Read the following document and produce:
1. A one-line takeaway (the single most important thing).
2. The three highest-leverage facts (claims, numbers, decisions).
3. An explicit list of what you cut and why.
Do not include filler or background. Treat me as someone who has 60 seconds to decide whether to read the original.
Document:
{{document}}
Distill a long doc
Example output
One-line takeaway: Postgres 17's logical replication can replace most ETL pipelines for sub-100GB warehouses, but only if you control schema changes.
Three highest-leverage facts:
1. Subscriber-side row filtering removes the biggest historical objection to logical replication.
2. Schema changes still require manual coordination - this is the trap.
3. Performance is within 5% of physical replication for the workloads tested.
What I cut: the entire benchmarks section (numbers without context), the author's editorial about MySQL, and three paragraphs of historical background.
Common mistakes
Don't use this on something you wrote yourself - you already know what's load-bearing. It's a prompt for digesting work from others. Also: don't trust the cut-list blindly. If the doc was about a decision and the model says it cut 'just the background,' check whether the background contained the actual reasoning. The fastest way to ship a bad summary is to assume the model picked the right things to keep.
More from Power Prompts
Ask clarifying questions first
Before answering my next message, ask me up to 5 short clarifying questions about audience, length, tone, format, and any constraints I…
Explain it for a specific reader
Explain {{topic}} to {{audience}} in under 100 words. Use only words and analogies they would recognize. No jargon unless you immediately…
Critique as a named skeptic
Read the draft below as if you were {{skeptic_role}}. Find the specific weaknesses they would push back on. Quote the exact lines you…
Why it works
Compression is the most common AI task and the one most people do badly. They paste a long doc and write 'summarize this' - the model returns a generic abstract that loses the load-bearing claims. This prompt forces three things: a one-line takeaway, the three highest-leverage facts, and the explicit list of what was cut. The cut-list matters more than people realize - it tells you whether the summary is trustworthy. If the model says it cut nothing important, push back; everything cut is a judgment call. Use this for meeting transcripts, research papers, long Slack threads, RFCs, and any document that someone else wrote and you have to react to.