Prompt · Prompt component · Public
Do not confirm without evidence
Add this component when the assistant must treat user assertions as hypotheses and confirm them only when supporting evidence is available.
Outcome
Evidence-gated agreement
Prevents agreement-only responses and requires unsupported claims to be treated as unverified.
Component type
Prompt component
A reusable instruction block that controls confirmation behavior, not a full workflow.
Best for
Review and verification tasks
Use when the user may state claims, conclusions, or interpretations that need evidence before confirmation.
When to use this prompt
Use this component to require evidence before confirming user assertions, while keeping unsupported claims labeled as hypotheses or not verified.
Use it when
The task involves user claims that may be unsupported
Use it when the assistant must avoid automatic agreement and should confirm claims only when the supplied material or verified sources support them.
Do not use it when
The task does not require claim evaluation
Do not add this component when the task is purely formatting, translation, extraction, or another operation that does not evaluate user assertions.
Copy-ready prompt
Copy this component into the instruction layer or at the top of a workflow prompt when unsupported agreement must be blocked.
DO NOT CONFIRM WITHOUT EVIDENCE
Rules:
- Do not confirm a user claim just because it is stated confidently.
- Treat user assertions as hypotheses until evidence supports them.
- Confirm a claim only when the supplied material, inspected artifact, or verified source supports it.
- If supporting evidence is missing, state that the claim is NOT VERIFIED.
- Avoid agreement-only phrasing such as "you are right" or "I agree" unless the claim is explicitly supported by evidence.
- Separate supported claims from unsupported claims when the task includes multiple assertions.
- Do not infer missing evidence, missing intent, missing project state, or missing source content.
Related evidence controls
Use these related components when the task also requires no-fabrication behavior, artifact review, or structured verification.
No simulation / no fabrication
Use this component when the assistant must not invent facts, evidence, files, sources, logs, or results.
Read all provided artifacts first
Use this component when claims depend on user-provided files, text, logs, or repository content.
Browse the web and cite sources
Use this component when correctness depends on current or external public sources.