Back to docs

Tool Page Understand

Category: Site Knowledge · Since: v18.35.0 · Tier: All

Analyze the current page's type, actions, form structure, and navigation. Returns a page model with affordances, structure, and element relationships.

Parameters

Parameter Type Required Default Description
_browserToken string No — Browser session token from a previous tool call

Examples

Analyze this page to understand what actions are available

What type of page is this and what can I do on it?

How It Works

  1. DOM Extraction — A single page.evaluate() traverses the DOM. It extracts interactive elements, headings, forms, nav landmarks, and semantic regions.
  2. Page Type Classification — Heuristic scoring classifies the page into one of 9 types: landing, form, dashboard, article, search, checkout, error, settings, or list.
  3. Affordance Mapping — Each interactive element maps to an action type (click, fill, select, submit, toggle). Each gets a predicted outcome and confidence score (0.5--0.9).
  4. Structure Analysis — Navigation separates from content via <nav> landmarks and position heuristics. Forms group by <fieldset> boundaries and label associations.
  5. Relationship Detection — Element relationships (contains, adjacent, same-form, same-group) are computed for structural connections.

Response

{
  "pageType": "form",
  "affordanceCount": 12,
  "formCount": 1,
  "ctaCount": 2,
  "topAffordances": [
    { "element": "#email", "action": "fill", "expectedOutcome": "Enter email address", "confidence": 0.9 },
    { "element": "#submit", "action": "click", "expectedOutcome": "Submit form", "confidence": 0.85 }
  ]
}

Page Types

Type Detection Signals
landing Hero section, CTA buttons, minimal forms
form 3+ input fields with submit button
dashboard Sidebar nav + multiple data sections
article Long text with heading hierarchy
search Search input + result list
checkout Payment/billing form fields
error Error codes, "not found" text
settings Toggle/select-heavy, "settings" in heading
list Repeated card/row structures

Performance

  • Full analysis: <500ms for pages with <500 elements
  • Lightweight skeleton: <50ms
  • Pages with >1000 elements: viewport-only analysis
  • Results cached in memory, invalidated on navigation

Research

Based on DOM-native intelligence that outperforms screenshot-based methods. See rtrvr.ai DOM Intelligence Architecture -- 81.39% on WebBench vs screenshot-based alternatives.

Related

From the Blog