01Each result can be checked against the page itself.
A reading records how AI is likely to read a page or document on the day it is visited: what came back when it was asked for, what it says about itself, and how each check came out. The evidence is kept with each result, taken from the page as it was delivered, so your team can open the page and see the same thing. In WORK|PACK, results across the estate are set in order of priority; Diagnostics is the detail behind each one.
02Start from what AI was actually given.
Before any check, the reading shows what came back when the page or document was asked for, and what it says about itself. Often, that is where the problem starts.
A page that never says when it was written.
Served in full in 93 milliseconds, with a title and description. Nothing on the page states a date, so AI cannot tell whether it is current or years old.
Look for Last changed: Nothing on the page says.
A well-made document AI never reaches.
Built well enough to measure 66.0 out of 100, it counts for 9.9, because AI is turned away before it reaches the file.
An example - this is made up for th purpose of illustration.
Underlined in red: AI is turned away before it reaches this document.
03Each result points to the exact item.
Each check states what it looks for, what good looks like, why it matters for AI, what it found and what to do, then lists the items behind the result.
A question in the page’s headings with no answer under it.
Of the 9 questions the page asks in its own headings, 1 has nothing written beneath it. AI picks up the question and repeats it; the answer is not there to go with it.
Boxed in red: Accessibility help?
Links that answer with an error.
Of the 120 addresses linked from these pages, 10 return an error. Whatever those pages said is missing from what AI knows of you.
Look for 404, Not there.
Words that come out jumbled.
In 10 lines of this document the letters run together, so what AI reads from it is garbled rather than its actual words. Broken sentences get quoted as though they were what you wrote, with your name on them.
Look for The text, first 500 characters: the table of contents with its words run together.
Nothing is perfect, and no reading is. Each result is what AI thinks and is seeing on the day of the visit; where your team finds the page says otherwise, the evidence is there to raise it with us.
04Where a fix can be written down, the steps come with it.
The steps, written for the person who makes the change.
This PDF has no title in its own properties, so AI knows it by its file name. The result says what to do, then gives the web team the steps: where to go, what to set, and how to tell it is done.
Outlined in teal: the steps for the web team.
→Arrowed in red: no title and no author recorded.
Where a fix depends on your own systems, the result states what to change and the evidence to work from, and we work through it with your team.
05When the change is made, it is read again.
The page or document is read again after the change. The result moves when AI is given something different, so the record shows what was found, what was changed and the reading that confirms AI now has it. For directors and officers, it is the record that the exposure was found and dealt with. Each correction changes what AI is given to read; WORK|PACK sets the order of the work.