Source envelope
URL, canonical URL, title, author, dates, source, retrieval time, and content hash.
Article summarization
Article Feed Summary
Combine reliable feed metadata with clean article content, preserve evidence blocks, and expose extraction limits before generating a summary.
article summarycontent extractioncanonical URLevidence blocksAPI output
Article pipeline
Build an article record, not a blob
Separate page metadata, normalized content blocks, extraction status, and summary output.
Content blocks
Headings, paragraphs, lists, quotes, tables, captions, code, and stable IDs.
Summary object
Abstract, key points, entities, dates, content type, questions, and evidence IDs.
Content selection
Extract with coverage signals
A page can be full, partial, metadata-only, blocked, or unsupported.
When feed content is enough
A substantial full-content feed field can be cleaner and more stable than fetching the destination page. Preserve the item record and summarize that content directly.
Record which field was selected and its meaningful text length.
When page extraction is needed
Fetch only under permitted policy, validate redirects and content type, remove boilerplate, and preserve meaningful structure.
Show the exact extracted version beside the summary during review.
API design guide
Design an API around observable states
Extraction can fail independently of generation.
Designing an Article Summarizer API: Inputs, Outputs, and Quality Controls
Design an article summarizer API with stable inputs, structured outputs, citations, length controls, and transparent failure modes.
Article SummarizationAPI DesignOpenAPI
Read guide