Lab Notes · 2

I Compared 5 Prompts for Summarizing Long Documents

Jun 5, 2026 · AI Note Lab

Five summary prompts scored — structure + audience + citations wins
Five summary prompts scored — structure + audience + citations wins

Summarization is the single most common reason people use AI, yet comparisons of "how to ask for a good summary" are surprisingly rare. So I took one 30-page public industry-trends report and had it summarized five times, changing only the prompt.

Method

The 5 prompts and how they did

① "Summarize this" (4 out of 7)

It did the basics but skewed heavily toward the front of the document. Only 4 of the 7 key points made it in. The important counterargument section near the end was dropped entirely.

② "Summarize this in 3 lines" (3 points)

Short meant clear, but also mushy. A sentence carrying zero information — "several challenges remain" — took up one of the three lines.

③ "Summarize as: core claims / evidence / limitations" (6 points)

Specifying a structure made quality jump. Giving it a dedicated "limitations" slot brought back the counterargument section that ① had dropped.

④ "A summary to report to my manager, focused on what's needed for decisions" (6 points)

Naming the audience made the sentences practical. It even added interpretation — "this is an opportunity for us because…" — which is both a strength and something to watch (the interpretation can be wrong).

⑤ ③+④ combined, plus "add nothing that isn't in the original; cite the source page for each item" (7 points)

The winner. All 7 key points included, and the page citations made verification fast. Zero hallucinated sentences.

What I learned

Final recommended prompt: "Summarize this document using the structure [core claims / evidence / limitations & counterarguments]. Focus on what I need for decisions in [my situation]. Add nothing that isn't in the original, and cite the page where each item appears."
← Previous experiment · ChatGPT vs Claude