Your Claude prompt isn't working. Here's how to fix it.

When a Claude prompt produces a bad result, most people either try again with slightly different wording (and get a slightly different bad result) or give up and conclude that Claude cannot do the thing. Neither approach is efficient.

There is a better way: diagnose the specific problem before changing anything.

Almost every prompt failure falls into one of five categories. Identifying which one you are dealing with tells you exactly what to change.

The five categories of prompt failure

1. The task is unclear

Claude does not know what you actually want. The prompt is vague enough that multiple reasonable interpretations exist, and Claude picked the wrong one.

Signs this is your problem:

• Claude's response is technically correct but not what you needed
• The output format is wrong (you wanted a list, Claude wrote paragraphs)
• The length is off (too long, too short, too detailed, not detailed enough)
• Claude interpreted the task differently than you intended

The fix: Add specificity. Name the output format, the length, the audience, and the constraints explicitly.

Before: "Write something about our new feature."

After: "Write a 150-word feature announcement for our product newsletter. The audience is existing customers who already use our core product. Focus on one key benefit — faster data import — and end with a link placeholder. No jargon."

---

2. Claude is missing context

Claude does not have information it needs to do the task well. It is making reasonable guesses about your situation, your audience, or your constraints — and guessing wrong.

Signs this is your problem:

• The response is generic and could have been written by anyone
• Claude got the tone wrong (too formal, too casual, too safe)
• Claude does not seem to understand your specific situation
• The content is accurate but not relevant to your actual use case

The fix: Tell Claude more about the situation. Who is the audience? What is the purpose? What has already happened? What constraints apply? The more Claude knows about your specific context, the less it has to guess.

A quick test: would a smart new employee, reading only your prompt, know exactly what you need? If not, they are missing context. Add it.

---

3. Claude is solving the wrong problem

You asked for X, but what you actually need is Y. This happens when you describe how you want something done rather than what you need to achieve.

Signs this is your problem:

• Claude does exactly what you asked for, but it does not solve your underlying problem
• You find yourself heavily editing the output to add the thing that was actually missing
• The response addresses the surface question but misses the real issue

The fix: Describe the goal, not the method. Instead of "write three options for this subject line," try "I need this email to get a 40% open rate from a list of people who have already ignored our last two messages. The subject line should create urgency without sounding desperate. Give me five options."

You are telling Claude what you need to achieve, not just what to produce.

---

4. Claude needs constraints it does not have

Claude is producing something that technically fits your description but is doing things you do not want — being too formal, too verbose, using buzzwords, generating content that does not match your brand, etc.

Signs this is your problem:

• The output is fine except for one persistent thing that keeps being wrong
• Claude keeps including something you do not want (caveats, bullet points, qualifications)
• The tone is off in a specific, repeatable way

The fix: Add explicit constraints. Instead of hoping Claude will intuit your preferences, state them directly.

"Do not use the word 'leverage.' Do not start with 'I.' Do not include qualifications like 'it's worth noting' or 'it's important to consider.' Keep each bullet to one line."

Negative constraints (what not to do) are often more effective than positive ones (what to do) for correcting persistent style problems.

---

5. The task is too big for one prompt

You are asking Claude to do too much in a single prompt. The output is unfocused because the prompt is unfocused.

Signs this is your problem:

• Claude gives you something that covers everything but does nothing well
• The response is a list of every possible point rather than the most important ones
• You wanted depth on one thing but got breadth across everything

The fix: Break the task into smaller pieces. Run them in sequence rather than in parallel.

Instead of: "Research our competitors, identify our differentiation, write positioning copy, and suggest a pricing strategy."

Run in order:

1. "Here are five competitors. Identify their key positioning claims."
2. "Based on those claims, what are the gaps they are not addressing?"
3. "Write three positioning statement options that target those gaps."
4. "For each option, suggest a pricing tier that supports it."

Each step produces something you can review before building on it.

---

The debugging process

When a prompt does not work, run this in order:

1. Read the output and name what is wrong. Not "it is not right" — specifically: wrong format, wrong tone, missing context, too generic, wrong interpretation of the task.

2. Match the problem to one of the five categories above.

3. Make one targeted change. Not a complete rewrite — a specific adjustment to address the specific problem.

4. Try again in the same conversation. "That is not quite right — [specific issue]. Try again with [specific change]."

Most prompt problems require two or three iterations, not ten. The key is being specific about what went wrong rather than throwing a completely different prompt at it.

The meta-skill

Getting good at prompting is mostly getting good at describing what you want clearly. The same skills that make you good at writing a brief, giving feedback, or explaining a problem to a colleague make you good at prompting.

The shortcut: before writing a prompt, write out in plain language exactly what you want Claude to do, who it is for, what format it should be in, and what the output will be used for. Then paste that as your prompt. It will not be perfect the first time — but you will know exactly what to adjust.

Further reading

• Prompt engineering documentation — the full reference for structuring prompts
• Improve your prompts in the developer console — Anthropic's built-in prompt improvement tool