What a model can really do with a contract

AI for lawyers and contracts · Lesson 1 / 20

The model does not know law — it knows text

A language model does not hold current regulation in its head and does not understand your deal. What it does well is handle a document as material: break it into parts, compare those parts against each other, and hunt for gaps. That is not a small thing. A large share of contract work is mechanical: checking that cross-references point where they should, that a defined term appears before its first use, that the payment deadline in one clause does not contradict another. Here the machine saves hours.

Three genuine strengths

Structuring. Turning forty pages of solid prose into a table of scope, timing, money, liability and exit takes the model a minute, with very little loss.
Comparison. Two versions, a counterparty draft against your template, an annex against the main body — the model finds discrepancies that the eye slides past in the third hour of proofreading.
Gap hunting. Give it a list of provisions a document of this type should contain and it will say honestly which are missing. A bad clause is visible; a missing one is silent.

How to frame the task

A request to just review the contract produces a polite retelling. A working request carries a role, a goal, an output format and explicit limits.

You are a contracts lawyer acting for the customer side.
Analyse the text below and return a table:
clause | what it does | risk to us (low/medium/high) | why
Do not assess legality and do not cite any law.
Separately list the provisions that are NOT in the text.

The ban on citations is not a formality. It removes the single most dangerous class of answers and keeps the model inside the zone where it is strong: working with the text you handed it. Applicability to your situation and jurisdiction stays with a human. This course is not legal advice.

Insight. The value of the model grows with volume, not difficulty. On two paragraphs you are faster alone; on two hundred pages of annexes you are not.

Common mistake. Asking the model whether a contract is legal or illegal. It will produce that verdict confidently and without foundation, because it knows neither your jurisdiction nor the facts of the deal.

Pro tip. Keep a short file describing your company: which side you sit on, your standing interests, your red lines. Paste it at the top of every analysis and the quality of answers changes noticeably.

Cheat sheet

The model is strong at structure, comparison and gap hunting.
State the role, the goal and the output format.
Ban citations to law directly in the prompt.
Judgements about applicability always stay with a human.

1. What is the model's core strength on a contract?

2. Why ban citations to law inside the prompt?

3. What is hardest for a human to notice while proofreading?

🔒 Answer the question correctly to move on to the next lesson.