Can an AI agent tell what my buttons and links do?
A program that reads your page finds a control by its name. A button that says "Add to basket" can be found and used. An icon with no label, or a clickable box that is not really a button, gives it nothing to go on.
What it actually is
Every button, link and form field on a page can carry a name: the words inside it, or a label attached to it. People see that name as the text or the icon. A program reading the page sees only what is written into the markup, and the name is how it knows one control from another. We look at the HTML your server sends and count the controls that have no name, and the clickable things that are not controls at all, such as a plain box wired up to behave like a button.
Why it matters
More of the web is now used by software acting for a person: a browser agent asked to book, buy or fill in a form, and assistive technology that reads a page aloud. Both work from the same description of the page, built from the roles and names of its controls. A control with no name is a door with no sign on it. The agent either guesses, or does not use it, and the task it was asked to do fails on your page and succeeds on a competitor's.
What good looks like
Real buttons and real links, each with a name that says what it does: words in the control, a label on the field, a described icon. Anything a person can press is something a keyboard can reach too. The names say what the control does, not just what it looks like.
Being honest about it
The evidence here is for agents that act on a page, not for crawlers that only read it, and most of it comes from vendors' own tests, so it counts for only a small part of the score. We can only see the HTML as sent: a control that script builds afterwards, or a handler attached by script, is invisible to us, so a clean result is not a guarantee. We also check that a name exists, not that it is the right one, and a wrong name is worse than a missing one.
Where most sites go wrong
Icon-only buttons are the usual cause: a menu, a search, a close, a cart, drawn as a picture with no words. The second is the clickable box, a styled block wired up with a script where a button should be, which a keyboard cannot reach and an agent does not recognise as a control. The third is the form field that relies on grey placeholder text, or a label that sits beside the field without being attached to it.
On the web
In October 2026, 53% of the sites we scanned passed the “Buttons, links and fields have names” check.
| Result | Share of sites |
|---|---|
| Pass | 53% |
| Worth a look | 44% |
| To fix | 3% |
| Couldn't judge | 0% |
Next.js sites pass it most often, at 75%. Webflow sites pass it least often, at 31%.
Goes with
- Sites that pass this are 5 times as likely to pass the “AI catalog” check too.
- Sites that pass this are 3.5 times as likely to pass the “API catalog” check too.
Not enough history yet.
Based on the sites Bot Appétit has scanned: ones people chose to scan, plus lists we scan for research. Not a random sample of the web.
Not sure where yours stands?
A free scan checks this and everything else in about 20 seconds — no signup. Or if it's already flagged and you'd rather it were simply handled, we do that too.
Flagged for this one? We handle it. No pitch, just a plain-English chat.
Book a call