5 AI Prompts Every Lawyer Should Know in 2026
March 23, 2026 · CounselKit
Most lawyers using AI in 2026 are doing some version of the same thing: pasting text into ChatGPT or Claude, asking a question, and getting back something that's either too generic to use or too confident to trust. The tools aren't the problem. The prompts are.
These five prompts address the most common failure modes in legal AI use. Each one is tool-agnostic — they work with ChatGPT, Claude, Gemini, or any other AI tool you use. Copy them, fill in the brackets, and see the difference immediately.
1. The Role Assignment Prompt
The single highest-impact change you can make is telling the AI who it is before asking it to do anything. Without a role, the AI defaults to a general-purpose assistant tone — friendly, hedge-y, and shallow. With a role, it adopts the precision and analytical framework you need.
The Prompt
I need help with a legal task. For this conversation, I want you to adopt the following role:
You are a legal professional with 10+ years of experience in document review, legal analysis, and drafting. You understand contract language, statutory interpretation, and the nuance of legal argument. You are direct, precise, and avoid unnecessary jargon while remaining technically accurate.
When you respond, adopt this expertise level:
- Flag assumptions you're making (e.g., "I'm assuming this is governed by state law because...")
- Identify what you don't know or can't verify
- Distinguish between what the law clearly says vs. what is typical practice vs. what requires further research
- Use professional, measured language — no marketing speak, no false certainty
Do you understand this role assignment? Confirm, and then I'll give you the specific task.
Use this before every major task. The confirmation step isn't just politeness — it forces the AI to acknowledge the constraints before it starts generating output, which measurably improves adherence to those constraints.
2. The Structured Context Loader
The second most common mistake is jumping straight to the question without giving the AI enough context. Generic input produces generic output — every time. This prompt creates a structured briefing that prevents the AI from filling gaps with assumptions.
The Prompt
I'm going to give you context for a legal analysis task. Please confirm you've understood the context before I ask you to analyze or draft anything.
Here's the context:
THE SITUATION:
[DESCRIBE THE LEGAL SITUATION OR TRANSACTION — what's happening, who the parties are, what's at stake]
THE KEY CONSTRAINTS:
- Relevant jurisdiction(s): [list them]
- Critical dates or deadlines: [if any]
- Parties involved: [list them with their roles]
- What's already been decided or settled: [what's fixed vs. what's in play]
- What's NOT in scope: [what are we explicitly NOT analyzing?]
THE DELIVERABLE I NEED:
[DESCRIBE WHAT YOU NEED — e.g., "a memo analyzing X," "a redline of section 4," "a summary for the client"]
Before you proceed, please confirm:
1. You understand the situation and the key parties
2. You've noted the constraints and dates
3. You understand what IS and ISN'T in scope
4. You're ready for the task
If anything is unclear, ask me to clarify. Don't proceed until we're aligned.
Taking 30 seconds to load context this way prevents pages of off-target analysis. It's the difference between briefing a colleague and shouting a question across the office.
3. The Anti-Hallucination Constraint
AI hallucination in legal work isn't a theoretical risk — hundreds of attorneys have been sanctioned for submitting AI output containing fabricated citations and nonexistent legal standards. This prompt is a guardrail you can add to any task.
The Prompt
Before I ask my question, I need to set clear constraints on what you should and shouldn't do.
DO:
- Analyze only what's in the provided document or context
- Flag any assumptions you're making
- Use only legal principles I've provided or that you can clearly identify
DON'T:
- Don't cite any case, statute, or regulation unless I have provided it to you
- Don't add background information about this industry or area of law from your training data
- Don't present uncertain information as fact
Red flags for me:
If you find yourself about to cite a specific case or statute that I haven't provided, stop and note it rather than proceeding.
Now, with these constraints in mind, here's my task: [YOUR ACTUAL TASK]
This prompt works because it explicitly names the failure mode and tells the AI what to do instead. Rather than silently hallucinating a citation, the AI flags that it would need to cite something and asks you to provide it. That's the behavior you want.
4. The Risk Tiering Prompt
This one is immediately useful for anyone doing contract review or document analysis. Instead of getting a flat list of issues, you get a prioritized framework that mirrors how you'd actually communicate findings to a client or partner.
The Prompt
I've uploaded [DOCUMENT TYPE — e.g., "a vendor services agreement" or "an NDA draft"]. Review it and categorize every notable issue into three tiers:
HIGH RISK — Issues that could result in significant liability, loss of rights, or material financial exposure. These need immediate attention and likely require negotiation.
MEDIUM RISK — Issues that deviate from standard practice or create moderate exposure. Worth flagging and discussing, but not deal-breakers.
LOW RISK — Minor issues, stylistic concerns, or provisions that are slightly unusual but within acceptable range.
For each issue:
1. Quote the specific language from the document
2. Explain why you've assigned that tier
3. Suggest a specific action (negotiate, accept, flag for discussion, request clarification)
After the tiered analysis, provide:
- A count of issues per tier
- Your overall assessment in 2-3 sentences
- Any provisions you expected to see but that are MISSING from the document
The requirement to quote specific language and explain the tier assignment prevents the AI from making vague, unsupported claims about risk. If it can't quote language to justify a "high risk" rating, it has to reconsider.
5. The Output Format Enforcer
This isn't a standalone prompt — it's a module you attach to any other prompt to get output you can actually use without reformatting. The AI defaults to wall-of-text prose. You need structure.
The Prompt
I need you to structure your output in a specific format so I can use it directly in my work. Here's how:
For this task, use this exact structure:
[SECTION HEADING]
- [Key point or finding]
- [Key point or finding]
[NEXT SECTION HEADING]
[Summary paragraph if needed]
Format requirements:
- Use bullet points, not dense paragraphs
- Use bold headers to clearly separate sections
- Use a table if comparing multiple items
- Use numbered lists ONLY if order/sequence matters
- Be concise: every line should earn its place
- Include [CITATION NEEDED] if you're flagging something you're uncertain about
- Don't use markdown formatting beyond bold, italics, bullet points, and headers
Specifying format before the task — not after — produces dramatically better results. Once you have a format template that works for your practice, you'll reuse it across dozens of different tasks.
The pattern behind all five
Each of these prompts follows the same underlying structure: tell the AI who it is, give it context, constrain what it can and can't do, and specify the output format. This isn't complicated or technical. It's just being explicit about what you need instead of hoping the AI guesses correctly.
The lawyers getting the most value from AI in 2026 aren't using different tools than everyone else. They're using the same tools — ChatGPT, Claude, Gemini — with more structured prompts. The prompts are the differentiator.
These five prompts are adapted from CounselKit, a pack of 24 structured AI prompts built specifically for legal workflows — document review, research, analysis, and drafting. Each prompt includes anti-hallucination guardrails and works with any AI tool.
Get all 24 prompts — $49