AEO / GEO audit checklist

The agent-legibility audit, given away. Work through it by layer: read, cite, act, transact, measure. Every box you cannot tick is a place AI is routing around you.

Score your site with the scorecard

Layer 1

Read: can a machine parse you cleanly?

Core content is in real HTML text, not trapped in images, canvas, or script-rendered blocks an agent cannot read.
Pages use a clear heading hierarchy that mirrors the content's structure.
Structured data is present: at minimum an Organization schema with sameAs links to your profiles.
Page-type schema is in place where relevant: Product, Service, FAQPage, Article, BreadcrumbList.
A clean markdown version of each page is available to agents, via a .md URL or content negotiation.
An llms.txt file exists and points agents at your most important pages.
robots.txt allows the AI crawlers you want (GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot, Google-Extended).
Pages load fast and do not depend on heavy client-side rendering to show their content.

Layer 2

Cite: will a model trust and name you?

Your name, category, and core claims are identical across your site, your profiles, and third-party mentions.
Key facts (what you do, who you serve, pricing where relevant) are stated plainly and consistently.
Sections lead with the answer, then support it, so a model can lift the claim and its evidence together.
Content includes original information: data, first-hand testing, named methodology, specific numbers.
Questions your buyers actually ask are used as headings, with concise answers beneath. See AEO.
Claims are specific and defensible, not vague or hedged into uselessness.
You earn genuine mentions on reputable, widely-read sources over time.

Layer 3

Act: can an agent use your site?

Key actions (search, get a quote, check availability) are reachable without fighting a complex UI.
Forms have clear labels and structure an agent can complete.
You have considered or shipped a Model Context Protocol interface for your highest-value actions.
Critical actions do not hide behind unlabeled buttons, modals, or steps that assume a human is clicking.

Layer 4

Transact: can an agent buy or book?

For ecommerce: product data, pricing, and availability are machine-readable and current, with Product and Offer schema.
Shipping, returns, and warranty terms are readable as data, not buried in a PDF or image.
You are tracking the emerging commerce protocols (ACP, AP2) covered in agentic commerce.
For services: availability and booking are exposed in a way an agent could act on, not just a human.

Layer 5

Measure: can you prove it works?

You have written down the prompts your buyers actually use to find someone like you.
You check whether AI assistants cite you for those prompts, on a regular cadence. Use the AI visibility tracker.
You track share of voice against named competitors, not just your own mentions.
You can trace some traffic or revenue back to AI sources. See AI visibility.

Want a number instead of a list? Run the agent readiness scorecard.

Using the checklist

How do I use this checklist?

Work through it layer by layer against your own site. Read and cite are where most brands have gaps, so start there. Each item is a yes or no you can check by viewing your source, your schema, and your presence in AI answers. Anything you cannot answer yes to is a task.

Do I need technical skills to act on it?

Some items are content work anyone can do, like writing question-first answers. Others, like adding structured data or markdown versions of pages, need a developer or a capable CMS. The checklist tells you what to fix. How you fix it depends on your stack.

Is this the same audit you run for clients?

It is the backbone of it. We give the framework away because the value is in the doing, not the list. A deeper engagement adds your specific prompt set, competitor share-of-voice, and implementation. But you can get a long way with this on your own.

Get a score with the scorecard