Verified against ChatGPT · 2026-08-13
Build a literature comparison matrix that actually lets you compare studies, not just list them
Produces a structured comparison matrix across a set of studies or sources on consistent dimensions you choose, so you can scan for patterns across methodology and findings instead of reading each source's summary in isolation.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Build a literature comparison matrix from the sources below, structured so studies can actually be compared against each other on the same dimensions, not just summarized one after another. SOURCES Study A (n=200, retail workers): found flexible scheduling increased retention 12%. Study B (n=45, healthcare workers): found no retention effect but improved self-reported wellbeing. Study C (n=310, mixed industries): retention improved only among workers with caregiving responsibilities. COMPARISON DIMENSIONS sample size and population, methodology (survey/experiment/observational), retention effect found, wellbeing effect found, key limitation stated by the authors WHAT I'M TRYING TO DECIDE OR ARGUE Deciding whether to recommend flexible scheduling company-wide or only for specific employee segments. For each source, extract only what's needed to fill each dimension in sample size and population, methodology (survey/experiment/observational), retention effect found, wellbeing effect found, key limitation stated by the authors — do not include a general summary of the source alongside the matrix; the matrix itself is the deliverable, not a supplement to a separate summary. If a source doesn't report information for a given dimension, mark that cell as "not reported" rather than leaving it blank or inferring a plausible value — a matrix that silently fills gaps with inference is more dangerous than one that's honest about what's missing, because the gaps are exactly what someone comparing methodologies needs to see. After the matrix, add one row of pattern notes per dimension: where sources cluster, where they diverge, and whether any divergence tracks a specific difference in method (sample size, population, measurement approach) rather than being unexplained disagreement. Only claim a divergence "tracks a methodological difference" if the sources actually differ on that specific method — don't manufacture a tidy methodological explanation for a disagreement that might just be genuine disagreement. Given what I'm trying to decide or argue, close with a short note on which specific cells in the matrix are most load-bearing for that purpose — the ones where the sources most directly bear on my actual question, as distinct from dimensions that are interesting background but don't move my argument either way. OUTPUT FORMAT A matrix table: rows are sources, columns are the comparison dimensions. Below it, one pattern-note line per dimension. Close with the load-bearing cells note.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Requiring the matrix to be the sole deliverable rather than a matrix-plus-summary combination forces genuine dimensional extraction instead of GPT-5.1's default tendency to fall back on prose summarization, which is a more natural output shape for the model but defeats the entire point of a comparison matrix — the value of a matrix is that identical dimensions sit in the same column across every row, letting a reader scan down and compare directly, and a parallel summary paragraph lets the model quietly avoid the harder job of forcing inconsistent source reporting into consistent categories. The explicit "not reported" rule instead of leaving cells blank or silently inferring a value targets a specific and consequential failure: a model filling a structured table under implicit pressure to look complete will often infer a plausible value for a missing dimension rather than admit the source didn't report it, and for a literature comparison this is actively harmful because gaps in reporting are frequently the most useful signal — for instance if none of the studies report attrition rate, that absence itself is worth knowing, not something to paper over with a guess. The instruction to only attribute a divergence to a methodological difference when the sources genuinely differ on that specific method guards against a related pattern where a model asked to explain disagreement reaches for the most available tidy explanation (different sample sizes, different populations) even when the sources are actually just similar-method studies that disagree, which would misrepresent the disagreement as more resolved than it is. Closing with which cells are load-bearing for the stated purpose keeps the matrix from becoming an undifferentiated wall of comparison points, directing attention to the specific evidence that actually bears on the decision at hand rather than treating every dimension as equally decision-relevant.
What you get back
Row for Study A: n=200/retail, survey-based, +12% retention, not reported (wellbeing), authors note self-selection bias in respondents. Pattern note on retention effect: two of three studies find a positive effect, but Study C's effect is concentrated specifically among caregivers, suggesting the retention benefit may not be uniform. Load-bearing cells: the population column and Study C's caregiver-specific finding matter most for deciding whether to recommend flexible scheduling company-wide versus segment-targeted.
Verified against
ChatGPT GPT-5.1 · 2026-08-13
Changelog
- 2026-08-13 — 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
