Written by: Laura Jeffords Greenberg - Head of AI Legal Academy

This article is based on an episode of The Only Constant podcast. Listen on Spotify or Apple Podcasts.

Lawyers are trained communicators. They write for judges. They negotiate for a living. They draft with precision, intent, and structure. And yet, when many lawyers sit down in front of AI, something strange happens.

They stop communicating clearly.

The "AI Amnesia" Problem

One of the most common patterns in AI training with lawyers is what I’ve started calling AI amnesia. Highly capable professionals suddenly:

• Ask vague questions

• Strip away critical context

• Assume the AI “knows” what they mean

• Accept generic answers they would never accept from a colleague

The output looks polished but feels off. The conclusion is predictable: “This tool isn’t very good.” In reality, the issue isn’t the model. It’s the missing context.

AI Can’t Read Your Mind

Many lawyers subconsciously assume AI understands more than it does. They assume it knows:

• Their role

• Their company

• Their jurisdiction

• Their risk tolerance

• What “acceptable” looks like

It doesn’t. AI has no grounding unless you give it one. It does not know your internal policies, your deal history, or which risks are business-critical. All of that context lives in your head.

When lawyers ask AI a high-level question without transferring that context, the model fills the gaps with general patterns. The output sounds fluent but feels wrong because it lacks the specifics that actually matter. This is why prompting often feels disappointing. The issue is not that the AI is shallow. It is that the legal context never left the lawyer’s brain.

The Fix: Stop Asking Questions, Start Giving Instructions

The most useful mental shift for lawyers is this: Stop asking AI questions. Start giving it instructions.

Imagine you are delegating a task to a junior colleague who understands the law in theory but doesn’t know your company, the deal history, or which risks you care about. You would never say, “Review this contract and tell me if it’s okay.”

You would explain:

• What the document is

• Why you are reviewing it

• Your role in the deal

• The company’s position

• What you want back

Prompting works the same way. Good prompting often looks like a clear instruction, followed by a context dump, followed by constraints. If you’ve ever worked across time zones, you know this instinctively. You provide more detail up front because you won’t be there to clarify later. AI is always asynchronous.

Before & After: How Context Changes Everything

The difference between a disappointing AI response and a useful one is rarely the model. It’s almost always the prompt.

Before: The Generic Prompt

“Review this contract and tell me if there are any risks.”

This prompt leaves almost everything unsaid. The AI will respond with a broad list of standard risks. It will sound reasonable. It will also be largely unhelpful.

After: The Same Task, With Legal Context

“You are an in-house commercial lawyer reviewing a SaaS agreement governed by English law.

We are the customer. This is a mid-market deal with a vendor we want to move forward with.

Our priorities are data protection, liability caps, termination rights, and audit rights.

Please review the agreement and:

• flag provisions that are outside market for a customer

• identify clauses that could create material risk post-signing

• distinguish between ‘must-fix’ and ‘nice-to-have’ issues

Do not summarise the entire contract. Focus only on the points that require legal or commercial judgment.”

This prompt does several important things:

• Assigns a role

• Provides jurisdiction

• Clarifies deal context

• Defines risk tolerance

• Constrains the output

The AI now has enough information to be selective. The output becomes shorter, sharper, and more aligned with how a lawyer actually thinks.

Prompting Is Now a Core Legal Skill

Prompting is not a technical add-on to legal work. It is legal work. It requires clarity of thought, prioritization of risk, and the ability to articulate intent. Lawyers already have these skills. They just need to apply them deliberately when working with AI.

The question is not, “Why isn’t the AI giving me better answers?” The better question is, “What assumptions am I making about what this system already knows?”

When lawyers adjust that mental model, prompting becomes less frustrating and far more effective.

Frequently Asked Questions (FAQ)

Q1: Why do lawyers struggle with AI prompting more than other professionals?
Lawyers are trained to be precise and to avoid ambiguity. Prompting can feel iterative and imprecise, which is uncomfortable. They also often assume the AI has more legal context than it does, leading them to ask overly broad questions.

Q2: What is the single most important thing to include in a legal prompt?
Context. Specifically, your role (e.g., "I am the customer's in-house counsel"), the governing law/jurisdiction, and your primary goal for the review (e.g., "flag deal-breakers," "ensure compliance with our policy").

Q3: How can I get better at prompting?
Practice treating the AI like a new junior associate. Before you ask a question, write down the key context you would give a human colleague. This simple mental shift from "asking" to "instructing" dramatically improves output quality.