Apply the requirements that match the website’s intended use; distinguish observations, limitations and checks that do not apply.
A framework for the intended experience
AgentReady describes an agent’s path as find, read and act. Its essentials distinguish being able to discover and read a site from completing actions through a product. An editorial website and a service that lets agents transact do not have identical requirements.
We use that distinction when agreeing an AX Audit. The report should identify the relevant resources and checks, record what was actually tested, and explain exclusions. Checking against the AgentReady essentials is not a claim that Tityra certifies a site, declares compliance or awards an AgentReady badge.
Find: discovery and access
We review available first-party guidance, links, sitemap entries and the stated crawl policy for the agreed sample. Retrieval checks examine whether the tested client receives the expected public resource or encounters an error, redirect or challenge.
These checks can collect status codes, headers and discovery-file contents automatically. Interpreting a restriction needs context: a private customer area should not be treated as a public page that must be opened. An expert may need to compare intended access with CDN or application behaviour before recommending a change.
- Declared discovery and crawl policy
- Actual responses for sampled public resources
- Redirect destinations and missing-page behaviour
- First-party guidance linked to the relevant resource
Read: content and meaning
We compare the initial HTML with the rendered page and inspect the structure that describes the main information. Checks can include headings, landmarks, links, tables and structured data, along with available authorship, date, source and usage context.
Automation can identify missing fields, parse errors and differences between views. It cannot reliably decide whether an answer is complete, a usage statement is appropriate or a page serves its intended audience. Those questions need review against the actual content and task. Automated accessibility results also do not establish full accessibility conformance.
Act: only where the product supports it
If agents are intended to complete a task, the assessment may examine documented interfaces, schemas, authentication boundaries, permissions and recovery behaviour within the agreed access. A website that only publishes information should not be told to invent a payment API or authorization server to improve a scanner score.
Where an action exists, check its real boundary. Can the agent distinguish preparing a request from submitting it? Does the owner still provide the required approval? Are errors and retry instructions usable? Production changes, purchases and private-data operations require the appropriate authorisation; an audit does not grant it.
What the report should make clear
Read each observation with its affected resources, method, date and evidence. Separate a confirmed failure from an inference, an unavailable measurement and a requirement that is outside scope. Score summaries should reflect the checks supplied, rather than imply that untested items passed.
AgentReady evolves. A report should identify the assessment date and the framework or checks used so later comparisons have a meaningful basis. Tityra’s methodology explains review and delivery; any deeper investigation or retest should have its own agreed scope.
Sources and further reading
See what your website delivers.
Choose an agreed scope and get findings your team can act on.
