Most AI legal outputs aren’t wrong, they’re just not helpful. The AI isn't broken, but the prompts you're using to coax contract assistance out of a chatbot probably are.
Here's how to write AI prompts to generate responses that actually move a deal forward.
Every legal analysis is intended to help make a legal decision: do we accept this clause, fight this position, or sign this deal? If your prompt doesn’t imply a decision, the AI's answer won’t either.
Weak: “Summarize this contract.”
Effective: “We’re the customer. What are the biggest risks in this SaaS agreement that could cost us money or slow down operations?”
When you write a prompt, make sure three things are obvious:
AI defaults to neutral positions, which isn’t how legal work gets done. Lawyers represent a position and you need to tell the AI to do the same.
You don’t need a paragraph-length prompt preamble. One line is usually enough: “Act as commercial counsel for a mid-market SaaS company.”
That tells the model what angle to take when ingesting a body of text. Leave it out and you'll get polite regurgitation of contract language, not actionable assistance.
Broad, non-specific prompts like “Review this contract” are how you get a wall of useless text in return.
Instead, tell the AI what you care about: “Focus only on limitation of liability, indemnity, and termination. Ignore everything else.”
You’re not dumbing the task down, you’re making it tractable.
A useful trick is to paste clauses separately and label them, since the model handles smaller chunks better than a 40-page upload. AI has all the legal instincts of a junior associate (none). Treat it like one.
Left alone, AI will hedge. The same biases that make chatbots relentlessly positive sycophants also prevent them from taking strong positions. Without specific instructions, they'll recite ten clauses or considerations and avoid telling you which ones matter.
Force the issue: “Rank the top 3 risks by likelihood and financial impact.”
Now it has to take a position.
That’s closer to how real legal advice works, but you'll get that from an AI if you demand it.
A lot of contract friction comes down to one question: is this standard?
Ask the AI directly: “How does this contract compare to market-standard agreements for SaaS deals of this size?”
You’ll get a baseline. It won’t be perfect, but it’s often good enough to locate the landmines in redlines or third-party paper.
Redlines are easy, getting to signature is harder.
Push the AI to go further than simply marking up language: “Give me a revision, a fallback, and a one-line justification for our preferred position that I can send to the other side.”
Now you’ve got a position, a concession, and a way to explain it.
That’s an actual negotiation toolkit.
Trying to do everything in one prompt usually produces something bloated and shallow.
Break your analysis up into a series of tasks that the AI can clearly complete:
You’ll get cleaner thinking at each step, and you can steer as you go.
Before you complete an Ai-assisted analysis, ask: “What am I missing?” or "What additional information would materially impact my position?"
This is where the AI model tends to be surprisingly useful. It’s good at pattern recognition, which means it’ll often surface edge cases you hadn’t considered or context you forgot to give the AI at the start.
"Measure twice, cut once" is sound advice in general and a critical best practice with generative AI.
This should go without saying, but it often doesn’t: chatbots don't have bar numbers, so anything you send out under your name should have been subject to your actual legal judgement.
Use AI output as a draft from a particularly literal-minded junior associate, not a final work product. Pressure-test anything that feels off, and don’t pass it along without first making sure it's worth putting your reputation behind.
If a prompt isn’t producing useful results, it’s usually missing one of these key components:
Role, who the model is supposed to be.
"You are a lawyer for a mid-sized SaaS company"
Context, what kind of deal or situation this is.
"You are evaluating a vendor agreement"
Task, what you actually want done
"Analyze the liability clause to determine if any obligations or constraints are outside industry standard for this market."
Constraints, what to focus on and what to ignore
"Our standard position is a liability cap of $10 million. We are insured for a maximum of $15 million per incident."
Output format, how the answer should come back
"Return your analysis as a bulleted list. Cite each concerning factor separately, explain why it is non-standard or unreasonable, and use laymen's terms suitable for briefing a CEO or CFO who isn't a lawyer."
Put those five pieces together and most prompts will produce valuable responses on the first try.
These follow the structure above and can be dropped into a chatbot for deal analysis with minimal edits.
Role: Act as in-house counsel for a SaaS company
Context: We’re the vendor reviewing a customer-proposed agreement
Task: Decide whether to sign, negotiate, or escalate
Constraints: Focus only on liability, indemnity, payment terms, and termination
Output: One recommendation -- sign, negotiate, or escalate -- with three short bullets explaining why in business terms
Role: Act as commercial counsel for a SaaS provider
Context: This is an enterprise customer indemnity clause
Task: Identify and assess risk
Constraints: Rank the top 3 risks by likelihood and financial impact; ignore drafting style
Output:
Role: Act as in-house counsel advising non-legal stakeholders in a SaaS company
Context: Agreement is near final and pending approval
Task: Summarize what matters
Constraints: Stick to financial exposure, operational obligations, and termination risk; avoid legal jargon
Output: Three short sections, key risks, key obligations, and a bottom-line recommendation
Role: Act as lead negotiator for a SaaS company
Context: Liability cap discussion with a large enterprise customer
Task: Develop defensible positions
Constraints: No more than three positions; align with typical market outcomes
Output: Each position with a one-line business justification and a fallback option
Role: Act as in-house counsel supporting contract operations
Context: Agreement has been executed and needs to be implemented
Task: Extract actionable obligations
Constraints: Include only obligations that require action; group by function
Output: Table with obligation, owner, timing, and risk if missed
Good prompts aren’t clever; they’re specific.
We discussed this topic in detail with LinkSquares's Ashley Jones and Airbus's Krista Russell in our latest webinar, The Top AI Prompts for Transactional Lawyers. Watch and learn how actual in-house counsels use next-level AI prompting to enhance and accelerate their practice.
Ready to use an AI platform that's designed by and for legal professionals, rather than the default chatbot that came with your word-processor? Take our LinkAI legal assistant for a spin and see how you can promptly improve your contract workflow.