What Does an AEO Audit Check? The Seven-Point Checklist, and What to Fix First

An AEO audit checks seven things: whether AI crawlers are allowed in, whether your pages can be indexed and quoted, whether your brand facts are clear and consistent, whether pages lead with a direct answer, whether structured data describes the page truthfully, whether other sites back up what you say, and whether you are measuring any of it. The order matters, because a failure early in the list cancels out everything after it.

The table below is the whole audit on one screen. Each row corresponds to a section with the detailed check. If you want a quick outside view before you start, the free scan in the AEO Grader tool gives you a baseline score in about a minute, and our technical AEO checklist goes deeper on speed and markup once you have the basics confirmed.

Start with robots.txt, because a single line there can remove you from an engine completely. OpenAI's crawler documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, and adds that it can take about 24 hours from a robots.txt update for its systems to adjust. Perplexity's crawler page describes PerplexityBot as the agent that surfaces and links websites in its search results. Anthropic's help article warns that disabling Claude-SearchBot may reduce your visibility in Claude's search results.

Not every AI user agent carries the same stakes. Google's list of common crawlers says Google-Extended does not affect a site's inclusion in Google Search; it governs whether crawled content may be used for Gemini training and grounding. Blocking it is a business decision, not an accident to fix by reflex.

Then look past robots.txt. Google's guidance on AI features lists ensuring that crawling is allowed in robots.txt and by any CDN or hosting infrastructure. Bot-protection rules at the CDN often return a 403 to unfamiliar agents while the site looks fine in a browser. Check your firewall logs for blocked requests from the search crawlers above. Severity: critical, because nothing else on this list matters to an engine that cannot load the page.

Being crawlable is not the same as being usable. For Google, the bar is explicit: the AI features page says a page must be indexed and eligible to be shown in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements. Two settings quietly break that eligibility.

The first is noindex, which keeps a page out of results altogether. The second is easier to miss. Google's robots meta tag documentation says the nosnippet rule applies to AI Overviews and AI Mode and also prevents the content from being used as a direct input for them. A max-snippet value of 0 is equivalent to nosnippet, and a data-nosnippet attribute can hide individual sections. These tags often arrive through a plugin setting or a legal team's request and then outlive the reason for them.

To check, run your most important URLs through the URL Inspection tool in Search Console, which shows Google's indexed version of a page and lets you test whether a URL might be indexable. Then search your templates for nosnippet and max-snippet. Severity: critical for the pages affected, harmless elsewhere.

Answer engines have to decide which entity a name refers to before they can say anything useful about it. If your brand shares a name with a better-known company, or your own pages describe you three different ways, the engine either picks the wrong entity or hedges. This check is about removing that doubt.

Write one canonical fact sheet: legal and trading name, one-sentence description, category, location or service area, founding year, and the URLs of your official profiles. Then compare it, line by line, with your homepage, About page, footer, social bios and directory listings. Any mismatch is a finding.

Back the fact sheet with markup. Google's Organization structured data guide says the markup helps Google understand your administrative details and disambiguate your organization in search results, and that there are no required properties. Include the sameAs property pointing to your official profiles. Severity: high, because confusion here produces wrong answers rather than missing ones, and a wrong answer about your pricing or location is worse than silence. For the longer treatment of entities, see our entity SEO guide; for a shop, clinic or trade business, the same check means matching name, address and phone everywhere, as covered in how local businesses show up in AI search.

Engines quote sentences, not pages. Open your ten most valuable pages and read only the first sentence under each heading. If that sentence does not answer the heading on its own, an engine has to assemble the answer from fragments, and a competitor with a cleaner sentence is easier to cite. Phrase headings as the questions buyers actually ask, answer directly underneath, then add the detail. Thin pages fail this check by default because there is nothing worth quoting.

Structured data comes next, and the test is honesty rather than volume. Google's AI features guidance asks you to make sure your structured data matches the visible text on the page. FAQ markup on a page with no visible questions, or Product markup with a price that differs from the page, is a liability. Validate each template and compare the JSON-LD with what a visitor sees.

One optional item belongs here too. The llms.txt proposal suggests a markdown file at the site root to help agents use a website. It is a proposal rather than a standard the major engines have committed to, so treat it as low priority: cheap to add, and never a substitute for checks 1 to 3. Severity for this block: medium to high, rising with how commercially important the page is.

Engines rarely trust a brand's own pages alone. When they answer a buyer question, they lean on review sites, directories, comparison articles, forums and news coverage. If those sources describe you inaccurately, or do not mention you at all, your own site cannot fully compensate.

You audit this from the outside. Ask each engine five to ten questions a buyer would ask, such as what you do, whether you are trustworthy, and who the best options in your category are. Write down every domain each engine cites. Then sort the list into three groups: sources that mention you correctly, sources that mention you incorrectly, and sources that cite competitors but not you. The second group is the quickest win, because correcting a listing you already have is easier than earning a new one. The third group is your outreach list.

Severity: high, but slow to change, which is why it sits after the technical checks in the fix order. Our glossary entry on cross-source corroboration explains why agreement across independent sites carries so much weight.

An audit without a baseline cannot prove that its fixes worked. Before you change anything, record what the engines say today, with the date, the exact question and the sources cited.

Then switch on the first-party data that is free. Google announced Search Generative AI performance reports in Search Console on June 3, 2026, showing impressions within AI Overviews and AI Mode by page, country, device and date, and the post notes that as of August 31, 2026 the reports had rolled out to all websites worldwide. Microsoft's AI Performance report in Bing Webmaster Tools shows how often your content is cited in generative answers across Copilot and Bing's AI summaries.

Neither report covers ChatGPT, Claude or Perplexity, so you still need a repeatable set of questions for those. We lay out a monthly routine in how to track AI visibility over time. Severity: low on the day of the audit, high three months later when someone asks whether the work paid off.

Fix in the order an engine meets your site: access, then eligibility, then identity, then content, then corroboration. A higher stage that fails makes every lower-stage fix invisible, so effort spent on copy while a firewall blocks the crawler is wasted. Within one stage, pick the item with the biggest impact for the least effort. Our step-by-step AEO process shows the same sequence as a project plan, and the AI Overviews guide covers the Google-specific eligibility details.

The free scan in the AEO Grader tool does not crawl your site. It runs one grounded Gemini and Google Search check, estimates technical readiness from what the web says about you, and returns an AEO score with a four-pillar breakdown and an action plan with code snippets. The crawl-based site audit is part of the $10 full report. Here is exactly what that audit does.

It fetches your homepage and up to four internal pages, preferring paths such as /about, /pricing, /faq, /blog and /services, plus your robots.txt, llms.txt and sitemap.xml. It reads robots.txt for GPTBot, Google-Extended, PerplexityBot, ClaudeBot and OAI-SearchBot and fails the access check if any of them is disallowed from the whole site. It then runs fourteen pass, warn or fail checks: Organization, Brand or LocalBusiness JSON-LD; FAQ or Q&A content; question-style headings; Article, Product or Service schema; llms.txt; AI crawler access; title length; meta description length; a single H1; a canonical link; a sitemap; robots.txt; average content depth; and Open Graph tags. A pass earns full credit, a warning half, a failure none, for a 0 to 100 technical score.

The findings become a prioritized fix list, combined with what the four engines said in the same report, such as which competitors they named and which domains they cited. Every item carries a severity, an impact and an effort rating, a reason and a concrete change. If the fix list cannot be generated, a rule-based fallback ranks blocked AI crawlers and missing Organization markup as critical. If our crawler cannot load your site at all, the report gives no technical score and no on-page advice, because it never saw the pages; a 401, 403 or 429 response produces one action telling you to check firewall and CDN rules for AI crawlers.

Be clear about the gaps: the audit does not read meta robots tags, so checks 2 and 3 above stay manual, and it does not grade page speed. The report includes one re-scan within 14 days, so you can confirm a fix registered. Methodology is on the methodology page. Nothing here guarantees a citation; it tells you what to fix and in what order.