FeedSummary.com Summary prompts

Markdown normalization

Markdown Feed Summary

Use a controlled Markdown profile as a readable bridge between messy source content and predictable, evidence-aware summary prompts.

CommonMarkheadingslistscode fencessource boundaries

Clean text, stable structure

Markdown as an intermediate representation

It keeps visible structure while removing most page-layout noise.

Heading hierarchy

Section paths guide semantic chunking and document-level synthesis.

Lists + quotes

Procedures, findings, warnings, and attribution remain distinct from paragraphs.

Code + tables

Typed blocks can be rendered predictably while complex data stays structured.

Normalization profile

Use a documented profile

Choose a baseline syntax and a small extension set, then version the conversion rules.

Keep outside the body

Title, canonical URL, authors, dates, source ID, retrieval time, and content coverage belong in a metadata envelope.

Do not force the summarizer to recover metadata from a prose preamble.

Keep inside the body

Headings, paragraphs, lists, quotations, links, code, captions, and simple tables should remain in reading order with stable block identifiers.

Unsupported embeds should become labeled placeholders, not silent omissions.

Workflow

Prompt boundaries remain essential

Markdown is readable, but source text is still untrusted data.

Build canonical blocks

Store element type, source location, ID, and text before rendering Markdown.

Render a limited dialect

Use stable headings, lists, code fences, links, and table rules.

Wrap each source item

Include unique IDs and unmistakable start and end boundaries.

Chunk semantically

Keep headings with paragraphs and tables with labels and units.

Validate structured output

Generate final Markdown from accepted fields rather than raw model formatting.

Feed Signals

Complete Markdown guide

Conversion rules, links, tables, code, boundaries, chunking, and round-trip testing.