Overview

To get the most detailed and tailored output from Harvey, it’s essential to include the right mix of “request,” “context,” and “output” in your prompt. This article provides tips and best practices to help you use Harvey effectively.

How Prompts Work

Harvey performs best when given clear, task-specific instructions in natural language. Think of Harvey as a junior colleague who needs precise directions and relevant context.

A successful prompt includes three key components:

Request: Use clear, specific language to phrase your prompt as a question or request.
- Example: Summarize the interim covenants…
- Example: Analyze the assignment clause…
- Example: Draft a memo…
Context: Provide background to help Harvey understand your request.
- What document have you uploaded?
- Who is your intended audience?
- What key issues should be addressed?
Output: Specify the desired format and style of the response.
- Example: …in a table
- Example: …as a bulleted or numbered list
- Example: …as a memo

How to Craft Your Prompt

Upload Supporting Documents: Uploading documents helps ground Harvey and reduces the risk of hallucinations (AI-generated incorrect or misleading information). Be aware that Harvey may not always provide citations, especially when:
- Multiple sources are referenced
- There are too many valid sources
- The information is inferred as a general concept from the document
Avoid Referencing Pages or Document Numbers: Harvey may not interpret documents and pages as you do. Instead, refer to sections or document names.
Include Sufficient Information in Your Prompt: Be specific and detailed, specifying the desired output format (e.g., structure the response as a memo with subheadings).
Consider Breaking Up Complex Prompts: Simpler queries often yield better responses. When dealing with large datasets or complex questions, break your prompt into multiple parts to allow Harvey to focus on one task at a time.
- Assist mode: Use follow-up questions instead of multi-part prompts.
- Draft mode: Break up complex prompts into multiple queries, or reivisons, to ensure thorough responses. Think of it as marking up the document with a pen. Revisions should be tethered to the first draft.
Be Mindful of Output Length: Harvey cannot generate a specified number of pages. Instead of asking for a “10-page summary,” guide Harvey by requesting a “thorough” or “detailed” output for longer responses, and use “brief” or “concise” for shorter ones.

Examples

Draft Mode: Summarization

Less Effective: Can you make me a detailed summary of the ruling?
• This prompt lacks detail on what the format of the summary should look like → Harvey guesses

Better: Summarize this ruling in a clear and well-organized memorandum, with subheadings, and discuss potential implications for other companies.
• This prompt lays out the format for the summary and indicates exactly how Harvey should present the summary.

Best: I am an antitrust lawyer advising my client, Microsoft, on its arrangement to bring ChatGPT as the exclusive AI provider to Siri searches on Apple's iPhone. Review this court ruling and thoroughly summarize the anticompetitive effects of Google's agreement with Apple and explain how I could argue against those for Microsoft's agreement with Apple. Also, summarize the pro-competitive justifications raised by Google and why the court rejected those, and explain how I could strengthen those arguments for Microsoft. The output should be in paragraph format with clear subheadings for each effect and justification.
• This is a great prompt when you have some baseline familiarity with the document. Harvey will focus its summary and provide more in-depth, targeted analysis.

Draft Mode: Using Revisions with two types of revisions: Targeted vs. Global

Global Example 1:
Prompt: Draft an email to a counterparty asking when we should expect comments to the merger agreement.

Revision: Make this more concise and passive aggressive.

Global Example 2:
Prompt: Draft a law firm client alert summarizing this court ruling. Ensure the client alert is broken into different subheadings. End the client alert with a bulleted list of key implications and takeaways.

Revision: This should have a more academic tone.

Targeted:
Prompt: Draft a non-solicit provision for an NDA restricting buyers from soliciting the target company's employees.

Highlight: “For purposes of this provision, "employee" means any person who is or was employed by Target or any of its affiliates at any time during the six (6) months prior to the date of this Agreement or at any time thereafter while this provision is in effect. This provision shall not prohibit Buyer from hiring any employee of Target or any of its affiliates who (a) initiates contact with Buyer without any solicitation or encouragement by Buyer or any of its representatives, (b) responds to a general advertisement or solicitation that is not specifically targeted at employees of Target or any of its affiliates, or (c) is no longer employed by Target or any of its affiliates at the time of hiring.”

Revision: Remove the defined term for employee and add a carve-out (d) for employees terminated in the last 6 months.

Draft Mode: Precedent-Based Drafting

Less Effective: Draft a customary definition of "Indebtedness" for a Merger Agreement governed by Delaware law in which both parties are public companies. [No document upload]
• When should this be avoided? This prompt is great when drafting from scratch, but grounding Harvey in a source of truth will ensure the definition is tailored to your use case, precedent and what is market to your practice area.

Better: Draft a definition of "Indebtedness" using these precedents. Create new paragraphs for each subsection and, for each subsection, explain in brackets the purpose for including the subsection and which precedent it is from.
• Why is this better? Providing Harvey with precedent grounds Harvey in a source of truth, but Harvey is still constructing the clause from scratch.

Best: Suggest ways to revise the below definition of "Indebtedness" to align more closely with these precedent merger agreements and explain the changes made.
• When is a prompt like this helpful? The precedent plus the supplied language establishes confines within which Harvey should provide feedback. Also, it’s always a great idea to ask Harvey to explain its work!

Assist Mode: Dividing Queries with Follow-ups

Less Effective: In part one, please generate a detailed chronological summary of events from these documents.

In part two, please identify any conflicts, gaps, contradictions or ambiguities in these documents.

In part three, generate discovery requests to resolve the issues identified in part two.

Better: Initial Prompt: In a table, generate a detailed chronology of the main events in these documents. In the first column, specify the date of the event. In the second column, name the event. In the third column, provide a description of the event.

Follow-up: Identify any conflicts, gaps, contradictions or ambiguities in these documents.

Follow-up: Generate discovery requests to resolve the issues identified above.

Draft Mode: Draft Officer’s Certificate

Less Effective: Draft Purchaser's closing officer's certificate.
• Harvey’s compute power is first focused on analyzing the document to understand who the Purchaser is and then focused on what an officer’s certificate is and what it should include. In many instances, Harvey may include non-essential information.

Better: Draft Purchaser's closing officer's certificate pursuant to Section 8.3(d) of the agreement.
• Harvey understands exactly where to look when drafting the officer’s certificate, so its full compute power is dedicated to the task at hand.

General Prompt Example

Less Effective: Check this email for spelling and grammar errors.
• Generative AI, while likely able to identify your spelling and grammar mistakes, is much more than a word processor that does routine, repeatable tasks like a spell check.

Better: Is there anything in this email that someone might find confusing or unclear? List ten questions that a recipient might have after reading this email.
• Think of Harvey like a second brain living in your computer that you can instantly bounce ideas off of and generate content you might not have otherwise thought of.

For further assistance, please contact [email protected]