What each part of the prompt does
- Findings with numbers
- Asking for the measured figures, not a description of them, lets the abstract state results precisely. An abstract that says "a moderate effect" when you have the exact coefficient reads weaker to reviewers and loses the detail readers scan for.
- Gap from your sources only
- Step 1 limits the gap statement to the prior work you supplied. This keeps Claude from describing a gap that a paper you have not cited already filled, which is a common objection from examiners.
- Fixed section order
- The abstract runs context, gap, aim, method, results, meaning; the introduction moves from broad problem to your aim. Both orders match what most journals and thesis guides expect, so you edit wording rather than rebuild structure.
- Visible citation gaps
- The lowercase source-needed marker and the closing list turn missing references into a checklist instead of hiding them behind invented citations, the most damaging mistake an AI draft can make in academic work.
Variations to try
Same idea, one thing changed.
Conference abstract only <task>Write a conference abstract from my findings. Do not add results, citations or claims that are not in my notes.</task> <inputs> Field and target journal or course: [FIELD] Research question: [QUESTION] Method in brief: [METHOD] Main findings, with the numbers I measured: [FINDINGS] Prior work I build on, with the sources I have read: [PRIOR WORK] Word limit for the abstract: [ABSTRACT LIMIT] </inputs> <steps> 1. List the gap my study fills in one sentence, using only the prior work I gave you. 2. Write the abstract in this order: context, gap, aim, method, key results with their numbers, and what they mean. Stay under [ABSTRACT LIMIT] words. 3. Add a title of no more than 15 words that names the main finding. 4. Where a claim needs a citation I did not supply, write [source needed] instead of inventing one. </steps> <format> The title, then the abstract as one paragraph. After it, a short list of every [source needed] spot and any finding you were unsure how to phrase. </format>
Plain-language summary too <task>Write the abstract and the introduction of a research paper from my findings. Do not add results, citations or claims that are not in my notes.</task> <inputs> Field and target journal or course: [FIELD] Research question: [QUESTION] Method in brief: [METHOD] Main findings, with the numbers I measured: [FINDINGS] Prior work I build on, with the sources I have read: [PRIOR WORK] Word limit for the abstract: [ABSTRACT LIMIT] </inputs> <steps> 1. List the gap my study fills in one sentence, using only the prior work I gave you. 2. Write the abstract in this order: context, gap, aim, method, key results with their numbers, and what they mean. Stay under [ABSTRACT LIMIT] words. 3. Write the introduction in four paragraphs: the broad problem, what is already known, the gap and why it matters, and my aim with a one-line preview of the method. 4. Where a claim needs a citation I did not supply, write [source needed] instead of inventing one. </steps> <format> The abstract as one paragraph, then the introduction under its own heading. Then a plain-language summary of 120 words for readers outside the field, with no jargon. After all three, a short list of every [source needed] spot and any finding you were unsure how to phrase. </format>
If the result isn't right
- Paste findings exactly as they appear in your results tables; rounding or paraphrasing them first is how wrong numbers end up in the abstract.
- If the draft overstates what your design can show, reply "my study is correlational, rewrite any causal wording" and Claude will soften the verbs.
- Check the length against the journal's own count, since some include the title and keywords in the limit.