Define a verifiable output format for a prompt

Автор: AILesson6 мин на настройкуПроверено на:ChatGPTПроверено: 2026-08-28

Быстрый ответ

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

1

Добавьте контекст

ваш текст остаётся в этом браузере. AILesson Prompts не отправляет его ни в модель, ни на сервер.

2

Ваш промпт

Незаполненные поля остаются видимыми как заполнители, поэтому вы всё равно можете скопировать и отредактировать промпт

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.
Попробовать в Playground
Конфиденциально по умолчаниюПромпт собирается локально в вашем браузере. Не вводите конфиденциальную информацию в сервисы ИИ, если ваша организация этого не разрешает.

От исходных данных к результату

Разобранный пример

Посмотрите, как конкретный контекст превращает этот рецепт в полезный результат

Реальный ввод

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.

Пример вывода

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.

Почему это работает

  1. 1

    Typed fields and explicit empty states turn output quality into observable rules.

  2. 2

    Cross-field validation catches internally inconsistent outputs that look well formatted.

Проверьте результат

  • 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?

Используйте уверенно

Часто задаваемые вопросы

Практические ответы о том, когда использовать этот рецепт, что нужно предоставить и где по-прежнему важна проверка человеком

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.

Другие способы изучения

Где этот рецепт применим

Продолжайте работу

AILesson · Рекомендуемые курсы

ваш следующий шаг: примените ИИ на практике

Перейдите от понимания ИИ к выполнению задач. Практикуйтесь в составлении запросов, проверке и улучшении результатов с помощью интерактивных уроков для работы и повседневной жизни.

Просмотреть все курсы