Compare agreement and conflict across multiple sources
Author: AILesson9 min setupTested with:ChatGPTReviewed: 2026-08-28
Quick answer
Align source claims by scope and method before labeling them consistent or contradictory. Provide: Labeled source material, Comparison question, Evaluation rules. Expected result: A claim-level comparison matrix with genuine conflicts, explainable differences, and verification priorities.
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
Compare the supplied sources at claim level. Use only the provided material; do not fill gaps from memory.
Labeled sources and metadata:
[sources]
Exact question, population, outcome, geography, period, and decision:
[question]
Inclusion, evidence preference, terminology, materiality, and citation rules:
[rules]
Inventory every source and flag missing date, authorship, method, sample, denominator, version, or access. Extract atomic claims with direct source IDs and preserve wording strength. Normalize definitions, units, time windows, populations, comparators, and outcome measures before comparing. Never call two claims contradictory merely because one is silent, uses a different metric, studies another population, or reports a different period.
For each aligned claim classify the relationship as direct agreement, compatible/partial agreement, genuine contradiction, apparent conflict explained by scope or method, independent/non-comparable, or insufficient detail. Show supporting passages in short paraphrase, not long quotation. For conflict, identify exactly which propositions cannot both be true under the same conditions and assess recency, directness, method, sample, conflicts of interest, and correction status without collapsing them into an invented score. End with stable findings, unresolved disputes, missing perspectives, highest-value verification steps, and a source-to-claim coverage audit.
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
Labeled source material
A: City Library pilot report, 26 Aug 2026, internal analyst. 214 opted-in adults, six weeks; 61% collected before expiry versus 54% among 198 opted-in adults in prior six weeks; not randomized; seasonal mix differed. B: Vendor case study, 3 May 2025. Says reminder texts 'raised pickup by 18%' across 12 libraries; no numerator, denominator, assignment method, or definition; vendor sells messaging service. C: University working paper, 12 Jul 2026. Randomized 1,204 adult patrons across four urban libraries; collection 58% control and 63% reminder, difference 5 percentage points, 95% CI 1-9; excludes patrons without mobile number. D: Rural consortium dashboard, Jan-Jun 2026. 47% of all reservations collected before expiry; no reminder intervention and includes children.
Comparison question
For adult public-library patrons in 2025-2026, do reminder texts change collection before reservation expiry enough to justify a larger local randomized test? Decision is about testing, not permanent rollout.
Evaluation rules
Include empirical sources A-D but distinguish relevance. Prefer direct randomized evidence for effect. Use percentage points. Treat a difference of at least 3 points as operationally material for test planning. Cite by source ID; do not infer vendor methods.
Example output
Inventory: A has population, counts, periods, and limitations but no random assignment. B lacks definitions, denominators, assignment, and uncertainty and has a commercial interest. C supplies randomized method, population, comparison, and interval; publication review status is unknown. D supplies a local-sector baseline but neither adult-only population nor intervention.
Aligned effect claim: A reports a 7-percentage-point period difference; C estimates a 5-point randomized difference with 95% CI 1-9. These are compatible/partial agreement: direction and point estimates align, but A cannot attribute effect and C's interval includes effects below the 3-point materiality threshold. They do not jointly prove the local effect. B's “18%” is non-comparable because relative versus percentage-point meaning, denominator, and method are absent; it is not direct contradiction or corroboration. D's 47% is independent/non-comparable: it describes a different mixed-age population with no reminders, so it cannot conflict with A/C.
Stable findings: randomized evidence C supports a possible positive effect, with uncertainty spanning below and above the operational threshold; local A supplies feasibility and a descriptive signal, not causation. Genuine contradictions: none identifiable from supplied text. Unresolved: transfer from four urban libraries to this library, outcomes for patrons without mobile numbers, message timing, opt-outs, and staff workload.
Decision implication: the evidence supports designing a larger local randomized test, not permanent rollout or a promised effect size. Verify C's full protocol/correction status first; ask B for definitions only if it would affect design; use A to set local guardrails; do not use D as an effect estimate. Coverage: A→local descriptive signal; B→unverifiable non-comparable claim; C→effect evidence; D→context only.
Why this works
1
Aligning populations, measures, and periods prevents unlike findings from becoming false contradictions
2
Atomic claim mapping makes disagreement traceable to evidence rather than source reputation alone
Check the result
Is each finding linked to exact source IDs with scope and wording strength preserved
Are silence, different measures, and different populations kept separate from genuine contradiction
Do unresolved conflicts lead to concrete verification steps instead of a forced verdict
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 “Compare agreement and conflict across multiple sources”?
For “Compare agreement and conflict across multiple sources,” prepare Labeled source material, Comparison question, and Evaluation 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 “Compare agreement and conflict across multiple sources” result not ready to use?
The result is not ready if it does not yet deliver the stated outcome—A claim-level comparison matrix with genuine conflicts, explainable differences, and verification priorities—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 “Compare agreement and conflict across multiple sources”?
The published test record for “Compare agreement and conflict across multiple sources” 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