Research

Create a verification plan for a claim

Decompose a claim into testable parts, evidence standards, source routes, and stopping rules

7 min setupTested with:ChatGPTReviewed: 2026-08-28

Add your context

Your text stays in this browser. AILesson Prompts does not send it to a model or server.

Exact claimDefinitions and scopeResearch constraints

Your prompt

Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt

Create a verification plan for the exact claim.

Claim, speaker, date, context, and intended use:
[claim]

Population, measure, period, geography, comparison, causality, and materiality:
[scope]

Time, access, existing sources, excluded methods, risk, and confidence needs:
[constraints]

Do not decide whether the claim is true yet. Parse it into atomic factual, definitional, comparative, causal, predictive, and normative components. Identify ambiguous terms, hidden denominators, time sensitivity, attribution, and the strongest plausible alternative interpretations. For each testable component provide what evidence would support, weaken, contradict, or leave it unresolved; preferred primary sources and fallback sources; exact retrieval or calculation steps; source independence and conflict checks; bias and data-quality risks; and a record template. Include falsification attempts, alternative explanations, decision labels, escalation for high-stakes uncertainty, and a stopping rule tied to the intended use.

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

Exact claim: A vendor sales deck dated 2 August 2026 says: ‘Our automated invoice matching cuts processing time by 60% and pays for itself within three months for mid-sized retailers.’ Procurement wants to cite this claim in an investment memo.
Definitions and scope: Mid-sized retailer is undefined. Our company processes 24,000 invoices/year across three currencies. Compare end-to-end staff time from invoice receipt through approved posting, not only matching step. UK operations. Need to separate observed vendor-customer results, causal effect, and payback forecast.
Research constraints: Five working days. Vendor may provide methodology and references; we can inspect internal time/cost baselines but cannot run a pilot yet or contact named customers without permission. Investment memo is a material £90k decision, so vendor marketing alone is insufficient. Current pricing quote expires 31 August 2026.

Example output

Atomic components: C1 the product is automated invoice matching; C2 “processing time” fell 60% in a defined population/period; C3 the product caused that change; C4 results apply to a retailer comparable to ours; C5 total costs are recovered within three months; C6 the statement remains current on 2 August. “Processing,” “mid-sized,” baseline, sample, currencies, exception rate, implementation cost, and payback formula are ambiguous.

C2 evidence: request the study protocol, before/after distributions, denominator, dates, included workflow steps, exception handling, and raw/aggregated records; support requires consistent end-to-end measure, while matching-step-only timing weakens applicability. C3: seek comparison design and concurrent process changes; alternative explanations include staffing, policy, invoice mix, and learning. C4: compare volume, currencies, ERP, exception mix, and staffing. C5: reproduce payback using £90k plus implementation, integration, training, support, and internal change cost against verified labour savings; test low/base/high adoption rather than accept one forecast.

Sources: vendor methodology and contract terms; independent audit or permitted reference validation; internal time sample, loaded labour costs, invoice/error/rework data; official product/security documentation. Record claim, exact source/date, method, population, measure, result, limitation, independence, status. Labels: supported, partly supported, contradicted, or insufficient. Stop after calculations reconcile and at least one independent evidence route supports material inputs; otherwise state “insufficient for memo” and recommend pilot/contractual performance condition.

Why this works

1
Atomic decomposition prevents one true detail from lending credibility to a broader unsupported claim.
2
Falsification and stopping rules make verification a bounded test rather than endless source collection.

Check the result

Are all factual, comparative, causal, and predictive components identified separately?
Does the plan actively seek contradictory evidence and independent sources?
Is the evidence threshold proportionate to the claim's intended use and risk?

More ways to explore