--- title: "How to Write Exec Summaries With AI" description: "Your exec will read 200 words. Here is the five sentence structure, a real before and after, and the edits a model never makes to its own summary draft." canonical_url: "https://builderscamp.com/guides/tools/ai-for-executive-summaries" date_published: "2026-09-18" date_modified: "2026-09-18" author: "Andre Albuquerque, Inês Lourenço" publisher: "Builders Camp" guide_class: "tools" --- # AI for executive summaries: writing for someone who reads 200 words **TL;DR:** An executive summary is five sentences: what happened, the number, why now, what it cost, and what you need from the reader. A model compresses a long document into that shape in one pass and reliably deletes the caveat that would change the decision, so the useful second prompt is asking it what it cut. Everything after that is your judgment about which reader distrusts which number. ## What is an executive summary actually for? It is not a shorter version of the document. It is the artefact a person reads when they will not read the document, and it has to leave them able to do one thing: agree, disagree, or ask the one question that matters. A summary that faithfully compresses all twelve pages in proportion fails this, because the twelve pages were written to be complete and the summary has to be decisive. That difference explains why a model's first draft feels smooth and lands flat. Asked to summarise, it summarises: it preserves the balance of the original, spreads the word count across every section, and produces something a reader finishes without a view. You wanted an argument compressed, not a document shrunk. ## Five sentences, in this order 1. **What happened or what you are asking for.** Present tense, no preamble. "Renewals in the enterprise segment are down and we want to move three engineers onto the admin console for six weeks." 2. **The number that makes it matter.** One figure, with the denominator. Not "renewals dropped", but "renewals dropped from 91 percent to 78 percent across 46 accounts". 3. **Why now.** The thing that changed since this was last discussed, which is what stops a reader filing it as background. 4. **What it costs, and what you traded.** Both halves. The cost alone reads as a bill; the trade reads as a decision. 5. **What you need, and by when.** A request with a verb and a date, addressed to the person reading. Four of those five sentences exist in your source document already. Sentence four usually does not, because trades get made in conversations and written down nowhere, which is why the summary is often the first place a project's real cost becomes visible. ## Before and after, on a real compression A model given a 9 page quarterly analysis and asked for an executive summary produced this: > This quarter saw significant developments across our enterprise segment. Renewal performance showed some softening relative to prior periods, driven by a combination of factors including product gaps and increased competitive pressure. The team has identified several potential paths forward and recommends prioritising investment in administrative tooling. Further analysis is ongoing and we will continue to monitor the situation closely. Every sentence is true and none is usable. "Significant", "some softening", "several potential paths" and "continue to monitor" are four hedges in four sentences. The edit: > Enterprise renewals fell from 91 percent to 78 percent across 46 accounts this quarter. Twenty-nine of the 34 exit interviews named the same gap: admins cannot manage seats without filing a ticket. This became urgent in August when two of our three largest accounts hit the same wall in the same week. We want three engineers on the admin console for six weeks, which pushes the billing revamp to Q2. Please confirm by Friday, or tell us which of the two you would rather delay. Same facts, half the hedges, and the reader now has something to disagree with. The last sentence is the one that changes how the email gets answered: it offers the reader a real alternative, so "no" produces information instead of silence. ## The prompt that gets you a usable draft Give the model the full document, not a description of it, and give it the reader. > Here is the full analysis. Write a five sentence executive summary for our VP of Engineering, in this order: what we are asking for, the number with its denominator, what changed recently, the cost and what it delays, the request with a date. No adjectives where a number exists in the source. Then, separately, list the three things you cut that someone approving this might need to know. The second half of that prompt does most of the work. A model asked what it dropped will name the dropped caveat accurately, because identifying it is an easier task than deciding it was load-bearing. Read those three, promote at most one back into the five sentences, and let the other two live in the document you attached. If you are summarising a decision rather than an analysis, the upstream artefact matters more than the prompt. [Claude Code meeting notes to decisions](https://builderscamp.com/guides/tools/claude-code-meeting-notes-to-decisions) covers turning a transcript into a record with owners and reversal conditions, and a summary written from that record needs almost no compression at all. ## The five edits a model will not make on its own - **Delete the background sentence.** A model writes one every time, because documents open that way. A reader who needs background will open the attachment. - **Swap each adjective for the number under it,** and if there is no number, cut the sentence rather than keep the adjective. - **Put the trade back in.** The model summarises what the document says; the document rarely says what you gave up to propose this. - **Promote one cut caveat,** chosen because it would change the reader's answer, not because it is interesting. - **Address the request to a person with a date.** "Feedback welcome" is how a summary becomes a thread nobody closes. ## What does this not fix? A model cannot know which of your numbers the reader already distrusts. If your VP has watched the same churn metric get redefined twice, the sentence built on it reads as a flag rather than evidence, and no amount of compression helps: you need the definition in the sentence, or a different number. That is a relationship fact, and it lives in your head. The other limit is sharper. Compression is lossy in a specific direction: it keeps what is stated most often and drops what is stated once, and the sentence that would reverse a decision is almost always stated once. Asking the model what it cut recovers most of that, and it does not recover the caveat that was never in the document because you knew it and never wrote it down. Read your own five sentences and ask what a sceptical reader would need that is not there. ## Where this sits next to the rest of the work Writing the summary well is a downstream skill, and it exposes upstream gaps fast. If sentence two keeps coming out weak, the problem is the metric, and [data storytelling](https://builderscamp.com/guides/glossary/data-storytelling) covers turning a number into a claim. If sentence four keeps coming out vague, the problem is that nobody priced the trade, and Business for Product Managers, a 1 week bootcamp with 2 live sessions and 8 microlessons, covers unit economics and building a business case with honest assumptions. If you are writing these for a leadership team rather than a manager, the operating cadence and stakeholder material in Head of Product is the level above this one. Product Storytelling is the direct match for the writing itself: narrative structure, framing, and the practical scripts for decks, memos and updates, in 1 week with 2 live sessions and 7 microlessons. [See the Product Storytelling bootcamp](https://builderscamp.com/bootcamps/product-storytelling?utm_source=guide&utm_medium=organic&utm_campaign=ai-for-executive-summaries) One habit worth stealing from people who write these well: keep the last three summaries you sent and check which ones got answered within a day. The pattern is rarely about length. It is almost always sentence five, and whether it asked a specific person for a specific thing. Start there before you start rewriting sentence one, and if you want the longer form the summary compresses, [how to write a one-pager with AI](https://builderscamp.com/guides/templates/how-to-write-a-one-pager-with-ai) is the document that should sit behind it. ## Frequently asked questions ### How long should an executive summary be? Five sentences, roughly 150 to 200 words, sized so the reader never scrolls. Length is not the real constraint: an exec who reads the whole thing and cannot say what you want from them has read a summary that failed at 40 words or 400. ### What does AI actually do well here? Compression and first-pass ranking. Given the full document, a model reliably finds the three or four load-bearing facts and drops the rest, which is the slow part when you wrote the document and cannot see it cold any more. ### What does AI do badly? It drops caveats. A model compressing a 12 page analysis treats the sentence that would change the decision as one sentence among many, so it disappears at the same rate as the background. Recovering that caveat is a manual step every time. ### Should the ask go first or last? Last, as a request with a date. First position belongs to the decision or the news, because a reader who stops after one sentence should still leave with the thing that happened, not with a task whose reason they never read. ### How do you stop the summary sounding hedged? Replace every evaluative adjective with the number underneath it. Significant becomes 18 percent, rapid becomes six weeks, concerning becomes the churn figure. If no number exists, the sentence was an opinion and belongs in the recommendation, not the summary. ### Can you ask the model what it left out? Yes, and it is the highest-value second prompt available. Ask it to list the three things it cut that a reader making this decision might need, then decide which one earns a sentence back. Models answer that question well and almost never volunteer it. ### Which Builders Camp bootcamp covers this? Product Storytelling, a 1 week bootcamp with 2 live sessions and 7 microlessons, covers written and visual narrative for stakeholder communication, including practical structures for memos and updates. ## Sources - [Builders Camp: Product Storytelling bootcamp](https://builderscamp.com/bootcamps/product-storytelling) - [Builders Camp: Business for Product Managers bootcamp](https://builderscamp.com/bootcamps/business-for-product-managers) ## How this guide was made Researched from Builders Camp's bootcamp, track and masterclass material and the sources listed on this page, drafted with AI, and fact-checked against every source cited.