Distinguish facts, inferences, and opinions in source material
Classify atomic statements by what the source observes, concludes, values, recommends, or leaves uncertain
6 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
Classify statements in the source as facts, reported claims, inferences, predictions, opinions, value judgments, recommendations, or uncertainty using the supplied rules.
Source material:
[material]
Genre, author role, audience, and purpose:
[context]
Classification definitions:
[rules]
Split mixed sentences into atomic statements. A fact label means the source presents a checkable statement, not that you have independently verified it; distinguish source-reported fact from externally verified fact. Identify linguistic signals, evidence or premise, missing inference steps, modality, attribution, and what could verify each statement. Do not label a statement opinion merely because it is controversial, or fact merely because it contains a number. Return a statement table, inference chains, mixed/ambiguous cases, places where grammar hides the actor or certainty, and a neutral rewrite that preserves attribution and epistemic status.
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
Source material
[Director's memo, §2] Customer complaints rose from 42 in April to 57 in May. The increase clearly shows that the new routing system is confusing customers. Because the system is fundamentally flawed, we should restore the old process immediately. Operations says the May increase may partly reflect a campaign that generated 18% more orders, but its analysis is not complete. If the current pattern continues, complaints will exceed 70 in June.
Context and purpose
Internal decision memo written by the service director for executives. We need to prepare a neutral evidence brief before a rollback decision. No external verification has been performed.
Classification rules
Fact: source-presented checkable past/current statement. Reported claim: attributed checkable statement not independently verified. Inference: conclusion from stated/unstated premises. Prediction: future claim. Opinion/value judgment: evaluative standard. Recommendation: proposed action. Uncertainty: explicit limitation or modality. One statement may receive a primary and secondary label only when inseparable.
Example output
S1 Complaints were 42 in April and 57 in May | source-presented fact, not independently verified | signal: past counts | verify against complaint-system extracts, definitions, and monthly denominators. S2 The routing system confused customers | inference presented with high certainty (“clearly”) | premise: count increase; missing: complaint reasons, exposure, order volume, comparison, timing. S3 The system is fundamentally flawed | value judgment plus inference | “fundamentally” lacks a stated standard. S4 Restore the old process immediately | recommendation | depends on S2/S3 and unprovided rollback costs/risks. S5 Operations says orders rose 18% due to a campaign | reported claim; causal attribution remains unverified | verify order data and campaign design. S6 Operations analysis is incomplete | reported fact/uncertainty. S7 Complaints will exceed 70 in June | conditional prediction | depends on undefined “current pattern”; calculation route not shown.
Inference chain: 42→57 complaints → assumed routing confusion → judged fundamental flaw → immediate rollback. The observations do not supply the middle causal steps.
Neutral rewrite: “The memo reports 42 complaints in April and 57 in May; the counts have not been independently verified here. The director attributes the increase to customer confusion caused by the new routing system and recommends immediate rollback, but the memo does not provide complaint-reason or exposure analysis. Operations reports 18% more May orders associated with a campaign and says its analysis is incomplete. A June count above 70 is a conditional forecast whose method is not stated.”
Why this works
1
Atomic classification reveals when one sentence combines observation, explanation, and recommendation.
2
Separating source-reported from independently verified facts prevents analysis from laundering claims.
Check the result
Were mixed sentences split before classification?
Does each label explain its evidence, linguistic signal, attribution, and verification route?
Are numbers, controversy, confidence, and truth kept conceptually separate?
ChatGPT