Guide · AEO Audit
By Vlad Radu · Published 2026-07-22 · Updated 2026-07-22
An AEO audit is a structured review of whether AI answer engines — ChatGPT, Perplexity, Gemini, and Google AI Overviews — can read, understand, and cite a page. It checks three layers: technical (rendering, schema, crawlability), content (lead answers, definitions, extractability), and authority (author signals, citations, entity consistency).
An AEO audit answers one question: if an LLM crawler fetched this page right now, would it be able to extract a citable answer? Classic SEO audits score keyword coverage, backlinks, and Core Web Vitals. An AEO audit scores the machine-readable signals that determine whether SearchGPT or Perplexity quotes you at all — see our AEO vs SEO comparison for the full contrast.
Server-side rendering (the H1 and body must appear in raw HTML), schema.org coverage across Organization, WebSite, WebPage, Article, FAQPage, HowTo, and Product where relevant, crawlability for GPTBot / PerplexityBot / Google-Extended / ClaudeBot in robots.txt, and a valid /llms.txt at the domain root.
A 40–60 word direct-answer paragraph at the top, explicit definitions of key terms, one H1 with a logical H2/H3 outline, and extractable units (numbered lists, comparison tables, stat-dense sentences) that LLMs can lift verbatim.
A named author with Person schema and sameAs links (LinkedIn, GitHub, or a recognized publication), outbound citations to authoritative primary sources, and consistent Organization / entity signals across every page of the site.
Work through these in order. Each step names what to check and what a pass looks like.
View source (Ctrl+U) and search for your H1 and main body copy. If they aren't in the raw HTML, LLM crawlers won't see them. Pass: the H1, first paragraph, and every schema block appear in the raw response before any JavaScript runs.
Paste the URL into Google's Rich Results Test and Schema.org validator. Pass: at least Organization, WebSite, WebPage, and Article (or Product/FAQPage where relevant) parse with zero errors and zero warnings on required fields.
Open /robots.txt and confirm GPTBot, PerplexityBot, Google-Extended, and ClaudeBot are not disallowed. Pass: no blanket Disallow, and any per-bot rules are intentional.
Fetch /llms.txt at the domain root. Pass: the file exists, opens with a one-line site summary, and lists your highest-value canonical URLs grouped by section. See our llms.txt guide for the exact format.
Read the first 40–60 words of the page. Pass: they answer the page's primary question in a single self-contained paragraph, with named entities, numbers, and no marketing throat-clearing.
Confirm one H1, a logical H2/H3 outline, and explicit definitions for any acronym or key term (ideally in dl/dt/dd or a bold lead sentence). Pass: a screen reader outline reads like a table of contents.
Look for quotable units: numbered lists, comparison tables, short definitions, and stat-dense sentences. Pass: at least one extractable block per H2 and one original statistic per ~500 words.
Confirm a named author byline with a linked bio, Person schema with sameAs (LinkedIn, GitHub, or a recognized publication), and consistent Organization schema across pages. Pass: an AI can resolve the author and publisher to a single verifiable entity.
Check whether the page cites primary sources (docs, standards, research) and whether your brand is mentioned on third-party sites that LLMs index. Pass: at least 2–3 outbound citations to authoritative sources and a handful of external mentions on the open web.
Ask ChatGPT, Perplexity, and Gemini a query your page is meant to answer. Pass: your page appears as a cited source, or the answer paraphrases content only your page contains. See our guide on checking if ChatGPT recommends your site.
Manual audits are irreplaceable for judgment calls: is the lead answer actually clear, is the author credible, are the citations authoritative? But the deterministic checks — rendering mode, schema coverage, heading structure, llms.txt, crawlability — are tedious to run by hand and easy to score inconsistently across pages.
aeo-audit.app runs the technical layer of this checklist automatically in under a minute: it fetches the raw and hydrated HTML, scores schema completeness across 14 types, checks llms.txt, and returns a readiness score with the specific blockers to fix. Use it for the deterministic pass, then run manual review on the content and authority findings.
A thorough manual audit of a single page takes 30–60 minutes: 10 minutes on rendering and schema, 15–20 on content and extractability, and the rest on authority signals and live citation testing. Automated tools like aeo-audit.app compress this to under a minute for the technical layer.
Both. Use an automated tool for the deterministic checks — rendering mode, schema coverage, llms.txt, heading structure — because they're tedious and easy to get wrong by hand. Reserve manual review for the judgment calls: is the lead answer actually good, is the author bio credible, are the citations authoritative.
Fix technical blockers first — a client-rendered page or missing schema makes every content fix invisible to LLM crawlers. Then rewrite the lead answer and heading structure. Authority signals (author bios, sameAs entries, external citations) come last because they compound over weeks, not hours.
Monthly for high-velocity pages (product, pricing, homepage), quarterly for evergreen guides, and ad-hoc after any major LLM retrieval change — a new crawler, a new schema type gaining support, or a visible drop in AI referral traffic.
Next step
Get a readiness score, schema coverage, llms.txt check, and prioritized blockers — free for the first URL.
Run free audit →