Build a balanced evidence table without forcing neutral or irrelevant material into two sides
7 min setupTested with:ChatGPTReviewed: 2026-08-28
1
Add your context
Your text stays in this browser. AILesson Prompts does not send it to a model or server.
2
Your prompt
Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt
Extract evidence supporting and opposing the exact proposition.
Proposition, alternatives, definitions, scope, and use:
[question]
Sources with IDs, dates, authors, methods, and passages:
[sources]
Eligibility, population/time fit, independence, quality, and inference rules:
[rules]
Use only supplied material. Extract the smallest evidence unit that bears on the proposition and preserve context, measure, denominator, period, and limitations. Classify each as supports, opposes, mixed/conditional, neutral/context, or irrelevant; do not force every source into two sides. Separate a source author's conclusion from underlying data and your interpretation. For each item provide source locator, evidence statement, direction, directness, quality, independence, applicability, and what claim strength it permits. Identify shared underlying data, conflicts, publication bias risk, missing counterevidence, and conditions that reverse the conclusion. Finish with a balanced synthesis that weights evidence quality rather than counting items.
Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.
From input to outcome
A worked example
See how concrete context turns this recipe into a usable result
Actual input
Claim or decision question
For a 120-person UK office, does replacing assigned desks with a 0.7 desk-per-person shared model improve space efficiency without materially reducing employees' ability to find a suitable workspace during peak days? 'Suitable' requires a desk in the needed quiet/collaboration zone. Use for pilot design, not a permanent decision.
Source material
S1 occupancy count, our office, six Tue–Thu weeks: peak simultaneous attendance 72–98; zones not recorded. S2 employee survey, 86/120 responses: 58 expect shared desks to be acceptable, 19 unsure, 9 oppose; hypothetical question. S3 facilities simulation: at 0.7 ratio (84 desks), average modeled utilization rises 51%→69%; assumes attendance independent by person and no zone constraint. S4 pilot at another 300-person firm: 0.65 ratio; 12 of 20 observed peak days had no free quiet-zone desks; total desks were still available; different hybrid policy. S5 vendor blog cites S4 and says sharing 'eliminates wasted space.'
Evidence rules
Direct internal observations high for our attendance but incomplete without zones. Simulations are conditional on assumptions. Surveys measure expectations, not actual workspace access. External firm evidence is indirect but relevant to zone risk. Vendor blog is not independent of S4. Preserve counts and do not define 'materially' without a threshold.
Example output
Supports efficiency: S3 models utilization rising 51%→69% with 84 desks; directness medium, conditional quality because independence and no-zone assumptions are untested. S1 shows many observed peaks below 84, but 72–98 means some peak counts exceed capacity; mixed rather than support. S2: 58/86 respondents expect acceptance; low directness for actual access and 34 non-respondents.
Opposes the combined proposition: S4 found no quiet-zone desk on 12/20 peak days despite free desks elsewhere; indirect applicability because ratio, scale, and hybrid policy differ, but directly demonstrates the zone-allocation mechanism. S1 also opposes an unconditional 84-desk model on days above 84 unless attendance is managed; it cannot assess zone suitability. S5 adds no independent evidence and its “eliminates” conclusion exceeds S4.
Missing: our peak distribution by zone and hour, team co-attendance, suitable-zone demand, day-level overflow frequency, and an agreed materiality threshold. Balanced synthesis: evidence conditionally supports higher modeled utilization, but does not establish acceptable peak access. The strongest internal observation already shows total-demand risk on some days, while external evidence makes zone mismatch plausible. Design a pilot with booking/overflow, zone-level counts, and a pre-agreed failure threshold rather than claim the combined proposition is supported.
Why this works
1
A mixed and neutral category prevents artificial symmetry and loss of conditional findings.
2
Weighting directness, quality, independence, and applicability is more meaningful than citation counts.
Check the result
Does each evidence unit retain its denominator, period, context, and limitations?
Are author conclusions, underlying observations, and analyst interpretations distinct?
Does the synthesis weight independence and applicability instead of counting sources?
ChatGPT