Technical AEO: a repeatable access and rendering audit

Inspect the response before the screenshot

Request a representative public service page and guide without a login. Record requested URL, final URL, redirects, status, content type, timestamp and relevant indexing headers. A visually attractive page returning a server error needs investigation. A branded “not found” message with a 200 response is not a real 404. Check an intentionally unknown path alongside the known page, and preserve the distinction in the audit.

Compare initial HTML with the rendered page

Save the response body before JavaScript runs, then compare the main explanation, headings and ordinary links with the browser DOM. A client-rendered skeleton may omit important source material even when the browser eventually displays it. Server rendering is not a universal guarantee either: content can still be hidden behind a consent gate, account requirement or broken data dependency. Test the pages and templates your buyers use rather than assuming the whole CMS behaves identically.

Resolve access policy by purpose

Review robots rules for the precise path and documented crawler role. OpenAI distinguishes OAI-SearchBot for search discovery, GPTBot for training and ChatGPT-User for certain user-initiated actions. Google Search controls and Google-Extended have different purposes. A business can choose a training policy independently from search discovery. This guide does not ask you to reverse a training opt-out to gain citations.

Inspect CDN behaviour separately

A robots allowance does not bypass a firewall challenge. Compare observed responses with authorised CDN configuration and, when available, verified crawler log evidence. A user-agent string can be spoofed; a test that merely sets that string cannot verify the requester’s identity. Scope fixes to legitimate public routes and retain authentication for private areas. Do not disable security globally to make one test turn green.

Check page identity and discoverability

Inspect self-canonical, indexing and snippet directives, relevant language alternates, sitemap entry and at least one ordinary internal path. A correct sitemap does not repair an inaccessible page or guarantee indexing. Validate meaningful headings and accurate structured data that matches visible content. There is no special AEO schema required for Google AI features; an Organization or Article label cannot legitimise invented business facts.

Add performance and interaction evidence

Use field Core Web Vitals when an authorised source has sufficient observations; use a labelled lab run for diagnostic work. Do not call a lab observation real-user performance. Test tables, navigation, locale switching, keyboard focus and the contact form’s validation and receipt behaviour. Record device, tool and date so a later reviewer can reproduce a problem. A fast page with a broken enquiry path still fails the customer’s task.

Write a finding another person can fix

Every row needs observed issue, likely impact, supporting response or screenshot, proposed change, owner, risk and post-change test. For an illustrative rendering failure, specify which service paragraph is absent from initial HTML and which template supplies it. After the change, save both the corrected HTML and a negative control confirming private routes remain protected. Close the access issue only after verification; keep indexing and AI-answer observations in separate records.

Download the blank CSV worksheet

CSV opens in spreadsheet applications. Keep evidence locally or in your authorised workspace; no answers are submitted by this download.

Continue with the related decision

• AEO learning paths
• AI crawler policies: search, training and verified log evidence
• Discuss the scope of AEO and AI-search work