TL;DR
- Don’t ask AI to “summarize.” Ask it to produce a decision document: one recommended decision, three conclusions, one number behind each, one named risk.
- The reusable chain: lock audience and decision first, then TL;DR, then evidence, then counter-view, then you hand-write the opening two sentences.
- Tool fit (June 2026): a 30-page report fits comfortably in any of the three majors; a 90-minute transcript or a multi-file pack is where 1M-token context (Claude, Gemini) beats ChatGPT Plus’s in-app ceiling of roughly 320 pages.
- Hard target: body under 250 words, single page, no font tricks. If it doesn’t fit, the thinking isn’t done yet.
What this covers
A repeatable 20-minute workflow to turn substantial source material — a long report, a 90-minute meeting, a deep analysis — into a 1-page executive summary that a CFO or board reads in 90 seconds and acts on. It follows the same answer-first logic McKinsey consultants use: the Minto Pyramid Principle, often called BLUF — Bottom Line Up Front. Lead with the conclusion; let the support follow.
Who this is for
PMs, analysts, founders, and chiefs of staff who present up — board decks, leadership reviews, budget asks. If you write the page and someone more senior decides on it, this is for you.
When to reach for it
When you have dense content and a thin slice of senior attention — say 15 minutes. If the next step is presenting the same material live, pair this with the 10-slide work-presentation outline workflow so the summary and the deck carry one identical message.
Pick the right tool for the source size
The model matters less than whether your source fits in context in one pass. Splitting a long transcript across chats is where summaries lose the thread. Figures below are current as of June 2026.
| Source | Fits in one pass? | Best tool | Why |
|---|---|---|---|
| 30-page report (PDF) | Yes, anywhere | ChatGPT Plus ($20), Claude Pro ($20), Google AI Pro ($19.99) | ~15–20K tokens; all three handle it |
| 90-min meeting transcript | Needs room | Claude (Sonnet 4.6 / Opus 4.7, 1M ctx) or Gemini 3.1 Pro (1M ctx) | A 90-min transcript runs 20–35K tokens, but you want headroom for follow-ups |
| Multi-file pack (report + transcript + emails) | Often overflows | Claude Pro or Google AI Pro | ChatGPT Plus in-app context tops out near 320 pages; full 1M is on the $200 Pro tier only |
| Nuanced / legal / political source | Quality over size | Claude Opus 4.7 | Strongest at preserving intent and hedges in dense prose |
Practical notes on ChatGPT (as of June 2026): Plus allows roughly 80 file uploads per 3-hour window, up to 10 files per message, 512 MB per file, and a ~2M-token hard cap per file. A 100-page PDF counts as one upload; 100 page-images would burn your whole allowance, so always upload the document, not screenshots. For tool-by-tool depth, see our ChatGPT vs Claude vs Gemini comparison.
Step by step
Every step ships a copy-paste prompt. Replace the angle-bracket placeholders with your real info.
-
Upload source + lock audience. Attach the report PDF / transcript / deep doc in the same chat, then send:
The attachment is <type, e.g. "30-page market report" / "90-min board transcript" / "compilation of 12 user interviews">. Audience: <e.g. "CFO. Cares about cost, risk, timeline. Not familiar with product detail."> Scenario: <e.g. "Decides Tuesday whether to approve $2M extra budget. Has 15 minutes."> Voice: <direct, no adjective stacks, no buzzwords> Before generating anything, tell me: 1. Your read of the source's structure (3 sentences max) 2. Which 3 parts are most relevant to <audience> and which 3 parts are largely irrelevant 3. What's missing that would let you write a summary that's actually useful for <audience> Do NOT generate the summary itself yet. -
Force the “why” — what decision do you want?
Based on the source and audience, give me 3 candidate "decisions the reader should make" — each as verb + object + how-to-measure. Examples: - Approve / reject the $2M Q3 channel budget - Move "renew vs switch" to the next monthly review - Hold the status quo (no action) Do NOT give vague "be informed" / "for reference" framings.Pick one ask. Even if AI picks wrong, you now know exactly what to correct.
-
3-bullet TL;DR — conclusions only, no descriptions:
The ask is: <selected ask from previous step> Now write a 3-bullet TL;DR. Each bullet must: - Be a judgment / conclusion, not a description ("X grew 22%" is description; "X grew 22%, above renewal threshold" is judgment) - Be 14 words or fewer - Not repeat points across bullets - Bullet 1 directly answers the ask; bullet 2 gives critical nuance; bullet 3 names risk / next step -
Evidence layer — one sentence per bullet, each must carry a number:
Continuing from the 3-bullet TL;DR, write 1 evidence sentence per bullet: - Must include a specific number / date / source citation (e.g. "report p.14") - No vague intensifiers ("significantly", "substantially", "materially") without magnitude - After the number, 1 short subordinate clause on why that number matters - 25 words or fewer per sentence -
Counter-view — “if I’m wrong, most likely because”:
Write 1 sentence of "counter-view / risk": if the TL;DR conclusion is wrong, what's the most likely reason? Requirements: - Not a catch-all ("macro changes", "market risk" — these say nothing) - Must point to 1 observable leading indicator (e.g. "If Q3 new logos drop below 8, this conclusion fails") - 20 words or fewer -
Hand-write the opening 2 sentences. AI openings are 90% “This report is based on…” / “Following recent analysis…” filler. Write your own:
Sentence 1: tell the reader, in 1 line, what decision they have to make after reading. Example: "By month-end, you decide: renew SaaS platform Y, or switch to in-house." Sentence 2: give your recommendation directly ("I recommend..."). Example: "Based on the last 90 days of cost and reliability data, I recommend renewing for 12 months." -
Force 1 page. Stitch the open + TL;DR + evidence + counter-view together. Body (excluding header) must be 250 words or fewer. Over budget? Cut with this prompt:
Below is a summary draft, currently <X> words, target 250 words or fewer. <paste draft> Cut to target length. Rules: - No font shrinking. No translation tricks. - Don't delete numbers, decisions, or the counter-view. - You CAN delete: adjective stacks, transition sentences, "as you can see" lead-ins. - Break long subordinate-clause sentences into two short ones. - Keep all [p.14]-style citations. Return the final version + how many words you cut. -
Send + archive as a template. Save this round’s source + ask + the prompt chain to
exec_summary_templates/<topic>.md. Next quarterly review or board update, swap the source file and the ask and reuse everything else. Over a year this is the difference between writing summaries and assembling them.
The shape of a 1-page summary
[Open line 1] The decision you face after reading this.
[Open line 2] I recommend ___.
TL;DR
- <conclusion 1, answers the ask>
- <conclusion 2, the nuance>
- <conclusion 3, risk / next step>
Why
- <conclusion 1>: <number + 1 clause on why it matters> [p.X]
- <conclusion 2>: <number + 1 clause> [p.X]
- <conclusion 3>: <number + 1 clause> [p.X]
If I'm wrong: <one observable leading indicator that would flip this>Common mistakes
- No 1-line ask. The reader finishes the page and still doesn’t know what you want them to do.
- Strongest evidence buried in paragraph 4. Senior readers skim top-down; lead with the number that decides it.
- Skipping the counter-view. A page with zero risk reads naive. Naming the one thing that would flip your call signals you’ve stress-tested it.
- Letting AI write the opening. Generic openers are the clearest machine tell, and executives smell them instantly.
- Splitting a long source across chats. Context resets between sessions; summarize the whole source in one pass (see the tool table above).
FAQ
- Is 1 page really enough for the C-suite? Yes. The page is the decision artifact, not the analysis. Attach the full deck or model as a linked appendix; if they want depth, they ask. Pages 2+ rarely get read.
- Bullets or prose? TL;DR in bullets, evidence layer in short prose. The mixed format reads fastest because the eye lands on the bullets first.
- Which model should I default to in 2026? For a single report under ~30 pages, any of ChatGPT Plus, Claude Pro, or Google AI Pro (all ~$20/month) is fine. For long transcripts, multi-file packs, or nuanced/legal sources, default to Claude (1M context, strongest at preserving intent) or Gemini 3.1 Pro (1M context).
- Will the reader know it was AI-drafted? Not if you hand-write the opening two sentences and own the recommendation. AI does the compression and the evidence pull; the judgment and the voice are yours.
- How do I keep the numbers trustworthy? Demand a page or source citation on every figure (the evidence-layer prompt enforces this), then spot-check two or three against the source before you send.