AEO Audit Template: What to Check and in What Order

An AEO audit checks whether AI search systems can access your pages, understand your brand, extract useful answers and find trusted sources that support what you say. Work through this checklist in order, from crawl access to tracking, because there is little value in improving content on pages that cannot be fetched. For help turning findings into an ongoing process, explore SearchFIT’s AI search resources.

This template is for in-house marketing teams and agencies auditing a brand’s visibility in ChatGPT, Perplexity, Google AI Overviews, Claude and Gemini. It combines technical checks with content, third-party coverage and measurement, rather than treating AI visibility as a schema-only or rankings-only problem. For complementary industry frameworks, review Revv Growth's AEO audit checklist and Go Fish Digital's GEO audit framework.

A useful audit ends with evidence, owners and priorities, not just a list of recommendations. Record what you tested, what you found, how serious it is and what you will change. Then repeat the same checks after fixes so you can see whether access, citations and brand representation have changed.

How to use this AEO audit checklist

Start with your most important market, audience and product or service category. Make a working list of the pages and questions that matter most to customers, such as pricing, product comparisons, implementation, alternatives and common problems. The audit will be clearer if every finding is tied to a page, prompt or business question.

Use the sequence below as a dependency order. A problem in an early tier can make later work less useful, but it does not mean you should stop documenting later issues. Note them, fix the blockers first, then return to the rest of the checklist.

For each checklist item, mark it as Pass, Needs work, Not applicable or Not tested. Add a URL, screenshot, log, prompt or other evidence. “Not tested” is not a pass, and “not applicable” should include a short reason so the next person understands the decision.

Tier 1: Check crawl access and technical delivery

Begin here. Confirm that an AI search product or its crawler can request the pages you want it to use. Check the public site, not just a logged-in preview or the version you see in a browser. A page may look complete to a person while delivering an error, a challenge page or very little useful text to a crawler.

• List the priority URLs to test. Include your homepage, key product or service pages, pricing, comparisons, help content and the strongest informational pages, because a crawl check is only useful when it covers pages that answer customer questions.
• Check robots.txt and relevant crawler rules. Look for accidental blocks affecting search or user-triggered fetch crawlers, and document separate decisions for training crawlers rather than treating every AI-related bot as the same thing.
• Verify the current bot names against reliable documentation. Crawler names and behaviour can change; Cloudflare’s AI Crawl Control bot reference is one place to check the bots it recognises before updating rules.
• Test the response code for each priority URL. Confirm that the intended public page returns a successful response without redirect loops, access-denied responses or unexpected login requirements, because a crawler cannot use content it cannot retrieve.
• Review CDN, firewall and bot-management settings. Check logs and rules for blocks, managed challenges, rate limits or other controls that could affect declared crawlers, because a robots.txt permission does not override a server-side denial.
• Inspect the raw HTML response as well as the rendered page. Confirm that key facts, product descriptions, prices where appropriate and page headings are present in the server response, because relying only on client-side rendering can make content harder for some retrieval systems to access.
• Check canonical tags, redirects and duplicate versions. Make sure each important page points to the intended canonical URL and that redirects do not send crawlers to an irrelevant or outdated destination.
• Test mobile and desktop delivery. Compare the content and functionality served to both, because mobile-specific scripts, consent layers or layout changes can hide essential information.
• Review page speed and server errors in logs. Look for slow responses, intermittent failures and timeouts on important pages, because an unreliable fetch can make otherwise useful content unavailable at the moment it is needed.
• Confirm that important content is not blocked by access controls. Check cookie walls, geographic restrictions, paywalls and login states, then decide deliberately which material should remain public and fetchable.
• Check sitemap coverage and index directives. Confirm that the sitemap includes the pages you want discovered and that those pages are not marked noindex or blocked by another directive.
• Retest after every technical change. Fetch the same URLs again and retain the result, because changing a robots rule or firewall setting does not prove that the page is now accessible.

Keep search access separate from training preferences. Soar identifies search-oriented bots such as OAI-SearchBot, Claude-SearchBot and PerplexityBot, and distinguishes them from training-related crawlers such as GPTBot, ClaudeBot and Google-Extended. You can review compiled tokens and rules in Soar's AI bots robots.txt guide. Treat this as a prompt to verify current provider documentation, not a permanent list: bot names and policies can change. If the distinction matters to your organisation, confirm each provider’s current guidance before changing robots.txt.

Do not assume that allowing a crawler guarantees citation, or that blocking one bot removes your content from every AI product. The audit question is narrower and testable: can the relevant service fetch the public page, and does it receive the intended information? Record what you can verify and avoid claiming more than the evidence supports.

Tier 2: Make your brand and entities unambiguous

Once pages are accessible, check whether a system can tell who you are and how your organisation relates to its products, people and other names. Entity clarity means making those relationships explicit and consistent. It does not mean adding every possible schema type or creating profiles where your business has no legitimate presence.

• Write a canonical brand description. Agree on the organisation’s official name, what it does, who it serves and the markets it operates in, because inconsistent descriptions across pages and profiles create avoidable ambiguity.
• Check the About and contact information on your site. Ensure the legal or trading name, location and contact details are current and consistent with your public business profiles, because these details help distinguish one organisation from another.
• Audit your Organization structured data. Check that the schema describes the real organisation and uses accurate fields, rather than adding unsupported claims or details simply because a template offers them.
• Review sameAs links individually. Link only to official, current profiles that clearly refer to your organisation, because a wrong or outdated profile can reinforce the wrong identity.
• Check product and service names across the site. Use a consistent name and explain what each offering does, because overlapping or frequently changing labels make it harder to connect a product with its parent brand.
• Review Product or SoftwareApplication data where it genuinely applies. Confirm that visible product facts and structured data agree, because markup should describe content a visitor can also verify on the page.
• Check author and reviewer profiles. Make the person’s role and relevant expertise clear, link to a trustworthy profile where appropriate, and avoid attributing work to a person who did not create or review it.
• Keep Article details accurate. Check the visible author, publication date and updated date against any Article markup, because stale or conflicting dates can undermine confidence in the page.
• Review third-party identity records. Check relevant company directories, professional networks and knowledge resources for old names, leadership details, acquisitions or product descriptions that need correction.
• Look for confusing namesakes. Search your brand and product names alongside your category, location and common misspellings, then note where another organisation or product may be mistaken for yours.
• Avoid manufacturing authority signals. Do not create a profile or entity entry that you cannot substantiate; focus instead on accurate, consistent information in sources where your organisation belongs.

Structured data can help machines interpret information, but it cannot make inaccurate or absent facts trustworthy. Check the page a person sees first, then verify that markup represents the same facts. If the page says one thing and the structured data says another, fix the source rather than choosing whichever version sounds more favourable.

Tier 3: Audit answers, passages and page citability

Answer engines may use a specific passage rather than the full page. Audit each important page at the level of its sections: could a reader understand the answer under a heading without needing to reconstruct it from several distant paragraphs? A useful passage states its subject, gives a direct answer and provides enough context to be accurate.

• Give each important page one clear job. State the audience, question and next step the page serves, because a page trying to answer unrelated questions often fails to give any of them enough depth.
• Match headings to actual questions or decisions. Use descriptive headings that tell readers what the next section answers, because vague labels make both scanning and passage selection harder.
• Answer the section question near the start. Lead with a concise, direct response before adding background, caveats or examples, because the reader and a retrieval system should not have to search through a long introduction for the point.
• Make each answer self-contained. Name the product, process or subject in the passage instead of relying on phrases such as “it” or “this approach” without a nearby reference, because extracted text may appear without the preceding paragraph.
• Define specialist terms on first use. Explain terms such as AEO in plain language, because not every visitor or system will interpret an acronym the same way.
• Use lists for genuine steps, options or requirements. Keep each point distinct and ordered where sequence matters, because a clear list is easier to follow than a dense paragraph that mixes several actions.
• Use comparison tables when readers need to compare like with like. Include the same relevant criteria for each option and explain the basis of comparison, because a table is only useful when it is fair and interpretable.
• Support measurable claims with named sources. Include the source, date or relevant context close to the claim, because an unattributed number is difficult to check and can weaken trust.
• Add original evidence only when you can explain its method. Identify how data was collected, what it covers and what it cannot show, because a statistic without context can mislead more than it helps.
• Use examples that clarify the answer. Show how a recommendation applies to a real task or scenario, because concrete examples help readers assess whether the advice fits their situation.
• Include limitations and conditions. State when a recommendation does not apply or what a result depends on, because useful answers are precise rather than universally confident.
• Check for information gain. Ask what a reader learns here that is not already obvious from a generic summary, because distinctive expertise, clear analysis or useful evidence gives the page a reason to be cited. For formats that earn citations, see our guide on content strategies for AI search visibility.
• Audit comparison and alternative pages. Address the named options directly, use consistent criteria, disclose relevant relationships and keep details current, because evasive or one-sided comparisons are less useful to buyers.
• Make product facts easy to verify. Keep features, integrations, availability, pricing conditions and implementation details in visible text where appropriate, because vague marketing language does not answer specific buyer questions.
• Remove contradictions across related pages. Check that guides, product pages, FAQs and pricing information agree, because conflicting facts can leave readers and systems unsure which version is current.
• Review internal links and page context. Link to relevant supporting explanations and use descriptive anchor text, because a passage is more useful when readers can follow its evidence or next step.
• Check freshness where facts change. Revisit claims about products, policies, regulations and market conditions on a schedule, because an answer that was accurate last year may now mislead.

The Princeton and Georgia Tech Generative Engine Optimization study reported visibility improvements of up to 40% for some tested content optimisation methods. In its reported results, citing authoritative sources was associated with a 40% improvement, adding statistics with 37% and adding quotations with 30%. These are findings from that study’s experiments, not a guaranteed outcome for your pages or a promise that adding a citation or statistic will produce the same result. Read the GEO study on arXiv and use it as a reason to test useful evidence, not as a target to copy mechanically.

A practical passage test is to copy a section without its neighbouring text and ask a colleague to identify the question it answers, who or what it describes, and what evidence supports it. If the passage becomes ambiguous on its own, revise the opening, name the subject and add missing context. You do not need to force every section into the same word count; completeness and clarity matter more than a fixed template.

Tier 4: Find third-party coverage and citation gaps

A site cannot fully control the information available about its brand. Review independent sources because AI answers may draw on sources beyond your domain, and customers also use those sources when comparing options. This tier is about understanding the external picture and correcting genuine gaps, not manufacturing mentions or trying to manipulate discussions.

• Collect a set of real category questions. Include informational, comparison, recommendation and problem-solving prompts, because a brand can appear for one type of question and be absent for another.
• Ask relevant AI search tools the same prompts. Save the date, wording, platform, answer and cited sources, because outputs can differ across engines and change over time.
• Record every cited source and named competitor. Include owned pages, publishers, review platforms, forums and directories, because the sources around a recommendation can reveal what information your own site does not provide.
• Group citations by source type and topic. Separate editorial coverage, reviews, communities, directories and official sources, because each type may shape a different part of a buyer’s decision.
• Compare the sources cited for your brand with those cited for alternatives. Look for a factual coverage gap, not just a mention-count gap, because a competitor may be cited for a specific useful explanation you do not publish.
• Check accuracy on important external profiles. Review product names, descriptions, availability, pricing claims and outdated business details, because incorrect public information can create a poor customer experience even when your website is accurate.
• Find credible places where customers already evaluate your category. Prioritise relevant publications, review sites, associations and communities, because a presence on an unrelated high-profile site is not useful evidence of fit.
• Read community discussions in context. Note recurring customer questions and concerns, but do not treat a small number of posts as representative of all customers, because forum conversations are anecdotal and may not reflect the whole market.
• Respond to errors through the source’s normal process. Contact the publisher or platform with clear evidence when a material fact is wrong, because corrections are more durable than publishing a contradictory claim on your own site.
• Build a useful earned-media plan. Offer subject expertise, original research or a genuinely helpful resource to relevant publishers, because the goal is credible information that serves their audience.
• Avoid fabricated reviews, undisclosed promotion and spam. These tactics can damage trust and may break platform rules, because apparent volume is not a substitute for genuine customer experience.
• Track source changes over time. Revisit priority citations and profiles after campaigns, product changes or corrections, because third-party pages can remain outdated even after your own site is fixed.

The State of AI Search Citations report from Metavert estimates that 85.7% of answer-engine citations point to third-party domains rather than a brand’s own sites. Treat that figure as the report’s estimate and check its methodology and scope before applying it to a particular sector. Its practical implication for an audit is to include external sources, not to assume a fixed share for every query. See Metavert’s citation report for its analysis.

When a competitor appears in an answer despite ranking below you in conventional search results, treat that as a diagnostic question rather than a contradiction. Check whether the competitor has a clearer answer to the specific prompt, more relevant independent coverage, or better-maintained product facts. Then test whether those differences are real and addressable. Higher organic position alone does not establish that your page is the best source for every generated answer.

Tier 5: Set an AI visibility baseline

A visibility baseline gives you a repeatable way to assess whether the work is changing how your brand appears. It is not a universal score for the quality of your marketing. Results vary by prompt and platform, so keep the query set and observation method stable enough to compare, while recording changes that could affect the answers.

• Define the audience, geography and category. Specify who is asking, which market they are in and what type of decision they are making, because an answer for one audience may not describe visibility for another.
• Build a balanced prompt set. Include questions about needs, product categories, comparisons, alternatives, implementation and brand-specific facts, because testing only your brand name will miss discovery and consideration queries.
• Use natural question wording. Include variations that resemble how a customer might ask, because a single carefully crafted prompt cannot represent every way people express the same need.
• Choose platforms that matter to your audience. Record the AI search products you test and why, because testing every available product may be impractical and may add noise without informing a decision.
• Save the exact prompt and complete response. Capture citations, recommendation order and relevant context, because a mention without the surrounding answer can be hard to interpret.
• Track brand mentions separately from clickable citations. A brand can be named without a link, while a page can be cited without a clear recommendation, because those are different kinds of visibility.
• Record the context and sentiment of each mention. Note whether the brand is recommended, described neutrally, compared, or criticised, because a bare mention count can hide a material misrepresentation.
• Track prompt coverage. Mark which priority questions return a relevant answer about your brand, because this shows where your visibility is missing rather than only how many times your name appears.
• Record cited URLs and source types. Note whether the answer cites your own site or an external source, because the source may reveal which page or third-party coverage is shaping the response.
• Repeat tests with consistent conditions. Keep prompt wording, platform and collection method as stable as practical, and note changes, because results from different setups are not directly comparable.
• Use manual checks to validate automated monitoring. Review a sample of outputs yourself, because a tool’s classification may not capture nuance, incorrect claims or the reason a source was selected.
• Keep a change log alongside the baseline. Record content updates, technical fixes and external campaigns, because a timeline helps you interpret movement without assuming that every change caused it.

For each result, store the observation date, prompt, engine, answer text, cited sources, brand mention, sentiment and any relevant notes. You can calculate simple measures from these records, such as the share of tested prompts where the brand was mentioned or cited. Define the denominator and rules before comparing periods, and do not present a small or changing prompt set as a market-wide measure.

SearchFIT’s AEO and AI visibility KPI guide can help teams choose measures for ongoing reporting. If you are comparing an internal workflow with a specialist platform or external provider, use SearchFIT’s comparison page to understand the differences and choose tools based on the audit work you actually need.

Tier 6: Connect AI referrals to analytics

AI visibility and web traffic answer different questions. A citation or mention may influence a decision without producing an immediate visit, while a referral visit does not by itself prove that a generated answer caused a conversion. Set up analytics to measure the traffic you can identify, and describe its limits clearly. If your data team needs to query referral patterns alongside product telemetry in plain English, D23 provides managed Apache Superset and AI analytics.

• Check whether AI referral sources appear in your analytics. Review referral reports and landing pages for traffic from services such as ChatGPT, Perplexity or Claude, because unclassified visits can be difficult to compare later.
• Create a documented source list. Record the domains and app referrers you intend to classify, then validate them against actual analytics data, because referral formats can differ and may change.
• Use a custom channel grouping only after testing it. Keep rules specific enough to avoid classifying unrelated traffic as AI referrals, because an overly broad rule can make reports misleading.
• Inspect landing pages from identified AI referrals. Check which pages receive visits and whether those pages answer the likely follow-up question, because a source report becomes more useful when it informs page improvements.
• Compare engagement and conversions carefully. Review sessions, key events and conversion paths alongside other channels, because differences can reflect audience intent, tracking limits or attribution rules rather than a simple channel effect.
• Review assisted paths as well as last-click reports. Look for cases where an identifiable AI referral is followed by another visit or channel, because a final conversion source may not show every earlier interaction.
• Document consent and privacy settings. Confirm that your analytics setup follows your organisation’s policies and applicable requirements, because measurement should not override privacy obligations.
• Check for unattributed and direct traffic. Do not label all unexplained visits as AI-driven, because referrer data can be missing for many reasons.
• Keep referral reporting separate from prompt visibility reporting. A visit shows that someone reached the site; a prompt test shows what an engine returned, because neither measure can stand in for the other.
• Set a regular review cadence and owner. Decide who checks the reports and how often, because data that no one reviews does not create a feedback loop.

Avoid assuming that AI referrals convert at a premium unless your own data supports that conclusion. Compare like with like, use a clear conversion definition and be careful with low volumes. Referral data can show observable visits, but it cannot capture every effect of an answer that a person saw and remembered without clicking.

Turn findings into a prioritised action plan

Once the checklist is complete, group findings by risk and effort. Fix access blockers before polishing content, correct material identity and accuracy problems before adding new claims, and prioritise page improvements around important customer questions. A minor schema warning should not automatically outrank a page that returns an error or gives buyers no direct answer.

Use a simple action register:

For each issue, write a recommendation that can be tested. “Improve AEO” is not actionable. “Add a clear explanation of supported integrations to the product page, confirm it matches the current product, and rerun the relevant prompts” names the change and the verification step.

Prioritise a finding when it affects an important customer question, blocks access to a valuable page, creates a material factual error or appears repeatedly across your prompt set. Consider effort and dependencies too. A page rewrite may need product input; a crawl fix may need a developer; an inaccurate directory entry may depend on an external platform. Record those dependencies instead of letting them disappear from the plan.

For agency teams, share the evidence and decision with the client before making changes that affect crawl policy, claims or public profiles. For in-house teams, include the people who own the website, product facts, analytics and brand information. An AEO audit crosses these responsibilities, so a recommendation without a clear owner often remains an observation rather than a fix.

The next step after your AEO audit

Work through the six tiers, save evidence for every result and assign the highest-priority fixes to named owners. Then rerun the same technical checks and prompt set after changes, so you can distinguish completed work from measurable change. If you need a platform to support repeatable audits, tracking and action planning, review SearchFIT pricing and choose a workflow that fits your team.