Back to all steps

Page perception

Step fields
question for AI (optional) (text)
Visual AI check of the current page. The worker takes a full-page screenshot and sends it to AI vision. If the field is left empty, the question is whether the page is visually fine (is the font readable, is the layout intact, did the styles and images load). If you enter your own question, the AI checks that instead (e.g. «did the video open»). If the AI answers that something is wrong with the page (unstyled raw HTML, overlapping/cut-off text, broken-image icons, an empty/white screen) the test fails and the model’s answer becomes the reason. If the AI answers that the page is fine, the step is considered passed and the test continues. It spends the test owner’s AI tokens: the step only runs AI vision when the balance is at least 5000 tokens; below that the check is silently skipped and the balance never goes negative. In preview/recorder it is a stub: the step is marked done, no AI call is made and no tokens are spent.
Usage example

Visual AI check of the current page. The worker takes a full-page screenshot and sends it to AI vision with a single question: is the page visually fine. No parameters.

What the AI looks at:

  • is the font readable — text not overlapping, cut off or garbled;
  • is the layout intact — blocks in place, nothing shifted apart;
  • did styles and images load — no unstyled raw HTML, no broken-image icons, no empty/white screen, no raw error text.

Step outcome:

  • the AI answers that something is wrong with the page → the test fails, and the model’s short explanation becomes the reason;
  • the AI answers that the page is fine → the step is passed and the test continues.

Spends AI tokens. The step only runs AI vision when the test owner’s balance is at least 5000 tokens (threshold configurable). Below that the check is silently skipped — the test does not fail and the balance never goes negative.

In preview / recorder it is a stub: the step is marked done, no AI call is made and no tokens are spent.

A typical combo — check the page after a navigation:

  1. Go to page — open the page
  2. Page perception