Web Research to Stated Workflow

Build sourced web-research workflows that turn known URLs or broader research questions into persistent Stated Pages for human review and action.

2026-09-15 · 2 min read

Use Stated research capabilities when an automation needs to turn public web information into a human-facing deliverable.

Choose the retrieval method

Use the Firecrawl-backed scraping workflow when you know the public URL to fetch. Use the Parallel-backed research workflow when the task is a broader research question that requires sourced findings across the web.

Do not use a broad research run merely to retrieve one known page, and do not treat one scraped page as comprehensive research.

Known-page workflow

Stated can fetch a public page, structure its content into blocks and retain a visible source link and fetch timestamp on created blocks.

The default single-page workflow is the conservative choice. Site crawling should be limited to the pages actually required because each fetched page consumes credits.

Grounded-research workflow

Grounded research preserves footnote citations and a visible Sources block. Choose quick research for normal questions and deep research only when the task genuinely requires a more thorough run.

Preview before writing

Both research paths support preview workflows when the calling application needs to inspect the structured result before modifying a publication. Provider credits still apply to successful research work even when preview mode does not write to a Page.

Create versus append

Omit a publication target to create a new Page. Supply the existing publication id, slug or Stated URL when the research should be appended to an existing Page.

For recurring deliverables, retain the publication identity in the system that owns the automation.

Source provenance

Do not remove source attribution when transforming retrieved content. Keep the distinction between source facts and AI interpretation visible to the audience.

Human-facing output

After research is written to Stated, use appropriate Page blocks for the audience. Tables should represent real comparable attributes. Charts should use real numerical data. Interactive blocks can collect the next human signal when the workflow requires one.

Example

For targeted monitoring:

Costs and controls

Research and scraping consume Stated Credits. Check plan usage before expensive or bulk operations, keep crawl limits conservative and use deep research only when the user explicitly requires it.

See the Firecrawl and Parallel workflow documentation for provider-specific behavior, and the workflow-deliverables API guide for persistent artifact patterns.