page.ai.analyzePageText()

Ask a free-form question about the visible text on the current page and receive a natural-language answer.

Asks a free-form question about the visible text on the current page and returns a natural-language answer. Useful for sanity checks, content audits, and verification tasks where you want a human-readable report rather than structured data.

Signature

page.ai.analyzePageText(
  analysisToRun: string,
  options?: {
    additionalContext?: string;
    gptClient?: GptClient | LanguageModel;
  }
): Promise<string>

Parameters

Parameter	Type	Default	Description
`analysisToRun`	`string`	—	Plain-English description of the analysis to perform
`options.additionalContext`	`string`	—	Any context relevant to the analysis that is not present on the page itself (e.g. "This is a developer changelog page")
`options.gptClient`	`GptClient \| LanguageModel`	Project default	Override the AI provider for this call

Returns

A string containing the AI's natural-language analysis. You can log it, attach it to the test report, or feed it into subsequent assertions.

Usage

import { test } from 'donobu';

test('changelog contains no deprecated API warnings', async ({ page }) => {
  await page.goto('https://docs.example.com/changelog');

  const report = await page.ai.analyzePageText(
    'List any mentions of deprecated APIs, breaking changes, or security warnings you find',
    { additionalContext: 'This is a developer-facing changelog for a REST API.' },
  );

  console.log(report);

  // Use the report as part of a structured assertion if needed
  expect(report.toLowerCase()).not.toContain('critical security');
});

Practical use cases

Content verification: confirm that key messaging or legal copy appears correctly
Deprecation audits: scan changelogs or release notes for potentially impactful changes
Accessibility spot-checks: ask whether the page contains any placeholder text or lorem ipsum that slipped into production
Localisation review: verify that no untranslated strings are present on a localised page

Notes

The analysis runs on the raw extracted text of the page, not the rendered HTML. Layout, visual hierarchy, and formatting are lost — results may be less accurate for pages where position and styling carry meaning. For structured extraction, prefer page.ai.extract.