Verified against ChatGPT · 2026-08-08
Turn a pile of sources into a defensible research-gap map instead of a vague 'more research is needed' line
Takes the sources you've already gathered and sorts what they actually agree on, conflict on, and never touch — so the gap you claim is one you can point to in the table, not one you're guessing at.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are helping me build a research-gap map from a set of sources I've already collected — not finding new sources, and not writing the polished "gaps in the literature" paragraph yet. Your job is to sort what I have into agreement, conflict, and silence, so any gap I claim afterward is one I can point to directly. RESEARCH QUESTION Does async standups improve delivery velocity for distributed engineering teams compared to synchronous daily standups? SOURCES (list each with its core claim/finding in one line) Smith 2023: async standups cut meeting time 40% but self-reported alignment dropped. Chen 2024: no significant velocity difference either way, small sample (n=12 teams). Patel 2022: async works only when paired with written decision logs. WHAT COUNTS AS "COVERED" HERE A source counts as covering velocity only if it reports a measured delivery metric, not just a survey of team sentiment. SUSPECTED GAP (if I already have a hunch) I suspect nobody has studied this specifically for teams spanning more than 6 time zones. HOW TO SORT Group the sources into three buckets: claims multiple sources converge on, claims sources actively conflict on (state both sides and which sources hold each), and sub-questions inside my research question that none of the sources address at all. For the conflict bucket, do not average the disagreement into a mushy middle position — name the disagreement as a disagreement, because a real gap-in-the-literature claim is often "these two camps contradict each other and nobody has reconciled it," not just "nobody looked." For the silence bucket, only list a sub-question as uncovered if it's a piece of my actual research question, not a tangential curiosity — an uncovered tangent isn't a gap, it's just something outside scope. If I gave you a suspected gap, check it against the sorted evidence and tell me plainly whether the sources actually support that being a gap or whether one of them already addresses it and I missed it. WHAT NOT TO DO Do not invent a source, a study, or a specific finding to fill in a bucket — if you don't have enough information from what I gave you to judge whether something is covered, say that explicitly instead of guessing. Do not write the final "gap statement" prose yet; this pass is the sorting, not the write-up. OUTPUT FORMAT 1. Convergence table: claim, which sources support it. 2. Conflict table: the disagreement stated as a disagreement, which sources are on each side. 3. Uncovered sub-questions: a short list, each tied to a specific piece of my research question. 4. A verdict on the suspected gap, if given — supported, partially addressed, or already answered by [source].
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The core failure this avoids is the model quietly averaging a genuine disagreement between sources into a hedged middle position — GPT-5.1's default synthesis behavior tends to smooth conflicting claims into something like "results are mixed," which erases exactly the information a gap statement needs, since a real contradiction between two credible sources is often the gap itself, not noise to be flattened. Forcing three explicit buckets (convergence, conflict, silence) rather than one blended narrative keeps the disagreement visible as a disagreement, with sources attributed to each side, so the writer can later argue "camp A and camp B haven't been reconciled" instead of just asserting a vague absence. The instruction to only count a sub-question as uncovered if it's tied to a specific piece of the stated research question closes a common failure mode where a model pads out a gap list with tangential curiosities that sound like gaps but were never actually in scope — this keeps the gap map defensible rather than padded for length. Explicitly forbidding invented sources matters because when a coverage definition is ambiguous, a model under instruction to be thorough will sometimes produce a plausible-sounding citation-shaped claim to fill a bucket rather than admit uncertainty; naming this as the one thing to never do gives the model a clear, checkable failure to avoid rather than a vague quality aspiration. Checking the suspected gap against the sorted evidence rather than accepting it at face value also protects against confirmation bias baked into the prompt itself.
What you get back
Convergence: all three sources agree written documentation matters for async setups. Conflict: Smith reports a velocity gain, Chen finds no significant difference — worth noting Chen's sample is much smaller and may be underpowered rather than genuinely null. Uncovered: none of the three sources measure outcomes specifically for teams spanning more than 6 time zones. Suspected gap verdict: partially supported — timezone-spread specifically is untouched, but the broader async-vs-sync velocity question already has conflicting rather than absent evidence.
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — 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
