Claude Prompt Generator

What is the Claude Prompt Generator?

The Claude Prompt Generator is a free tool that turns a task description into a ready-to-use prompt for Anthropic's Claude. It comes with a role, the task, constraints and an output format, and you choose whether it comes wrapped in XML tags, as Anthropic recommends for long prompts, or as plain prose.

How to use the Claude Prompt Generator

1. Describe what you want Claude to do in the first box, such as "Write a launch email for a technical SEO product".
2. Add the audience, tone, constraints and anything Claude must not say in Context. It is optional, but it is what makes the prompt reusable.
3. Choose XML tags or Plain text under Structure, then press Generate prompt.
4. Copy the prompt from the result panel, or press Open in Claude. The first time, sign up free to reveal the result.

Why use XML tags in Claude prompts?

Anthropic's prompting guidance recommends XML tags to separate the parts of a prompt: instructions, context, examples, input data and output format. Tags stop Claude mixing up your instructions with your material, and let you point at a part by name, as in "using the report in <report> tags".

No tag names are special. Claude was not trained on a fixed set, so pick names that describe their contents and use them consistently. A short prompt reads fine as prose. Tags pay off when one prompt carries reference material, rules and a required format together.

What does the generated Claude prompt contain?

It is written to give Claude a role, the task, the constraints it has to respect and the shape of the answer you want back. The answer format is the part people skip, and it is why two runs of the same prompt come back looking nothing alike.

It also tells Claude what to do when information is missing. Left unsaid, a model fills the gap quietly, and you cannot tell the invented part from the answer.

Does the prompt go in the system prompt or the user message?

Split it. On the API, Claude's role and standing rules go in the system parameter, and the task and the material you want worked on go in the user message. In the Claude app there is no system field, so paste the prompt as the first message, or save the standing part as a Project's instructions and send only the task.

Keep your input apart from the instructions either way. Pasted together as one block, Claude has to guess which lines are orders and which are data, which is the confusion tags exist to prevent.

What does a good Claude prompt template look like?

A good Claude prompt template has fixed slots for the role, the task, the material, the rules and the output format, with the material kept in its own block. In tag form that is a short role line, then <instructions>, <context>, <document> or <input>, <constraints> and <output_format>, each holding one kind of thing. In plain form it is the same order with a heading or a blank line between parts.

The template earns its keep when you reuse it. Write the standing parts once, leave the input block empty, and each new run means pasting one document, not rewriting the instructions. If you change the template, change one slot at a time and run it again on the same input, so you know which change moved the output.

The generator gives you a first version of this template filled in for your task. Treat the role and constraints as a draft: they come from what you typed in Context, and anything you did not mention has been inferred.

Choose XML tags when the prompt will carry a document, a set of rules and a required format together. Choose Plain text for a short request, or when people who find tags noisy will read and edit the prompt. Claude handles both; the tags are there to keep long prompts unambiguous.

Should you give Claude examples in the prompt?

Yes, whenever the shape of the answer matters more than its wording. One example of the finished output teaches length, tone and layout faster than a paragraph describing them. Anthropic's guidance is to wrap examples in their own tags, such as <example>, so Claude reads them as samples rather than as instructions or as the material to work on.

Use a few examples that differ from each other. A single example gets copied too closely: the same opening line, the same number of points, sometimes the same facts. Two or three that vary in topic but share the format show Claude which parts are fixed and which are free.

The generator does not invent examples for you, because a made-up example is worse than none. Add your own below the generated prompt, inside an example tag if you chose XML.

If the output must follow a strict structure, such as JSON for a script, show the exact keys in the example and say that nothing should appear outside it.

How do you stop Claude inventing answers?

Give Claude explicit permission to say it does not know, and tell it what to do instead. A line such as "if the document does not contain the answer, say so and stop" removes the pressure to fill the gap. The generator already writes an instruction for missing information into every prompt; check that it names the behaviour you want.

For work over a long document, two habits help. Put the document near the top and the question at the end, which Anthropic's long-context guidance recommends. And ask Claude to quote the passages it relied on before it answers, so every claim can be traced to a line you can check.

None of this makes an answer correct. It makes a wrong answer easier to spot, which is the realistic goal.

When the task is analysis rather than lookup, ask for the evidence and the answer in separate tags, such as <quotes> and <answer>, so you can read the evidence first.

Why is my Claude prompt not working?

Usually one of four things. The instructions contradict each other, such as "be thorough" next to a tight word limit, and Claude satisfies one at the expense of the other. The output format is described rather than shown. The material is pasted into the instructions, so Claude treats part of your document as orders. Or the tags are inconsistent: a block opened as <report> and referred to later as "the document".

Recent Claude models follow instructions closely and do what you asked, not what you meant. If you want extra effort, a particular level of detail or a suggestion beyond the brief, say so. If a prompt that worked last month now gives a different kind of answer on a newer model, read it as literally as the model does and look for the sentence that allows the new behaviour.

If the page itself is the problem, check that Task is filled in, since it is the one required field.

One more check: the Open in Claude button opens Claude in a new tab but does not carry the prompt across. Copy the prompt first and paste it into a new chat, so an earlier conversation does not change how Claude reads it.