Home › Prompt Builder
AI Prompt Builder
Fill in as much as you need and watch the prompt assemble itself on the right. You do not have to complete every field. The score tells you which parts you have skipped and whether that is likely to cost you anything.
Build your prompt
Your prompt
What makes a prompt score well
The score is not a grade on your writing. It measures how much ambiguity you have removed. Task is worth the most because a prompt without one is just a topic. Context is worth almost as much, because it is the field people skip most often and the one that changes output the most. Role, format, tone, and constraints each close a smaller gap.
Anthropic's own prompt engineering guidance puts being clear and direct first, ahead of any clever technique, and recommends giving the model the context a new colleague would need to do the task. That is the principle this builder is shaped around.
If you want the research view, Sahoo and colleagues catalogued 41 distinct prompting techniques in a systematic survey published on arXiv in February 2024, grouped by application. Most of them are refinements on the same foundation: tell the model precisely what you want and what good looks like.
Sources: Anthropic, prompt engineering overview in the Claude developer documentation (docs.anthropic.com); Sahoo, P. et al., "A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications," arXiv:2402.07927, February 2024 (arxiv.org). Both linked above.
How to Use the Prompt Builder
The builder assembles a prompt from four fields that together account for most of the difference between a useful response and a vague one. Role sets who the model should act as, which shapes vocabulary and level of detail. Task states the single thing you want produced, phrased as an instruction rather than a topic. Context supplies the facts the model cannot infer, including audience, constraints and any material it should work from. Format specifies the shape of the output, meaning length, structure and whether you want prose, a list or a table.
What Each Field Does in Practice
Role is the field people skip and should not. Act as a hiring manager reviewing applications produces different output from act as a career coach, even with an identical task. Task should contain one verb: draft, summarise, rewrite, compare. Two tasks in one prompt reliably produces a worse version of both. Context is where specificity pays most, since audience and constraint do more to shape an answer than any instruction about tone. Format is what makes output usable without editing.
Weak Prompts Versus Strong Ones
A weak prompt reads: write something about our new product. It has no role, an unclear task, no context and no format, so the model guesses on all four. A strong version of the same request reads: act as a B2B copywriter, draft a 150 word product announcement for existing customers of a small accounting firm, focus on the time saved rather than the feature list, and return it as three short paragraphs with no headings.
The second prompt is longer, and length is not the point. It is specific on four dimensions the first left open, which is why it produces something you can use rather than something you have to rewrite.