How to Use a Claude “Grill Me” Prompt

Use a Claude “grill me” prompt by naming the subject, giving it reliable material, and asking for one challenging question at a time. Tell it to wait for your answer, explain any gap, and end with a short list of what to practice or change.

Treat “grill me” as a prompting pattern, rather than a special command you must find. fenbs can keep the version of the prompt you approve and the lessons from each session, so you can repeat the exercise without rebuilding the instructions.

How do you start a useful questioning session?

1. Choose one purpose: rehearse an explanation, test your knowledge of supplied material, or examine the weak points in a proposal.
2. Provide the source or proposal. Name what is in scope and what should be left out.
3. Set the difficulty and audience. A first practice session needs different questions from a final presentation rehearsal.
4. Ask for one question at a time and a pause for your answer. Define whether feedback comes immediately or after the full round.
5. Set a finish line, such as five questions, then request a summary of gaps and next actions.

Start small enough that you will finish. A topic such as “everything about marketing” encourages scattered questions and vague feedback. “Question me on the three audience assumptions in this campaign proposal” makes the exercise assessable. You can see whether the questions actually test what you wanted to examine.

Help me test my understanding of [topic].
Use only the material below as the answer reference.
Ask one question at a time and wait for my answer.
Start at [difficulty] for an audience of [audience].
After each answer, state what was correct, what was missing, and the relevant source passage.
If the source does not settle the point, say so.
After five questions, summarize my gaps and give me three practice tasks.
Material: [paste or attach approved source]

The structure follows the general principle in Anthropic’s prompting guidance↗ (opens in a new tab): make the requested output and steps explicit. The important instruction is to wait. Without it, you may receive a complete question sheet and answers before you have practiced recalling anything.

How can Claude test what you actually know?

Ask for a mix of question types. Definitions test whether you recognize terms. Explanations test whether you can connect ideas. Scenarios test whether you can apply them when the wording changes. A balanced session can reveal a gap that memorized phrases would hide.

For example, suppose you are preparing to explain a design system to a new teammate. Give Claude the approved documentation. Ask it to test the difference between a component and a usage rule, then present a case where the teammate must choose between them. Check its feedback against the supplied documentation.

• Recall: explain a concept in your own words without looking at the reference.
• Application: decide what to do in a realistic example and state the relevant rule.
• Comparison: explain when one approach is appropriate and another is less suitable.
• Teaching: answer a short follow-up from someone who has never encountered the topic.

Specify whether hints are allowed. If you want to practice retrieval, ask for a hint only after you request one. If you are learning a new subject, a short hint may help you continue. Keep the session’s purpose clear so a change in difficulty does not look like evidence that your knowledge suddenly improved.

How do you challenge a proposal instead of studying?

Change the role from examiner to skeptical reviewer. Give the proposal’s goal, constraints, evidence, and open choices. Ask Claude to question assumptions that would change the recommendation. Mere disagreement is not a useful test; a question should reveal information you can gather or a choice you can make.

Question this proposal as a skeptical but constructive reviewer.
Goal: [desired outcome]
Constraints: [time, people, scope]
Proposal and evidence: [material]
Ask one question at a time about the weakest important assumption.
After my answer, explain whether that assumption is supported, unresolved, or contradicted by the evidence.
Do not invent facts about customers or results.
Finish with the questions we still need to answer before a person decides.

Say you are proposing a shorter client intake form. Good questions might explore which information is necessary before the first call or how missing information will be collected later. A question about an unrelated company expansion would not help evaluate that proposal. Redirect the assistant when the discussion leaves the agreed decision.

Keep the reusable questioning method in fenbs Skills and prompts, while the actual choice belongs in Decisions and rules after a person makes it. A persuasive AI critique is still a proposal for your consideration.

How do you keep the feedback fair and grounded?

Define the assessment criteria before answering. For a concept explanation, they might be factual accuracy, completeness, and a relevant example. For a proposal, they might be evidence, feasibility, and fit with the stated goal. Do not let polished wording become the only sign of a good answer.

Claude can produce incorrect information, including during an exercise. Anthropic’s guidance on reducing hallucinations↗ (opens in a new tab) recommends allowing uncertainty and grounding claims in supplied material. Ask for the supporting passage when feedback depends on a fact. Check that the passage actually supports the correction.

If the source is ambiguous, mark the result unresolved. Do not count an unsupported correction as a knowledge gap. Likewise, a score from Claude is a practice aid, not an official grade or evidence that a proposal has passed an independent review. Your criteria and source quality limit what the exercise can establish.

What should you save after the questions?

Save the specific gap and the action it calls for. “Study more” provides no useful starting point. “I confused the exception with the default rule; review the two examples and explain the distinction again” gives the next session something concrete to test.

Separate what you learned from what the assistant merely suggested. A lesson should reflect something supported by the session. A proposal may produce an unanswered research question instead. Keep that question open until you have evidence, rather than turning it into a confident project fact.

How to do this with fenbs

1. Create a free account, free for one person with no credit card, and add a project for the subject or proposal you are working on.
2. Save the questioning prompt in Skills and prompts. Include its purpose, source requirements, question cadence, feedback criteria, and stopping point.
3. Connect Claude through Integrations. Ask it to read the project context and saved prompt before starting the questioning session.
4. Record supported lessons and add tasks for the remaining work. Save progress with the next question to revisit before you stop.

fenbs does not grade your answers or verify Claude’s feedback. For a quick practice round, a single chat is enough. The shared record helps when you want the prompt, lessons, and outstanding work available in later sessions.

If you are new to the assistant, read what Claude is. You can adapt the same written questioning pattern for Claude Code when discussing code, while the AI prompts hub has other reusable conversation formats.