Verified against ChatGPT · 2026-08-09
Turn a pile of papers into a literature review that shows where the studies actually disagree
Builds a literature review organized around points of disagreement between studies rather than a paper-by-paper summary, so contradictions surface instead of getting smoothed into one comfortable narrative.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I'm giving you a set of papers or paper summaries on the same general topic. Write a literature review that is organized around where these studies agree and, more importantly, where they disagree — not a list that summarizes each paper in turn.
TOPIC AND REVIEW PURPOSE
Whether standing desks measurably reduce lower-back pain in office workers, for an internal wellness-benefit recommendation.
PAPERS OR SOURCES
Six studies pasted in as abstracts, ranging from a 2015 RCT with 40 participants to a 2024 meta-analysis of 12 studies.
KNOWN METHODOLOGICAL DIFFERENCES
The older studies self-reported pain on a 1-10 scale weekly; the newer meta-analysis used a validated clinical instrument administered monthly.
REVIEW SCOPE BOUNDARY
Only office-worker populations — exclude studies on manufacturing or industrial standing workstations.
STRUCTURE
For each major claim relevant to the topic, group the sources that address it and state explicitly whether they converge or conflict. Where they conflict, do not resolve the conflict yourself by picking a side — instead, identify whether the disagreement traces to a methodological difference (different sample, different measure, different time period) named in The older studies self-reported pain on a 1-10 scale weekly; the newer meta-analysis used a validated clinical instrument administered monthly., or whether it looks like a genuine unresolved contradiction in the field. Order the review by claim, not by paper — a paper that touches three separate claims should appear in three different sections, not get one summary paragraph that flattens all three findings together.
WHAT NOT TO DO
Do not write a review that reads as a string of "Smith et al. found X. Jones et al. found Y." sentences with no synthesis between them — that's an annotated bibliography, not a review. Do not manufacture consensus language ("the literature broadly agrees") when what you actually have is two or three papers that happen not to contradict each other yet.
OUTPUT FORMAT
1. A short framing paragraph stating the review's scope.
2. One section per major claim, each ending with an explicit convergence/conflict verdict and, if conflicting, the likely methodological cause.
3. A closing paragraph naming the single biggest open contradiction the field hasn't resolved.
4. A note on any source that fell outside Only office-worker populations — exclude studies on manufacturing or industrial standing workstations. and was excluded, and why.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Asked to summarize a set of papers with no other instruction, a model's default organizing principle is the input order — one paragraph per paper, in the sequence they were pasted — because that requires no synthesis judgment and is the safest way to guarantee every source gets mentioned; the result reads complete but never actually compares anything. Restructuring the task around claims rather than papers forces a different kind of pass over the material: the model has to first extract what each paper actually found, then re-group by finding rather than by source, which is the step that surfaces disagreement, because two papers addressing the identical claim sit next to each other instead of three sections apart. Explicitly forbidding invented consensus language matters because language models have a documented tendency toward hedge-free, harmonizing phrasing when synthesizing multiple sources — "the literature suggests" is a smoother sentence to generate than "these two sources disagree and neither is obviously more reliable," so without an explicit instruction against it, real disagreement gets written over with false unity. Tying each conflict back to named methodological differences gives the model a legitimate way to explain disagreement without picking a winner — it's the difference between "these studies contradict each other" (an unhelpful dead end) and "these studies used different pain-measurement instruments and different sample sizes, which plausibly explains the gap" (an actual analytical finding a reader can act on).
What you get back
Claim: standing desks reduce self-reported lower-back pain. Four of six sources report a modest reduction; two report no significant effect. The two null-result studies both used the validated clinical instrument rather than self-reported weekly scores, suggesting the positive results may partly reflect a less sensitive measurement approach rather than a true larger effect. Verdict: conflicting, likely methodology-driven rather than a genuine contradiction in the underlying effect.
Verified against
ChatGPT GPT-5.1 · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
