Review a prompt bound for a reasoning model

In a nutshell: Check a prompt against the rules for reasoning models, where chat-model habits work against you

Use it when: You are moving a working prompt onto a reasoning model and want to know what to take out.

MIT

Compiled Command

Check a prompt against the rules for reasoning models, where chat-model habits work against you. Applies when: You are moving a working prompt onto a reasoning model and want to know what to take out. ## Write an instruction a model follows the first time How to answer, from the prompt-engineering/write-an-instruction-a-model-follows Framework. These govern the way you write your answer. They are not requirements to check the document against. - **Put the instruction before the material** — Others recommend that you place instructions at the beginning of the prompt. - **Say what to do rat

Characters: 2,108
Field: Prompt engineering
Authors: DAIR.AI

What it draws on

The instructions should be simple and direct.

Add response constraints in the instruction whenever applicable.

Avoid ambiguous and unnecessary instructions.

This means that you want to be more careful with how you use CoT, and potentially avoid using it with reasoning models altogether.

Start with low thinking effort and evaluate the quality of the response.

We recommend using XML as the default mode for structuring generated content unless there is a hard requirement to output the content in JSON.

Sources

Prompt Engineering Guide — Prompting Reasoning Models
DAIR.AI — Licensed MIT
Source last checked 2026-09-21
Prompt Engineering Guide — General Tips for Designing Prompts
DAIR.AI — Licensed MIT
Source last checked 2026-09-21

What it draws on

Keep the instruction simple and directFrom Prompt Engineering Guide — Prompting Reasoning Models

State the response constraints in the instructionFrom Prompt Engineering Guide — Prompting Reasoning Models

Cut ambiguous and unnecessary instructionsFrom Prompt Engineering Guide — Prompting Reasoning Models

Be careful with chain-of-thought hereFrom Prompt Engineering Guide — Prompting Reasoning Models

Start at low thinking effort and judge the answerFrom Prompt Engineering Guide — Prompting Reasoning Models

Structure generated content as XML unless JSON is requiredFrom Prompt Engineering Guide — Prompting Reasoning Models

Sources