How to write effective prompts for Claira to get accurate, consistent results from AI-powered document review.

Prompting Overview

Writing effective prompts is the key to getting accurate, consistent results from Claira. A prompt is your set of instructions to the AI -- it tells the model exactly what to look for, how to analyze the document, and what format to return the result in.

Why prompts matter

Think of writing a prompt as giving instructions to a meticulous but literal-minded colleague. If you say "tell me about this document," you will get a vague, general response. If you say "determine whether this document discusses the patent licensing agreement between Acme Corp and Beta Inc, and respond with Relevant or Not Relevant," you will get a precise, usable answer.

The quality of your results depends almost entirely on the quality of your prompt. Spending time on prompt writing is not overhead -- it is the most impactful investment you can make in your review workflow.

Core principles

Be clear and specific

State exactly what you want the AI to do. Vague instructions produce vague results.

◈Too vague

Is this document relevant?

◈Clear and specific

Determine whether this document is relevant to the breach of contract claim between Acme Corp and Beta Inc regarding the 2024 supply agreement. Respond with "Relevant" or "Not Relevant".

Define key terms

Do not assume the AI knows your case-specific terminology. If a term has a particular meaning in your review, define it.

◈With defined terms

A document is "Relevant" if it discusses, references, or relates to the supply agreement dated March 15, 2024 between Acme Corp and Beta Inc, including any amendments, negotiations, or disputes arising from that agreement. A document is "Not Relevant" if it does not mention or relate to this agreement.

Ask for structured output

Tell the AI exactly how to format its response. This makes results consistent across thousands of documents and easier to use downstream.

◈Structured output format

Respond in exactly this format:
Determination: [Relevant / Not Relevant]
Confidence: [High / Medium / Low]
Reason: [One sentence explaining your determination]

Keep prompts focused

Each prompt should have a single, clear objective. If you need multiple types of analysis, consider running separate prompts rather than cramming everything into one.

Limit the AI's options

When you want a classification, list the exact allowed values. When you want a summary, specify the length. Constraints make output predictable.

◈Constrained response

Classify this document into exactly one of the following categories: Contract, Invoice, Correspondence, Internal Memo, Other. Respond with only the category name and nothing else.

Tell the AI what to do when information is missing

Documents do not always contain the information you are looking for. Your prompt should tell the AI how to handle that.

◈Handling missing information

If the document does not contain enough information to make a determination, respond with "Insufficient Information" and briefly state what is missing.

Prompt examples by workflow

Relevance review

◈Relevance review prompt

Review this document and determine whether it is relevant to the following issue: [describe the issue in detail].
Respond with:
Determination: [Relevant / Not Relevant / Potentially Relevant]
Reason: [One sentence explaining your determination]

Privilege review

◈Privilege review prompt

Analyze this document for attorney-client privilege. A document is privileged if it is a confidential communication between a client and their attorney (or attorney's agent) made for the purpose of seeking or providing legal advice.
Respond with:
Privilege: [Privileged / Not Privileged / Needs Review]
Basis: [Brief explanation of why the document is or is not privileged]

Objective coding

◈Objective coding prompt

Classify this document into one of the following document types: Email, Contract, Invoice, Memorandum, Presentation, Spreadsheet, Other.
Respond with only the document type.

Entity extraction

◈Entity extraction prompt

Extract the following information from this document:
All person names mentioned
All company or organization names mentioned
All dates mentioned
Format your response as:
People: [comma-separated list, or "None found"]
Organizations: [comma-separated list, or "None found"]
Dates: [comma-separated list, or "None found"]

Key fact finding

◈Key fact finding prompt

Identify the three most important facts in this document that relate to [describe the case issue]. For each fact, provide the fact and a direct quote from the document that supports it.
Format your response as a numbered list.

Quick checklist

Before you run a prompt at scale, confirm:

Did you state clearly what you want the AI to do?
Did you define any case-specific terms?
Did you specify the exact output format?
Did you set limits on the response (allowed values, length)?
Did you include instructions for when information is missing?
Are you using the same wording for the same task across documents and reviewers?

Next steps

Ready to write your own? See Custom Prompts for practical guidance.
Need help tuning model behavior? See Model Settings.
Need better scan quality? See Understanding Results.

Need help? Contact us at support@claira.to.

Prompting Overview

On this page