The ACDQ Checklist for ChatGPT Prompts

ACDQ is a way to check whether a prompt names the task, supplies context, defines useful evaluation and leaves room for clarification. Treat it as a briefing aid, not a tested formula that guarantees a better answer.

Key Takeaways

• Specify the work rather than inventing professional credentials.
• Supply constraints and evidence the answer must respect.
• Ask for assumptions and tradeoffs you can inspect.
• Verify the output against the original task.

What the four letters mean

Element	Useful question	Example
Act	What work should the assistant do?	Compare two onboarding sequences
Context	What facts and limits matter?	Audience, current flow, permissions and deadline
Deliberation	What should the evaluation consider?	Error recovery, implementation effort and unresolved risks
Questions	Which missing facts could change the answer?	Whether signup is required before browsing

The original shorthand used “Deep Thinking.” Here, the useful interpretation is explicit evaluation criteria, not the phrase “think deeply” as a guarantee. Ask for conclusions, assumptions and supporting reasons you can check.

Rewrite a vague request

Vague: “Act as a world-class marketer and make me a content plan.”

A more reviewable prompt:

Plan four posts for a small hotel explaining room choices.

We have approved photographs of three room categories.
We can publish once a week and have no verified customer-result data.
Our audience is couples comparing a short city stay.
The goal is to answer practical questions before an availability search.

For each post, give the audience question, proposed angle,
required asset, draft caption and facts we must verify.
Do not invent testimonials, rates, inclusions or availability.
Explain which ideas depend on missing information.
Ask up to three questions if their answers would change the plan.

This is an illustrative brief. Its improvement is visible in the constraints and output requirements; it is not evidence of a measured model-performance gain.

Evaluate the answer against the brief

Check whether the output used the supplied facts, respected exclusions and exposed uncertainty. A fluent answer can still invent an offer, miss a constraint or produce something impossible to film.

Use a simple review table: requirement, where it appears in the answer, problem and correction. If the output fails, revise the missing context or ambiguous instruction rather than adding more claims about the model's supposed expertise.

The content-calendar prompt and video concept prompt apply this structure to specific tasks.

Do not turn a role into authority

“Act as a lawyer,” “certified trainer” or “security expert” does not grant qualifications or make advice correct. For consequential decisions, use appropriate verified sources and qualified review. Keep private information out unless the workflow and permissions allow it.

For code or factual research, require a way to inspect the result: tests, source references or explicit unknowns. Do not accept fabricated citations or a claimed test run without evidence.

Keep the checklist small

For a simple rewrite, one clear sentence may be enough. Add context when it changes the answer and questions when the missing facts matter. The project-planning essay makes the same point about human briefs: uncertainty should be managed, not disguised by confidence.