How to Make Claude Brutally Honest About Your Work

To make Claude brutally honest about your work, request direct criticism tied to evidence, review criteria and specific improvements. Ask it to state what it cannot judge, and make clear that insults or invented flaws are not useful feedback. fenbs can retain the review prompt and the changes a person approves so that the next critique starts from the same standard.

The useful part of “brutally honest” is permission to name a real problem plainly. A harsh tone does not make the criticism accurate. You want to learn which part fails, why it matters and what you can do about it.

What should you ask Claude to review first?

1. Provide the actual work: the draft, proposal, design description or code that needs review.
2. Explain its audience, goal and constraints so the review has a clear standard.
3. Choose a few criteria, such as clarity, evidence, feasibility and usefulness to the reader.
4. Request the most consequential weaknesses, each with a location, reason and suggested fix.
5. Ask Claude to distinguish an observed flaw from a preference or an unanswered question.

Anthropic’s prompting documentation↗ (opens in a new tab) emphasizes clear instructions and relevant context. For a review, that means explaining what the work must accomplish before asking whether it succeeds. “Tell me if this is bad” gives the reviewer no usable threshold.

Review the work below for its stated audience and goal. Give a candid verdict, then up to three weaknesses ranked by their effect on the outcome. For each, identify the passage or feature, explain the problem, cite the evidence and propose a concrete improvement. Separate factual errors, missing evidence and personal preferences. Say when you cannot judge. Keep useful strengths that the revision should preserve. Be direct and respectful.

Audience:
Goal:
Constraints:
Work:

For a broader introduction to the assistant, see what Claude is. Here, start with one piece of work you can revise today rather than asking for a sweeping judgment of your ability.

Why is asking for harshness a poor test of honesty?

A sentence can sound fearless while making an unsupported claim. “Nobody will buy this” is forceful, but a draft usually cannot establish how every potential customer will behave. “The proposal never identifies the buyer’s problem” is a claim you can inspect and repair.

Set a standard for criticism that is equally demanding whether the verdict is positive or negative. Claude should support praise with a visible strength and criticism with a visible weakness. If the work meets the criteria, permit it to say so rather than requiring an arbitrary number of faults.

Avoid a forced persona such as an abusive boss or a hostile investor. Describe the review behavior you need: name the problem promptly, explain its consequence and show the smallest useful correction. Respectful wording leaves more attention for the substance of the review.

What should an actionable criticism contain?

Ask for enough detail to check the judgment and act on it. The following is an illustrative review of a service proposal, not a report about an actual business. Notice how a specific observation creates a practical revision instead of merely changing the mood.

Rank issues by their likely effect on the goal. An unsupported promise matters more than an awkward heading. Have Claude identify the two or three changes worth making first so a review does not become a long inventory of minor preferences.

For visual work, describe the audience and intended action as carefully as you describe appearance; our Claude Design guide introduces that setting. The review still needs criteria a person can inspect rather than a broad claim that a design feels wrong.

How do you keep uncertainty in the verdict?

Require a distinction between what the work shows and what would need outside evidence. A missing explanation is visible in a proposal. Whether the proposed service has market demand requires information beyond that text. Claude should not turn the first observation into proof of the second.

Anthropic’s guidance on reducing hallucinations↗ (opens in a new tab) recommends allowing uncertainty and grounding claims in sources. In a critique, explicitly permit “I cannot determine that from this material.” Ask which missing fact would most affect the conclusion.

For the claim that this plan will fail, separate: what is directly supported by the supplied material; what you inferred; what remains unknown. Replace unsupported certainty with a narrower statement. What one piece of evidence would most change your assessment?

Read the cited passages yourself. A reference makes a claim easier to inspect, but it does not guarantee the interpretation is sound. If the critique concerns an important technical or professional decision, use an appropriate independent check before accepting it.

Save the agreed review standard with its scope. Decisions and rules in fenbs can keep a person’s criteria visible to connected assistants without turning an uncertain critique into an approved decision.

How should you respond when the criticism is wrong?

Challenge the claim with the relevant evidence. If Claude says the proposal omits a deadline, point it to the paragraph containing the deadline and ask it to reconsider that finding. Do not ask it to defend its first verdict regardless of what the evidence now shows.

Then ask whether the correction changes the priority of the remaining issues. A flaw in one criticism does not prove the whole review is useless, and a useful suggestion does not validate every claim. Approve changes individually according to the goal and constraints.

Keep the final choice separate from the assistant’s recommendation. You might accept a clearer opening and reject a change that conflicts with a client requirement. Record that reason so the next review does not reopen the same suggestion without new evidence.

If you use another assistant for the same work, the separate guide to making ChatGPT brutally honest covers that neighboring workflow.

How can you tell whether the review helped?

Revise the highest-priority issue and compare the result against the original goal. Can a reader now identify the problem, understand the promise or follow the required action? A useful critique should lead to an observable improvement or a clear reason to keep the original.

Test the prompt on work with a known strength as well as work with a known flaw. If it invents problems in both, add a stronger evidence requirement. If it overlooks the flaw, make the relevant criterion more explicit. Review quality is the test; severity of tone is not.

How to do this with fenbs

1. Create a free account and add the project whose work needs review.
2. Save the critique prompt in Skills and prompts, including audience, criteria and the requirement to identify uncertainty.
3. Ask a connected assistant to read the prompt and project context before reviewing the work you supply.
4. Choose which changes to accept. Have the assistant record your decision and reasons, then save the remaining revision tasks and progress.

fenbs helps when review standards and approved changes need to survive the conversation. It does not certify a critique or replace your judgment; a single review can stay in a chat. Explore the AI prompts hub for more ways to make a request assessable.