GEO audit checklist

Start with one page and one job

A GEO audit should answer a narrow question: can this specific page support the answer you want an AI system to give? Choose the page, the audience question, and the intended answer before reviewing implementation details.

This is a page-level checklist, not a promise of inclusion or citation. Search and AI products make their own retrieval and response decisions. Use the checklist to find observable weaknesses, then test outcomes separately.

1. Confirm access

• The public URL returns the intended page.Check the canonical URL without authentication, location-specific gates, or an interstitial that replaces the main content.
• The important answer is present in the page text.Do not require a click, animation, image, or client-side interaction to reveal the only useful explanation.
• Robots rules match your intent.Review robots.txt and page-level directives for the search and AI user-agents you intend to allow. OpenAI and Perplexity publish separate crawler documentation; Google publishes its crawler list. See sources.
• The canonical points to the preferred version.Avoid sending crawlers to a duplicate, parameterized, or obsolete version of the page.

2. Make the answer explicit

• Name the question in the heading or opening context.A reader should know what the page resolves before reading the full page.
• Give a direct answer early.State the useful conclusion before the supporting detail. Do not make the reader reconstruct it from slogans.
• Keep the page focused on one primary job.Separate unrelated questions when combining them makes the answer harder to locate or quote.
• Define specialized terms.Use the same name consistently for the company, product, category, audience, and important concepts.

3. Clarify the entities

• Identify who is speaking.Name the organization or author responsible for the page and link to supporting identity pages where useful.
• State what the offering is and who it is for.Category, audience, use case, and scope should be written in plain language.
• Disambiguate similar names.Add location, industry, product type, or another meaningful qualifier when the entity could be confused with something else.
• Keep visible copy and structured data aligned.Google says structured data should describe the page it appears on. Treat schema as a machine-readable description of visible facts, not a substitute for them. See sources.

4. Support consequential claims

• Separate facts, observations, and interpretation.Readers should be able to tell what was measured, what was observed, and what you conclude.
• Attach a source to externally verifiable claims.Link to the original documentation, research, dataset, or named evidence close to the claim.
• Publish method, date, sample, and limitations for original research.Do not present a small test, illustrative example, or changing product behavior as a universal result.
• Label fictional or composite material in the article text.Put the label where text extraction preserves it and repeat it beside highly quotable fictional copy.

5. Make useful passages quotable

• Write self-contained sentences.A quoted sentence should retain its subject, context, and important qualification.
• Use descriptive subheadings.Headings should identify the answer below them instead of relying on clever or generic labels.
• Use tables and lists for true comparisons and procedures.Keep row labels, units, time periods, and comparison bases explicit.
• Put limitations beside the claim they qualify.Do not isolate essential context in a remote disclaimer.

6. Establish a citation baseline

• Choose questions before checking results.Include brand-free discovery, comparison, and known-brand retrieval questions.
• Record the exact product, date, question, and run.AI product behavior changes. A reproducible log is more useful than an undated screenshot.
• Separate mentions, citations, and no-search responses.Record what is visible without inferring an unseen retrieval process.
• Repeat tests after one controlled page change.Compare the new observations with the baseline; do not call one favorable response a stable ranking.

How to interpret the audit

Do not collapse the checklist into a universal score. Failures in access can prevent the rest of the page from mattering; weak evidence can make a clear answer unsafe to reuse; and a well-built page can still be absent from a particular response. Use the result as a prioritized repair list.

1. Fix access or wrong-page problems first.
2. Rewrite the primary answer and entity context.
3. Remove, qualify, or source unsupported claims.
4. Improve the most reusable passages.
5. Establish a baseline, publish one controlled change, and retest.