# Seven Summarization Prompts A set of prompts for turning long source material into different kinds of summaries. Each one does a different job. Pick by what you need, not by which sounds best. --- ## How to Use These **1. Paste the source text first, then the prompt.** Every prompt in this set says "the text above." That wording is deliberate. Instructions placed after a long block of text get followed more closely than instructions buried above it. Paste your article, transcript, or document into the chat, hit return a couple of times, then paste the prompt underneath. **2. Pick one prompt. Don't stack them.** Combining two of these produces a longer, muddier output than either would alone. If you want both an outline and an action plan, run them as two separate requests on the same source. **3. Don't remove the word limits.** The caps are the part that makes these work. Without a hard number, "concise" is a hope rather than an instruction, and the model will pad. If a cap is too tight for your source, raise the number — but keep a number. **4. Notice which prompts invent content.** Prompts 3, 6, and 7 can generate material that wasn't in your source. Each one instructs the model to label that material. Read the labels. If you're summarizing something where accuracy matters, prompts 1, 2, 4, and 5 stay closest to the original. **5. If the output is too long, say so and name a smaller number.** "Redo that at 200 words" works better than "make it shorter." --- ## Quick Reference | # | Use it when you need | |---|---| | 1 | Everything covered, explained simply | | 2 | Only the points that change your thinking | | 3 | To actually learn the concept | | 4 | To make a decision | | 5 | To see the structure | | 6 | To remember it later | | 7 | To do something with it | --- ## 1. Full Coverage in Plain Language Everything important from the source, explained simply, capped at 500 words. > Summarize the text above. Cover every major point and leave nothing important out, but > explain each one in plain language a 12-year-old could follow. Maximum 500 words. Use > bullets, one idea per bullet. Use only what the source says — if it leaves something > unexplained, note the gap rather than filling it. --- ## 2. Decision-Changing Points Only Strips background and anything a reader could already guess. > From the text above, extract only the points that would change a reader's decision or > contradict a common assumption. Skip background, repetition, and anything a reader could > guess. Maximum 300 words, maximum 10 bullets. If fewer than 10 points meet the bar, give > fewer. --- ## 3. Three-Pass Concept Walkthrough What it is, how it works, what experts know that beginners don't. > Explain the concept in the text above in three passes: (1) what it is, in 100 words, no > jargon; (2) how it works, in 300 words, introducing terms as you go; (3) what an expert > knows that a beginner doesn't — edge cases, caveats, disagreements — in 300 words. Draw > only on the source. Mark anything you add from general knowledge as such. --- ## 4. Four-Header Brief for Making a Call Insights, problems, solutions, actions, with urgency flagged honestly. > Rewrite the text above as a decision summary under these four headers: Key Insights / > Problems / Solutions / Recommended Actions. Maximum 400 words total. Under Recommended > Actions, list only steps the source actually supports, and note which are time-sensitive > versus which are not. --- ## 5. Three-Level Indented Outline Topic hierarchy plus a section for connections the hierarchy can't show. > Convert the text above into an indented outline. Three levels maximum: main topics, > sub-topics, supporting detail. Maximum 8 main topics. After the outline, add a short > section listing connections between topics that the outline structure can't show. --- ## 6. Key Points Plus Labeled Recall Aids Source material and invented memory devices kept visibly separate. > Two parts. First: the key points from the text above as short bullets, 200 words maximum, > source material only. Second, under the header "Memory Devices": analogies and mnemonics > to help me retain those points. Label the second section clearly as your own invention, > not content from the source. --- ## 7. Ordered Steps with Completion Criteria What to do, what each step needs, and how to know it's finished. > Turn the text above into an execution plan. List the concrete steps the source supports, > in order, with what each step requires and how to know it's done. Maximum 12 steps. Flag > any step where the source is vague about method. Do not invent templates or examples — > if the source doesn't provide one, say so. --- ## If You Want to Modify These Three things are doing the work in every prompt here. Keep them when you write your own: - **A hard cap.** A word count or an item count. Not "concise," not "brief." - **A selection rule.** Something the model can actually apply, like "would change a reader's decision." Not "high-impact" or "critical." - **A fidelity clause.** An instruction to use only the source, and to name gaps instead of filling them. Drop any one of the three and the output drifts toward long, generic, and partly invented.