Author: AILesson9 min setupTested with:ChatGPTReviewed: 2026-08-28
Quick answer
Extract comparable claims, observations, methods, and limits from several documents while preserving source lineage. Provide: Research or decision question, Labeled source materials, Evidence and table rules. Expected result: A traceable evidence table with source inventory, claim clusters, conflicts, and synthesis limits.
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
Organize the supplied documents into a traceable evidence table for the exact question.
Question, definitions, population, period, alternatives, decision, and claim threshold:
[question]
Sources with IDs, metadata, methods, locators, and relevant text:
[sources]
Eligibility, extraction unit, quality, independence, applicability, quotation, privacy, unknowns, and schema:
[rules]
Inventory sources before extraction and identify duplicate versions, citations that reuse the same underlying dataset, missing pages, and inaccessible attachments. Use one row per smallest evidence unit that can support a distinct claim; a source may create several rows, while a repeated claim without new data is not new evidence. Separate reported observation, author interpretation, recommendation, and analyst inference. Preserve population, sample, denominator, measure, comparison, period, effect/value, uncertainty, method, limitations, and exact locator. Classify relevance and direction only against the stated question. Do not harmonize incompatible metrics, count dependent sources as independent, repair missing values, or treat absence of mention as evidence of absence. Return: source inventory; evidence table; claim-to-source matrix; dependency map; conflict table; evidence gaps; synthesis weighted by directness, quality, independence, and applicability; and rows needing human verification.
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
Research or decision question
Should a 12-week pilot of optional four-day workweeks be approved for the 46-person product department? Decision concerns delivery reliability and employee strain, not permanent adoption. Compare evidence from January–June 2026; 'delivery reliability' means committed milestone met by date; strain means reported workload or overtime. No success threshold has been approved.
Labeled source materials
S1 Delivery report v2, Ops, 8 Jul 2026, pp.3–5: product teams met 38/45 milestones Jan–Mar and 31/39 Apr–Jun. April introduced a planning-system migration; seven delayed milestones were tagged migration-related. No workweek change occurred. S2 Staff pulse, People, 30 Jun, sec.2: 37/46 responded; 21 report frequent overtime, 10 occasional, 6 none; question does not define hours. 28/37 favor trying an optional four-day week. S3 Union memo, 2 Jul, p.1: cites S2's 28/37 and recommends a trial with no pay reduction; no new survey. S4 Vendor case study, rev May, pp.2,7: another 120-person software company reports on-time projects 74%→81% during a six-month four-day pilot; project definition changed in month three and no comparison group. S5 Finance note, 10 Jul, p.2: pilot analysis estimates USD 18,000 temporary coverage, based on two contractors for 12 weeks; rates listed, procurement not confirmed.
Evidence and table rules
Eligible: internal operational records, surveys with denominator, comparable external cases, and cost evidence. One row per measure/claim. Quality fields directness, method, independence, applicability, limitations. Short excerpts max 15 words. Unknown='Not reported'. S3 depends on S2. Do not compare milestone and project rates as identical. No personal data.
Example output
Source inventory: five current labeled sources; S3 is analytically dependent on S2, not corroboration. No missing pages are stated. S4 is external and its metric definition changed.
Evidence rows: E1/S1 p.3–5—delivery observation, Jan–Mar 38/45=84.4% milestones met; direct internal baseline, no pilot exposure. E2/S1—Apr–Jun 31/39=79.5%; seven delays tagged migration-related, but the tag does not prove all difference was caused by migration. E3/S2 sec.2—21/37 frequent, 10/37 occasional, 6/37 no overtime; direct self-report among respondents, undefined hours and 9 nonrespondents. E4/S2—28/37 respondents favor an optional trial; preference, not expected delivery effect. E5/S3 p.1—trial recommendation using E4; no independent observation. E6/S4 p.2,7—reported projects 74%→81%; indirect, changed definition, no comparison group, not numerically comparable with S1 milestones. E7/S5 p.2—estimated USD 18,000 coverage based on stated rates/two contractors/12 weeks; planning estimate, procurement unconfirmed.
Conflict/gaps: no source observes this department under a four-day schedule; no approved thresholds, overtime hours, workload baseline, eligibility/take-up estimate, customer impact, or pilot measurement plan. Weighted synthesis: internal evidence establishes current delivery and respondent strain but cannot predict treatment effect. External evidence suggests feasibility only weakly. Cost is estimable but not approved. The table supports designing a controlled, reversible pilot with preregistered milestone, overtime, workload, coverage, and stop criteria; it does not by itself support declaring the policy beneficial. Human verification: check S1 migration tagging and S5 rate arithmetic.
Why this works
1
Evidence-unit rows keep a document's data separate from its interpretation and recommendation.
2
A dependency map prevents several retellings of one dataset from appearing to be corroboration.
Check the result
Can every row be traced to a stable source, version, page/section, and exact evidence unit?
Are observations, author conclusions, recommendations, and analyst inferences distinct?
Are shared data, incompatible measures, conflicts, missing material, and applicability limits visible?
Use it with confidence
Frequently asked questions
Practical answers about when to use this recipe, what to provide, and where human review still matters
What should I prepare before using “Organize multiple sources into an evidence table”?
For “Organize multiple sources into an evidence table,” prepare Research or decision question, Labeled source materials, and Evidence and table rules. Replace placeholders only with information you can verify. If a detail is unknown, preserve that uncertainty explicitly instead of asking the model to infer it.
When is the “Organize multiple sources into an evidence table” result not ready to use?
The result is not ready if it does not yet deliver the stated outcome—A traceable evidence table with source inventory, claim clusters, conflicts, and synthesis limits—from the supplied evidence, or if it relies on unresolved assumptions, missing approvals, or invented details. Use the checks as release gates: revise the source inputs or assign a named, authorized reviewer instead of polishing an unsupported output.
Which AI tools have recorded tests for “Organize multiple sources into an evidence table”?
The published test record for “Organize multiple sources into an evidence table” lists ChatGPT as of 2026-08-28. This confirms recorded runs, not guaranteed compatibility or identical results in later product versions. For another tool or version, keep every constraint visible and repeat the result checks before use.
ChatGPT