Ask for a narrative, not a summary: a better prompt for reading long reports with LLMs
Low_Helicopter_1354 · reddit · 2026-09-20
Instead of pasting a long report and asking for a summary—which flattens the argument into one gray paragraph—the author asks the model to walk through it as a presentation, one beat at a time, flagging where the argument actually turns.
This reveals which sections deserve their own treatment and which are supporting detail. Long context does the heavy lifting, but only when asked for structure instead of compression. A useful refinement: have the model flag beats where it is inferring rather than quoting, so you know what to verify against the source.
More from Apps
- Claude beats GPT image 2.5 at on-brand design 9/10 times with a brand guide — sahilypatel · 2026-09-20
- Comprehensive 55-minute Codex Desktop Tutorial Covers Skills, MCP, TikTok and Blender — aziz4ai · 2026-09-20
- Stalkr adds keyword groups to benchmark your brand vs. competitors, with API and MCP access — marclou · 2026-09-20
- WIRED Hands-On: Meta's Muse AI Agent Is Better at Surveillance Than Getting Tasks Done — nordicinst · 2026-09-20
- Developer builds iOS widget to show Codex usage limits at a glance — Dimillian · 2026-09-20
- Astra aces the pelican test in Nautilo: co-creative design without prompt-and-pray — Dan_Jeffries1 · 2026-09-20