Chain of Thought: How AI Reasons Through Legal Problems

When an AI tells you "This clause violates state law," that's a claim. But how did it reach that conclusion? Chain-of-thought is asking the AI to show its work—to explain each reasoning step so you can verify if the logic holds. In legal contexts, this is critical because flawed reasoning can lead you to make bad decisions.

Here's how it works: Instead of asking "Will I have a claim for breach of contract?" you ask "Walk me through the elements of a breach of contract claim and explain which ones this situation satisfies." The AI then lists the elements, evaluates each one against your facts, and shows where the claim is strong or weak. Now you can judge whether the reasoning is sound.

The power of chain-of-thought is that it reveals when AI is making assumptions or skipping steps. If the AI says "This is a breach because they didn't deliver on time," chain-of-thought would force it to show: "Does the contract specify a delivery date? Did they miss it? Did that delay cause you harm?" If any of those answers is "no" or "unclear," the claim weakens. You see this instead of just getting a conclusion.

In practice, chain-of-thought is especially useful for analyzing citations and legal arguments. If you ask an AI to "cite the law supporting your answer," it might hallucinate a statute. But if you ask it to "explain which specific statute or regulation applies, what it says, and how it applies to these facts," you can check each step. Most hallucinations fail when you demand a paper trail.

Another application: evaluating AI-drafted arguments. Say you've asked AI to draft an argument for why a landlord owes you a refund. Use chain-of-thought to have the AI explain: "What are the landlord's likely counterarguments? How would you address each one?" This prepares you for negotiation or dispute resolution because you understand not just your position but also its weaknesses.

The technique also works in reverse. If someone else makes a legal claim to you—a collector demanding payment, a business making a threat—ask an AI to "evaluate this argument step by step. What assumptions is it making? What would need to be true for this to be valid?" This helps you assess whether you should take the claim seriously.

A practical example: A contractor claims you owe them $10,000 for work they started but didn't finish. You could ask an AI: "Under contract law, does a contractor have a right to payment if they abandon a job? What conditions would trigger that right? Given these facts [describe], does the contractor meet those conditions?" The step-by-step reasoning shows you whether their claim is legitimate.

The limitation: Chain-of-thought makes hallucination less likely but doesn't eliminate it. An AI can reason perfectly from false premises. So if an AI cites a statute, verify that the statute actually exists and says what the AI claims.

Try this: Find a legal question you're unsure about and ask two versions: (1) a direct question ("Am I liable for...?") and (2) a chain-of-thought version ("Walk me through the steps of liability. What would I need to do to be liable? Do my facts meet those conditions?"). Compare the answers. The chain-of-thought version will reveal reasoning you can judge, while the direct answer might just be a guess.