Safe Prompt Habits

Why prompts matter

The quality of AI output depends heavily on the quality of the input. A vague prompt produces vague output. An ambiguous prompt produces output that may address a different question than the one you had in mind. A prompt that omits critical context produces output that fills that gap with assumptions.

Good prompt habits reduce errors, make output easier to review, and produce results that are closer to usable on the first pass. This lesson gives you a structure you can use for any AI task in a professional setting.

The five-part structure

A well-structured prompt for professional work includes five elements. Not every prompt needs all five — but when output quality matters, all five should be present.

1. Role

Tell the AI what role to take. This shapes the tone, vocabulary, and level of detail in the response.

• "You are a professional editor reviewing a client communication for clarity."
• "You are an experienced legal intake coordinator."
• "You are a business analyst summarizing a set of intake notes."

Without a role, the model defaults to a generic helpful assistant. With a role, the output is more likely to match the register and approach you need.

2. Context

Provide the relevant background the model needs to do the task correctly. This is where most prompts fall short. The model cannot assume details about your firm, your client, or the specific situation — you have to provide them.

• What type of firm or practice is this for?
• What is the document or situation being worked on?
• Are there constraints the model should know about? (jurisdiction, client type, document stage)

Apply the privacy rules from Lesson 4: provide the context the model needs, redact what is not necessary.

3. Task

Be specific about what you want the model to do. Vague tasks produce vague output.

• Weak: "Help me with this email."
• Strong: "Draft a follow-up email to a prospective client confirming their intake appointment and listing three documents they should bring."

The more precisely you describe the task, the less room the model has to interpret it in a direction you did not intend.

4. Rules

Set constraints on what the model should and should not do. This is where you establish guardrails specific to the task.

• "Do not include legal advice or opinions."
• "Do not mention fees or billing."
• "Use plain English. Avoid jargon."
• "Do not include information I have not provided — flag gaps instead."
• "Keep the response under 200 words."

Rules are especially useful for preventing boundary crossing — the failure mode from Lesson 2 where the model does more than you asked.

5. Output format

Specify how you want the output structured. This makes the output easier to use and easier to review.

• "Return a bulleted list."
• "Write this as a professional email with a subject line."
• "Return a two-column table: issue in the left column, recommended action in the right."
• "Write three short paragraphs, one for each topic."

The full prompt template

A worked example

Using prompts from your firm's library

GuideHerd Academy includes a prompt and template library as part of your subscription. These prompts have been designed for common professional services workflows and include the guardrails built in. Using a library prompt is faster than building from scratch and reduces the chance of missing an important constraint.

When you adapt a library prompt for a specific task, keep the rules and output format sections intact. Those are where the safety guardrails live.