# How to Write a Summary That Keeps the Source Honest URL: https://aginsi.com/journal/how-to-write-a-summary Type: blog Locale: en Published: 2026-10-22 Updated: 2026-10-24 --- > A working method for how to write a summary: choose the format, find the claim, paraphrase without distortion and verify every figure. Last spring a client sent me a 58-page evaluation report and asked for "a quick summary by tomorrow". The first draft I produced ran to three pages and said almost nothing. Knowing how to write a summary is less about cutting words than about deciding, in advance, what the reader will do with the result. A good summary states the source's main claim, its evidence and its limits in roughly one tenth of the space, in your own words, without adding opinions. ## What is a summary actually for? A summary is a substitute reader. Someone who will never open the original should still be able to repeat its argument correctly, and someone who will open it should know where to look. That sounds obvious, yet most weak summaries fail on the first test. They describe the document ("this report discusses funding models") instead of reporting it ("the report finds that multi-year grants cut administrative costs by a fifth, but only for organisations above 20 staff"). The first version could be written without reading. The second could not. Before writing anything, answer one question in a single sentence: who reads this, and what will they decide or do afterwards? A colleague deciding whether to read the full paper needs orientation. A director deciding on a budget needs findings and a recommendation. Same source, two different summaries. ![Pencil underlining a single key sentence in a printed paper](https://fdzlnqpwsaniezitwiuw.supabase.co/storage/v1/object/public/cms-media/aginsi/2026-10/a33b2c-i1.webp) ## Which kind of summary are you writing? Three formats get confused all the time, and the confusion produces most of the bad drafts I see. An abstract belongs to academic writing. It runs 150 to 250 words, follows purpose, method, results and conclusion, and exists so a scholar can decide whether the full paper deserves their afternoon. An executive summary belongs to business and policy documents. It is longer, usually five to ten per cent of the report, written for people who must act, and it leads with findings and recommendations. The [University of Waterloo's engineering guidance on executive summaries](https://ece.uwaterloo.ca/~dwharder/Reports/Executive_summaries/) puts the rule plainly: it should contain nothing that is not in the full document. A plain summary, the kind a course or a briefing asks for, sits between them. It covers the whole text proportionally and keeps the author's order. Skip the abstract format if your reader is not a specialist. The jargon that helps a scholar skim will stop a manager cold. ![Three documents of different length laid side by side](https://fdzlnqpwsaniezitwiuw.supabase.co/storage/v1/object/public/cms-media/aginsi/2026-10/a4c39a-i2.webp) ## How do you find the main point before you write? Read in two passes, and write nothing during the first. The first pass is structural. Read the title, abstract or introduction, every heading, the first sentence of each paragraph and the conclusion. In about ten minutes you should be able to say what the author claims. If you cannot, the text is either badly organised or you are missing background, and a summary written at that point will be a guess. The second pass is selective. Now read the sections that carry the argument and mark three things only: the claim, the strongest piece of evidence for it, and the author's own stated limit. Highlighting is choosing. Marking more than a tenth of the page means you have not chosen yet. Then close the document. Write the main point from memory in one sentence. If you must look back to do it, you have not understood it yet, and no amount of polishing will fix that. ## What does the structure of a good summary look like? Use this order, adjusted to length: - The claim, in one sentence, with the author or source named. - The evidence, two to four points, each with its figure or finding. - The limit or condition: what the source does not show, or where it applies. - If the reader must act, the recommendation, last and clearly attributed. Keep the author's order unless your reader has asked for a different one. Reordering feels like improvement but it quietly changes the emphasis, and the emphasis is part of the argument. On length, a common target is 10 to 20 per cent of the source for a short text, and far less for long reports. A 40-page paper rarely needs more than 300 words. Past that, you are rewriting, not summarising. ## What does a weak summary look like next to a good one? Take an invented but typical source: a 30-page study of remote work in a mid-sized firm. The weak version reads: "This study looks at remote work and its effects on employees and the company. It covers several factors and offers some conclusions." Every word is true and none of it is usable. The stronger version reads: "Across 240 employees over 18 months, the firm found that remote staff delivered the same output as office staff but reported weaker ties with colleagues outside their team. The authors recommend scheduled in-person weeks, and note that the sample excludes new hires, so the finding may not hold for onboarding." Three sentences, a claim, a figure, a recommendation and a limit. The difference is not length. It is that the second version required the writer to decide. That decision is the entire skill, and it is why the first pass matters more than the second. ## How does the method change by type of source? The same steps apply, with a different emphasis for each source. For a research paper, lean on the abstract, the results and the limitations section, and say what kind of study it was. A survey of 200 people and a randomised trial deserve different verbs. For a long report, read the executive summary last, as a check on your own, and pull the numbers from the tables rather than the prose, because prose tends to round generously. For an interview or a meeting, the claim is often implicit. Summarise decisions and open questions, not the order in which people spoke. For a news article, keep who, what and when, and flag anything the reporter attributes to a single unnamed source. Your summary should not make it sound firmer than the original did. ## How do you write it in your own words without distorting it? Paraphrase by meaning, not by synonym. Swapping "increase" for "rise" in each sentence is not paraphrase, and it is how accidental plagiarism happens. Instead, hold the idea, look away from the page and say it as you would to a colleague. Three habits protect accuracy. Keep numbers exact. If the source says 23 per cent of 412 respondents, do not write "about a quarter of those surveyed" unless that rounding serves the reader. When you must round, round in the direction that does not flatter the claim. Keep hedges. "Suggests" and "demonstrates" are different verbs. Authors choose them deliberately, and replacing a hedge with a certainty is the commonest way a summary becomes wrong. Keep your opinion out. A summary reports. If you disagree with the source, write a second paragraph labelled as your comment, or write a separate note. Mixing the two destroys trust in both. ![Laptop and notebook with a short handwritten summary paragraph](https://fdzlnqpwsaniezitwiuw.supabase.co/storage/v1/object/public/cms-media/aginsi/2026-10/a54075-i3.webp) ## Where do AI tools help, and where do they not? Used well, a model is a fast first reader. Used badly, it is a confident stranger. In my own client work, I let a tool produce a draft of the structure, then check every figure against the page. That check takes ten minutes and has caught a wrong number in roughly one draft out of five. What the tools do well: finding the section that carries the argument, producing a first-pass outline, and compressing a long document into a draft you can then correct. What they do badly: preserving hedges, keeping the author's order under length pressure, and attributing claims to the right source when a text quotes several. A read-it-later tool with built-in highlighting changes the order of work, because you mark the key passages first and summarise from your marks rather than from the whole text. If your notes already live in a workspace, an assistant inside it can draft from what you have pasted, which keeps the source and the summary side by side. For readers who keep a local library of notes, plain Markdown files make summaries easy to link back to their sources, which is the habit that matters most over a year. The rule I give clients: let the tool shorten, never let it decide what matters. That decision is the summary. ## How do you check a summary before you send it? Run four checks, in this order. First, the stand-alone test. Hand it to someone who has not read the source and ask them to say the main point back. If they cannot, the claim is buried. Second, the figure check. Every number in the summary should be findable in the source in under a minute. Third, the addition check. Underline anything in your summary that is not in the source: a judgement, an example, a connecting phrase that implies causation. Delete it or label it. Fourth, the language pass. Short sentences, one idea each, no filler. A grammar and style checker catches slips here, though it will not catch a wrong claim. ## What should you skip? A few habits that look careful and are not. Skip the opening sentence that announces the document. "This paper examines..." tells the reader nothing they could not infer from the title. Skip quotation as a shortcut. One short quotation that carries the thesis is fine. Three means you did not condense. Skip adjectives about quality. "A comprehensive and insightful study" is a verdict, not a summary. Skip the closing "overall". The final sentence should be the last piece of content, not a summary of the summary. ## What would we do tomorrow morning? Take the longest unread document on your desk. Spend ten minutes on the structural pass, write the one-sentence claim from memory, then add evidence and the limit, in that order. Aim for under 200 words. Compare it with the source's own abstract or conclusion. Where they disagree, one of you is wrong, and finding out which is the part of the work that makes you a better reader. Do that three times and the method stops feeling like a method. ## FAQ ### How long should a summary be? A common target is 10 to 20 per cent of a short source and far less for long documents. A 40-page paper rarely needs more than 300 words, and an abstract-style summary usually runs 150 to 250. ### What is the difference between a summary and an abstract? An abstract is an academic overview of a paper, built around purpose, method, results and conclusion. A summary condenses any text proportionally and keeps the author's order, without assuming a specialist reader. ### How do I write a summary without plagiarising? Read, look away and restate the idea as you would to a colleague. Swapping synonyms sentence by sentence is not paraphrase. Name the source and use a short quotation only when the exact wording carries the point. ### Should a summary include my opinion? No. A summary reports what the source says. If you disagree, put your view in a separate, clearly labelled comment so the reader can trust both parts. ### Can AI write a summary for me? It can produce a useful first draft and find the section that carries the argument. It often loses hedges and misattributes claims, so check every figure and attribution against the source before using it. ### What is the first sentence of a good summary? The source's main claim, with the author or publication named and a figure where one exists. Avoid openings that only announce the document, such as 'This paper examines'.