The Problem
Describing exactly what you want is hard. You write instructions, get back something close but not quite right, then refine them and try again. For tasks with specific formats or styles, this back-and-forth is frustrating.
The Solution
Show the AI a few examples of what you want, and ask it to follow them.
This technique is called "few-shot learning" - you're teaching the AI through examples rather than rules.
Structure
I need you to [task description].
Here are examples of the format I want:
Example 1:
[Your example]
Example 2:
[Your example]
Example 3 (optional):
[Your example]
Now do the same for: [your actual input]
Before You Rely on This
The examples on this page are made up to show the technique. No test results have been published for it, so try it on your own work and judge it by what you find.
This guide is about how to word a prompt. It does not make a client's document or information safe to put into an AI tool. What you may enter depends on the plan your firm has, its agreement with the vendor and your firm's policy: Claude Projects for Client Matters sets out the checks for one tool. To practise, use a precedent or a published document that contains no client information, and refer to parties by role.
Why It Works
Anthropic's guidance for Claude says: "Examples are one of the most reliable ways to steer Claude's output format, tone, and structure." It calls the technique few-shot or multishot prompting (checked 29 September 2026; the page is listed under "Source"). That is one vendor writing about its own models. Other vendors publish their own guidance.
Example Use Cases
1. Contract Clause Drafting
Prompt:
Draft a limitation of liability clause for our SaaS agreement.
Here are examples of the style and structure I want:
Example 1:
"Neither party shall be liable to the other for any indirect or
consequential loss arising out of or in connection with this Agreement."
Example 2:
"Each party's total aggregate liability under this Agreement shall not
exceed the fees paid by the Customer in the twelve (12) months before
the claim arose."
Now draft a clause limiting liability for data processing errors,
following the same style and structure.
2. Time Entry Descriptions
Prompt:
Convert my work notes into time entries for billing.
Examples of the format I want:
Example 1:
Input: "Call with client contact re contract"
Output: "Telephone attendance on client regarding commercial contract terms; advice on liability cap and termination provisions (0.3 hours)"
Example 2:
Input: "Reviewed NDA and made changes"
Output: "Review and revision of non-disclosure agreement; amendments to confidentiality obligations and term provisions (1.2 hours)"
Now convert these:
- Meeting with client about employment issue
- Research on redundancy law
- Drafted witness statement
3. Client Communication Tone
Prompt:
Rewrite this email to a client in the tone/style shown in these examples:
Example 1:
"Thank you for your email. I've reviewed the contract and have
identified three areas requiring amendment. I'll prepare a marked-up
version by close of business tomorrow."
Example 2:
"I understand your concerns regarding the lease terms. The landlord's
position is commercially aggressive, but there are several points we
can push back on. I suggest we discuss strategy tomorrow morning."
Now rewrite this:
"The other side rejected our offer. We need to change our approach."
Pro Tips
- Start with a few examples and add one when the output does not match. Anthropic's guidance for Claude is to "Include 3–5 examples for best results" (checked 29 September 2026)
- Examples should be diverse (show range of scenarios, not identical cases)
- Include edge cases if your task has them
- Match the complexity - don't show simple examples for complex tasks
- Label examples clearly if you're showing input → output transformations
Common Mistakes
❌ Inconsistent examples: Showing contradictory formats confuses the AI
❌ Bad examples: Showing incorrect examples will get you incorrect outputs
✅ Clear, consistent, relevant examples that represent your desired output
Quick Reference
Template:
I need you to [task].
Here are examples of what I want:
Example 1:
[Perfect example of desired output]
Example 2:
[Another perfect example, slightly different scenario]
Example 3 (optional):
[Third example if needed for clarity]
Now apply the same approach to: [your actual input]
Use it for:
- Drafting with specific house style
- Formatting outputs (tables, bullet points, etc.)
- Matching tone and register
- Standardising variable content (time entries, case summaries)
- Showing AI your firm's conventions
Source
Checked on 29 September 2026.
- Anthropic, Prompting best practices: the section "Use examples effectively"