Precision is how specifically a prompt states its requirements and how narrowly it asks a language model to interpret them. Higher precision reduces the range of plausible outputs and usually improves consistency; lower precision leaves room for variation. Choose the level according to the job: exploration benefits from latitude, while work that must fit an existing system or pass review needs tighter specifications. section 3.1 section 3.1.2
To make a prompt more precise, specify the requirements that determine whether its output is usable:
- Scope: name the task, audience, system, and boundaries rather than asking for a generic artifact. section 3.1.1
- Required content: list facts, components, steps, or failure branches that must appear. section 3.1.1 section 3.1.2
- Output requirements: state the format, section order, length limits, and tone when they affect use or review. section 3.1.1 section 5.2.1
Specify a password-reset diagram for review
“Create a sequence diagram for a password reset” permits a plausible sketch but leaves system boundaries and failure handling unspecified. A more precise prompt names the User, Web App, Auth API, Email Service, Token Store, and Audit Log; requires reset request, token generation, signed-link email, validation, password update, token invalidation, and audit write; and asks for labeled success and invalid-token branches. Review the result against that list: a diagram that omits token invalidation or the invalid-token branch is not yet ready for the intended review. section 3.1.2
Table 3.1 When to use each precision level
| Choose low precision when |
|
|---|---|
| Outputs must fit existing systems, formats, or workflows. |
You are exploring options or searching for new directions. |
| Multiple people need consistent results. |
Variation is useful and human review is available. |
| Clear success criteria already exist. |
Success criteria are still emerging. |
| Revision cycles are costly or time constrained. |
Iteration is cheap and part of the process. |
| Accuracy and implementation readiness matter most. |
Novelty and idea generation matter most. |
Use the precision-level comparison to decide whether consistency and implementation readiness outweigh variation and idea generation. section 3.1.2
Precision is not the same as directness or brevity. It controls the specificity of requirements; directness controls whether instructions are explicit, and brevity controls wording length. More detail costs tokens and can cause filler or failure when requirements conflict. Start with moderate precision while exploring options, then tighten the prompt around a selected direction. Check the output against each stated requirement rather than assuming a detailed prompt guarantees compliance. section 3.0 section 3.1 section 3.1.2
