AI Website Assessment

What the AI does, and what it does not

Fixed-rule findings are recorded as pages are visited, so they exist even if every model call fails. The Basic assessment uses no AI at all. Intermediate adds the AI work above, and Advanced adds a second model that reviews the findings and can disagree with the first.

What the model sees

• It sees page content: the text and structure of the pages being tested, and screenshots if the model you chose can read images.
• It does not see your password. The model works with a placeholder. Provelex fills in the real value itself, and only into a password field.
• It may see what your application displays after signing in. So use a test account and test data wherever you can.
• The Privacy Policy names the AI providers and where data can go.

Which models can be used

DeepSeek is the default and the cheapest. You can choose Kimi, ChatGPT or Anthropic instead, or a pair that check each other. The run screen shows the price of your selection and the difference from the default. If a provider is unavailable, a request may be routed through OpenRouter.

Where human review remains important

• Business rules. A model does not know that your discount should never exceed 20%.
• Risk. Whether a security finding matters depends on your data, your users and your obligations.
• Accessibility. Whether a page actually works with a screen reader needs people.
• Critical journeys. Payments, sign-up and anything involving money or safety deserve a person's attention as well.