A useful audit connects a reproducible observation to an affected task and a practical verification step.
Agree the website, sample and task
Confirm permission to assess the website, then choose a question the audit should answer. Can a reader identify when an article was published and who wrote it? Can a customer tell which product variants the listed price applies to? Choose pages and templates that help you answer the question.
Record the domain, included URLs or sampling method, exclusions and access conditions. Count assessed resources separately from report pages. Note whether you checked the initial response, a rendered browser view or both. Describe the findings within the limits of that sample.
Collect evidence before assigning a score
For each resource, inspect retrieval, status and useful content. Compare the initial HTML with the rendered page, then review the document outline, landmarks, links, tables and metadata that carry meaning. Check whether structured data agrees with visible information.
Record the time, URL, retrieval method and relevant location or excerpt with the evidence. A network error, a deliberate access restriction and missing HTML content require different explanations. If you couldn't measure something, record it as untested and explain why.
- Which response did the tested client receive?
- What essential information was present or absent?
- Which pages or templates are supported by the evidence?
- Which checks were not performed?
Try the HTML delivery example
These two example pages describe the same fictional service. The first adds the service details with JavaScript. The second includes them in its initial HTML.
- Scope: two local HTML files and one shared service-content script. No external requests, account, consent state or payment flow.
- Question: is the service description available in the initial main content without executing JavaScript?
- Check version: html-delivery-v1. The supplied verifier reads the files and executes the example script against a minimal document stub.
| Check | Before | After |
|---|---|---|
| Service heading in the initial main content | Present | Present |
| Service description in the initial main content | Absent | Present |
| Consultation next step in the initial main content | Absent | Present |
Repeat the check
Save all four files in the same folder and run the verifier with Node.js. It confirms that the first page's script can insert the missing details and that the second page already contains them. For a separate visual check, open both HTML files in a browser, then disable JavaScript and reload.
node verify.mjsRecommended change
Deliver the essential description and next step in the initial document, while keeping JavaScript for enhancement. The after page demonstrates that change for this fixture.
This check confirms the delivery difference in the local files. HTTP responses, browser rendering and accessibility need separate checks on the authorised website.
Separate findings, severity and dimension scores
A finding documents an issue and the evidence for it. Its severity tells you how much the issue matters within the agreed scope. A dimension score summarises a group of checks, so a high score can still leave an important finding to address.
Read coverage, thresholds and limitations alongside the numbers. Before comparing scores, check which pages and checks each assessment covered. When a score is low, read the underlying findings and recommendations to decide what to fix.
Make the recommendation testable
Describe what should change and how the team will know it worked. “Improve agent readiness” is too vague. “Include the service description under its heading in the initial HTML, then confirm it is available without JavaScript” gives a developer something concrete to implement and verify.
Prioritise by the affected task and the reach of the template. Correcting a delivery problem across a shared template may matter more than adding another discovery file. Record which templates each recommendation affects so the team can estimate the work.
From finding to task
- Observation
- It's unclear which product heading goes with each value in a comparison table.
- Next step
- Use appropriate header cells and associations, preserve the visual layout, and verify that each value can be interpreted with its row and column context.
Review, deliver and follow up
Tityra requests an agreed website scope and an authority declaration before work is released. Individually requested reports receive staff quality review before release. Included weekly AX Audit reports are delivered automatically after the required checks. Expert Review adds deeper investigation within an agreed scope. Our AX audits check against the AgentReady essentials where they apply to that scope.
Use the report to plan changes, discuss evidence and track implementation. Marking a finding resolved records your progress in the workspace. If you arrange a follow-up assessment, record the new date, sample and methods so you can compare what was tested.
