Skip to main content

Prompt guides

How prompting differs for code, writing, and analysis

How the right prompting approach shifts between coding tasks, writing tasks, and analytical tasks — and how that connects to choosing the right model for each, not just the right words.

All prompt guides · Last reviewed:

Prompting for code: precision over persuasion

A coding prompt benefits most from precision — the exact function signature, the language and version, the constraint the code has to satisfy, an example of the expected input and output. Vague framing that would be harmless in a writing prompt ("make it good") gives a coding task almost nothing to work with, since there is no single correct shape to a request like that.

Prompting for writing: audience, tone, and one good example

A writing or editing prompt benefits most from the context and constraint guidance in how to write a clear, specific prompt, linked below — who the piece is for, the tone it should hold, and a length or structure constraint. An example of the target voice, per few-shot prompting, linked below, often does more work here than a longer description of the tone would.

Prompting for analysis: asking for the reasoning, not just the conclusion

An analytical task — weighing options, interpreting data, working through a decision with several factors — usually benefits from chain-of-thought prompting, linked below: asking the model to lay out its reasoning rather than only stating a conclusion gives you something to check, and tends to surface a missed factor or a weak assumption. This is the same shape of task as choosing a model for complex reasoning, linked below, which covers which model is built for exactly this rather than repeating that guidance here.

Questions people ask

Does one prompting technique work best across all three task types?
No — precision matters most for code, audience and tone matter most for writing, and asking for visible reasoning matters most for analysis. Most tasks benefit from a mix, weighted toward whichever of these the task actually needs.
Does the model I choose matter as much as how I write the prompt?
Both matter, and they are different levers — this guide is about wording; see choosing a model for your task, linked below, for the model-fit side of coding, writing, and reasoning-heavy tasks specifically.
Is a coding prompt just a writing prompt with different words?
No — a coding task usually has a single correct or working shape, so precision about the exact requirement matters more than it does for most writing, where several different phrasings can all be good.

Practice on a real conversation

ClawAI’s routing modes already lean toward a fitting model per task — Auto and High Reasoning for analytical work, for example — so a well-written prompt and a fitting route work together rather than as separate choices.