Methodology
How we look at a page
Always the same 6 checks, in the same two windows. Here is what they are, what we leave out on purpose, and why there is no model doing the measuring.
What we open
We open the address in a real browser. We do not just request the code and read it: we open the page, let it load everything it needs, and look at what ends up on screen. It is slower, but it is how we know where each element actually ends up.
The only thing we click
Whatever closes the cookie notice. Nothing else. We do not fill in forms or submit anything. We do not buy anything either. We look at what is there on arrival.
When something repeats
We name the first three in the order they appear on the site, and say how many more there are.
The 6
Always the same ones. We do not pick them per page: if they changed each time, two reports of the same site could not be compared.
The minimum for an overlay to count
If it covers less than a seventh of the screen, we measure it but leave it out of the report. We would rather say nothing than tell you something you could not tell apart from nothing.
When a check gets cut short
It can stop before listing everything. When that happens, the sentence in the report says so.
What we do not look at, on purpose
This list matters as much as the one above.
- Whether the page is pretty, whether the copy convinces, or how the brand is doing. It is not that it does not matter: it is that it cannot be measured by opening the page, and if we said it we would be giving an opinion in the same tone we use to measure.
- What happens after you touch something. We open, measure and leave: we do not follow a checkout to the end or complete a signup to see what happens.
- Text that lives inside an image. If what matters is painted into a JPG, we see a JPG.
- Nothing that asks for a password, and no more than 11 pages per site.
- What is not declared. If something behaves like a button but the code does not say it is one, to us it is not a button. It may not be one to whoever arrives either. Same with forms: we only see the ones the page declares as such.
In what order
Which changes appear is decided by the goal you chose and a fixed order of checks. Their order comes from three things we measured: how much each change overlaps with that goal, whether we could check it fully or only in part, and on how many pages it appears.
They do not go by severity, and that is deliberate. What matters most depends on your business, on the time of year and on what you can change this week—and we have measured none of that. An order by severity would be an opinion dressed up as a result.
Get people to write or call me
- The forms
- The first screen and the buttons
- The links
- Contrast and alt text
Get people to buy or sign up
- The first screen and the buttons
- The forms
- The links
- Contrast and alt text
Make it clear what we do
- The first screen and the buttons
- Contrast and alt text
- The links
- How it looks when shared
Make it usable by anyone
- Contrast and alt text
- The links
- The first screen and the buttons
- The forms
A general review
- All 6, with no emphasis on any one
Map version: objetivos.v5 · the same one used to seal the report.
Why there is no model doing the measuring
I gave the page to a model and the model wrote the report. It came out well. Well enough for me to believe it.
The writing was not the problem. The problem was that there was no way to know what it had looked at.
So I separated the two things. The sentences are written beforehand, one by one, for each thing the instrument knows how to check: how it is said, what measurement backs it, and what cannot be claimed with it. Then what we measure on the page decides which ones appear and with what numbers.
A laboratory does exactly this. It does not write a new text for each patient: it writes once what each value means, and then measures yours.
AI helps build the instrument. It does not do the measuring.
We lost some fluency, and the model's habit of covering whatever occurred to it: this finds less. In exchange, it measures something text cannot tell you: where each thing ends up when the page opens.
Start with an address
You can run three analyses without creating an account and without a card. The report has written on it what was looked at, in which window, and how long it will be there.
No account needed.