Define a verifiable output format for a prompt

Author: AILesson6 min setupTested with:ChatGPTReviewed: 2026-08-28

Quick answer

Convert vague output wishes into fields, types, statuses, constraints, and validation rules. Provide: Prompt task and downstream use, Current output request and examples, Format and validation constraints. Expected result: An output contract with schema, human-readable example, empty states, and validation checks.

Add your context

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

Prompt task and downstream useCurrent output request and examplesFormat and validation constraints

Your prompt

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

Design a verifiable output contract for the prompt task.

Input, transformation, reader/system, decision, and failure consequences:
[task]

Current format wording, desired examples, rejected outputs, and used fields:
[output]

Serialization, parser limits, length, locale, ordering, required/optional fields, and review:
[constraints]

Define the smallest output structure that supports the downstream use. For every field specify name, purpose, type, required status, allowed values or pattern, units, evidence link, unknown/empty representation, and cross-field rules. Separate data from commentary and machine fields from reviewer notes. Avoid asking for mutually incompatible formats, ornamental sections, or prose inside fields meant for parsing. Include deterministic ordering where relevant. Do not use null, empty string, zero, false, and “unknown” interchangeably. Return: output-design rationale; formal or tabular schema; exact prompt clause; valid complete example; valid empty/unknown example; invalid examples with reasons; validation checklist or pseudocode; and human-review fields that cannot be mechanically verified.

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

Prompt task and downstream use: Extract explicit commitments from a meeting transcript. A workflow imports the result into a task-review screen; no task is created automatically. Reviewers need action, owner, due date, status, and evidence locator. Missing owners/dates must remain visible. Treat suggestions as non-commitments. A wrong owner or invented due date creates accountability risk.
Current output request and examples: Current instruction says ‘Give me a clean structured list.’ Good output had one row per commitment and timestamps. Rejected outputs mixed a summary with bullets, used ‘soon’ as a due date, and omitted owner when absent. Reviewers do not use priority or sentiment.
Format and validation constraints: Strict JSON, no Markdown fences or extra prose. Parser accepts UTF-8 JSON array. ISO 8601 dates only when exact calendar date is stated. Required keys in fixed order for snapshots. Status values proposed, confirmed, completed, cancelled. Evidence contains speaker and timestamp. Human reviews commitment classification and identity.

Example output

Design: one array item per explicit commitment; omit summaries, priority, and sentiment. Unknown owner/date use null because absence is meaningful; an empty string is invalid. “Soon” remains in evidence but cannot populate due_date.

Schema: action string, required, nonblank; owner string|null, required key; due_date string|null, required key, YYYY-MM-DD only; status enum, required; evidence object, required, with speaker nonblank string, timestamp HH:MM:SS, excerpt nonblank string ≤200 characters. Key order: action, owner, due_date, status, evidence. Cross-field rule: completed status requires evidence that completion was stated; due_date must be null unless an exact date is in evidence.

Exact prompt clause: “Return only a UTF-8 JSON array. Each explicit commitment is one object with keys in this order: action, owner, due_date, status, evidence. Use null for an unstated owner or exact due date; never omit keys or convert relative wording to a date. status must be proposed, confirmed, completed, or cancelled. evidence must contain speaker, HH:MM:SS timestamp, and a source excerpt no longer than 200 characters. Return [] when no explicit commitment exists. Add no prose or Markdown.”

Valid: [{"action":"Send the revised budget","owner":"Nora","due_date":null,"status":"confirmed","evidence":{"speaker":"Nora","timestamp":"00:14:22","excerpt":"I’ll send the revised budget soon."} }]. Valid empty: []. Invalid: due_date "soon" fails ISO rule; omitted owner fails required-key rule. Human review still verifies whether the excerpt is a commitment and whether “Nora” resolves to the correct person.

Why this works

1
Typed fields and explicit empty states turn output quality into observable rules.
2
Cross-field validation catches internally inconsistent outputs that look well formatted.

Check the result

Does every field serve a real downstream decision, display, or check?
Are unknown, absent, not applicable, and zero values unambiguous?
Can mechanical checks and human judgment responsibilities be distinguished?

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 “Define a verifiable output format for a prompt”?

For “Define a verifiable output format for a prompt,” prepare Prompt task and downstream use, Current output request and examples, and Format and validation constraints. 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 “Define a verifiable output format for a prompt” result not ready to use?

The result is not ready if it does not yet deliver the stated outcome—An output contract with schema, human-readable example, empty states, and validation checks—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 “Define a verifiable output format for a prompt”?

The published test record for “Define a verifiable output format for a prompt” 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.

More ways to explore