contact details, pricing, and supported features should agree across owned public channels.
Check: Maintain a canonical fact sheet and review material profiles after product changes.
20. Can the team repeat its AI observations?
AI answers vary. Record the exact prompt, system, mode, date, location where relevant, response, cited URLs, and account state.
Check: Repeat the same small prompt set over time. Treat changes as observations, not proof that a specific optimization caused them.
How to prioritize findings
Use risk rather than a made-up universal score:
| Priority | Examples |
|---|---|
| Urgent | Private data exposed, false public claims, broken purchase or sign-in flow, important pages unintentionally blocked |
| High | Incorrect canonical or indexing controls, material structured-data mismatch, widespread broken links |
| Medium | Missing evidence, confusing page structure, stale profiles, incomplete monitoring records |
| Low | Optional formatting improvements with no demonstrated user harm |
Assign an owner and verification method to each action. A fix is complete only after the live or built output has been checked.
Frequently Asked Questions
How often should I run an AI visibility audit?
Use a schedule based on change rate and risk. Recheck after material product, pricing, routing, indexing, or content changes. A stable small site may need less frequent review than a frequently updated marketplace.
Can I do the audit without technical skills?
You can review claims, page clarity, public profiles, and repeatable prompts without code. A developer may be needed for access controls, response codes, performance, canonical URLs, and structured data.
How long does a full audit take?
It varies with site size, access, and evidence quality. Do not rely on a fixed duration promise. Define the pages, systems, and checks before estimating the work.
What is the difference between an AEO audit and an SEO audit?
There is substantial overlap. Both depend on useful content, crawlability, indexing, technical quality, and evidence. An AI visibility review adds repeatable observations of generated answers and checks claims about optional AI-oriented files.
Which items have the biggest impact?
Start with security, false claims, broken core journeys, and accidental access or indexing problems. Then use your own measurements to decide which remaining changes matter for your site.
Related resource
For a narrow starting snapshot, run a SkillAEO anonymous audit: 5 GPT questions, up to 3 audits per day, 1 per domain, subject to daily capacity.