Legal AI Prompts: 8 Templates to Draft and Review Contracts in an SME
A good legal prompt does not ask AI to write a contract from scratch. It gives the tool your template, the governing law and your negotiating position, and it forces the tool to flag what it does not know. Used this way, AI prepares the work; signing off on clauses that bind the company stays a human job.
This article is for SME owners, office managers and HR or purchasing teams who handle contracts without being lawyers: service agreements, terms and conditions of sale, NDAs, purchase orders, amendments. It covers what AI can reasonably do with a contract, the confidentiality precautions, eight ready-to-use prompts and a review routine. The basics of writing instructions are in the prompt engineering guide for SMEs, the starting point of this series.
What AI can and cannot do with a contract
AI is useful for structuring, rewording, comparing and spotting issues; it is not reliable for deciding on its own what protects your company. The line runs between preparing a document and taking responsibility for its legal effect.
An LLM produces plausible text based on what it saw during training. It does not know your history with this client, it can mix rules from several countries, and it sometimes cites statutes that do not exist or have since been amended. In a LinkedIn post, an approximation costs little. In a liability or termination clause, it can cost you a dispute.
The table below gives a simple benchmark for an SME.
| Use | Risk level | What to do |
|---|---|---|
| Rephrase a clause in plain language for the owner | Low | Use freely, then reread |
| Compare two versions of a contract | Low | Use, then check flagged differences in the original |
| Check that a document contains required wording | Moderate | Use with a checklist you provide |
| Adapt a clause from your template to a new case | Moderate | Use, and get it reviewed if the financial stakes are high |
| Spot one-sided clauses in a contract you received | Moderate | Use as a first read, never as the final opinion |
| Draft a full contract without a starting template | High | Avoid, or have it validated by a professional |
| Conclude that a clause is valid or void | High | Leave it to an in-house counsel or a lawyer |
One more point of caution for companies operating in France: giving legal advice and drafting legal documents on behalf of others, on a regular and paid basis, is governed by French Law No. 71-1130 of 31 December 1971 and reserved to authorised professions. Drafting your own contracts with the help of a tool does not raise that issue. If, however, your business consists of supplying ready-made contracts to your clients using AI, have your situation checked by a legal professional.
Before you prompt: confidentiality and framing
A contract almost always contains sensitive data (names, amounts, commercial terms): only paste it into an AI tool from a properly configured business account, after removing anything the task does not need. This rule prevents most incidents.
Three habits before each use:
- Use a business account. The enterprise plans of the main tools let you exclude your conversations from model training. Check that setting in your subscription instead of assuming it. We cover this in our article on AI and data confidentiality.
- Pseudonymise. Replace the other party's name with "Client A", replace amounts with orders of magnitude when the exact figure is not needed, and remove the signatories' personal data. The data minimisation principle of the GDPR applies here too.
- Respect existing confidentiality commitments. If the contract you are analysing contains its own confidentiality clause, sending it to a third-party service may be an issue. Professions bound by secrecy face even stricter constraints, covered in our article on AI and professional secrecy.
On framing, most legal prompts miss one piece of information: the governing law. Write it out explicitly: "French law, contract between businesses" (or between a business and a consumer). Without it, the tool may suggest a clause shaped by US practice, with notions such as "consequential damages" that have no direct equivalent in your contract.
The structure of a reliable legal prompt
An effective legal prompt has six components: the expected role, the governing law, the context of the parties, your position, the output format and an obligation to flag uncertainty. The first five improve quality; the sixth limits silent errors.
- Role: "You are assisting a non-lawyer SME owner." This sets the language level and avoids needless jargon.
- Governing law: France, B2B or B2C, and possibly the sector (construction, software, transport).
- Context: who the parties are, who proposed the contract, the financial stakes, the term.
- Your position: are you the provider or the client, the supplier or the buyer? A limitation of liability clause reads very differently depending on your side of the table.
- Format: table, list of clauses, text ready to paste, one-page memo.
- Uncertainty: ask the tool to separate what is in the document from what it infers, and to cite no statute without labelling it "to be checked".
If you use these instructions every week, put components 1, 2 and 6 in your assistant's standing instructions. The difference between those and a one-off request is explained in our article on system prompts vs user prompts. The general logic is detailed in the anatomy of a good prompt.
8 ready-to-use legal prompts
Each template below starts from an existing document (your template, the contract you received, your terms of sale) rather than a blank page. Replace the items in square brackets.
1. Adapt your contract template to a new assignment
You are assisting a non-lawyer SME owner. French law, business-to-business contract. Here is our standard service agreement template: [paste template]. New assignment: [description, duration, amount, deliverables, specifics]. Adapt only the clauses that need to change for this assignment. For each change, show: the original clause, the new wording, the reason for the change. Do not modify the liability, termination and intellectual property clauses: if they seem unsuitable, flag it in a separate section called "Items to validate".
Starting from your template keeps consistency with your other contracts. The last instruction stops the tool from quietly touching the clauses that carry the risk.
2. Draft a specific clause from your need
French law, B2B contract. I am the [provider / client]. I need a [reversibility, late delivery penalties, confidentiality, non-solicitation...] clause for the following situation: [context]. Offer two versions: a balanced one and one that protects me more. For each, explain in two sentences what it changes in practice. If a statutory rule governs this type of clause, mention it and label it "reference to be checked".
Asking for two versions gives you room to negotiate. The plain-language explanation lets you choose without legal training.
3. Review an incoming contract against your negotiation grid
This is the most profitable prompt for an SME, and the one you rarely see in template lists. It assumes you have written down, once and for all, your acceptable positions: what lawyers call a "playbook".
Here is our negotiation grid: [for example: liability cap equal to the annual contract value; 30-day payment term; termination notice of at least 3 months; no exclusivity clause; French law and the courts of our registered office]. Here is the contract proposed by [the client / the supplier]: [paste]. For each item of the grid, produce a four-column table: our position, what the contract says (exact quote and article number), gap (compliant / minor gap / major gap), suggested rewording. Then list the contract clauses that do not match any item of the grid.
The exact quote is essential: it lets you check within seconds that the tool has not invented or distorted a clause. To make this kind of table more reliable, giving one or two examples of well-filled rows helps a lot, following the few-shot prompting technique.
4. Spot one-sided clauses
French law. I am the [position] and I did not draft this contract: [paste]. Identify clauses that create markedly heavier obligations for me than for the other party: liability, termination, penalties, automatic renewal, unilateral price changes, ownership of deliverables. For each clause, quote it, explain the concrete risk in one sentence and suggest a question to ask the other party. Do not conclude on the legal validity of the clauses: that will be reviewed with an adviser.
Since the 2016 reform of French contract law, the Civil Code provides that in a standard-form contract, a non-negotiable clause creating a significant imbalance is deemed unwritten (Article 1171). AI can help you spot such clauses; whether a judge would strike them out is another matter.
5. Check the payment terms in your terms of sale and invoices
Here are our general terms and conditions of sale: [paste] and an invoice template: [paste]. Check that the following items are present and mark each one "present", "missing" or "ambiguous", with the quote:
- the payment term;
- the late payment penalty rate and how it applies;
- the 40 euros fixed recovery fee;
- [other items specific to your business]. Do not suggest any rule other than those in this list.
Here, you supply the rule, not the tool. The checklist items come from the official French government page "Délais de paiement entre professionnels et pénalités de retard" on entreprendre.service-public.gouv.fr: the payment term must appear on the invoice and in the terms of sale; absent agreement, it is in principle 30 days after receipt of the goods or performance of the service; if agreed, it cannot exceed 60 days from the invoice date, or 45 days end of month if that option is set out in the contract. Late payment penalties cannot be lower than 3 times the legal interest rate, and the 40 euros fixed recovery fee must be stated. Reread that page before building your list, as these rules change.
6. Compare two versions of a contract
Here is version 1 that we sent: [paste] and version 2 returned by [the other party]: [paste]. List every difference, including single-word changes ("may" turned into "shall", a modified deadline, a comma that changes the meaning). Present them in a table: article, version 1, version 2, practical effect for us. End with the three changes that deserve the most attention.
A standard text comparison tool finds the differences; AI adds the interpretation. Still check every difference deemed important in the original document.
7. Translate a clause into plain language
Explain the following clause to an SME owner with no legal training: [paste]. In five sentences maximum: what the clause requires me to do, what it requires the other party to do, what happens if something goes wrong, and a worked example with figures if relevant. If the clause is ambiguous, say so and give both possible readings.
This prompt also helps prepare a negotiation meeting or explain a contract to a business partner.
8. Prepare the meeting with your lawyer
Here is a contract I need to sign: [paste]. Stakes: [amount, term, importance of the client]. Prepare a memo for my lawyer: a ten-line summary of the contract, the five points that worry me and why, the precise questions to ask, and the background information they might ask me for.
For an SME, this is often the best legal use of AI: arriving at your adviser's office with an analysis that is already structured cuts billed time and makes the discussion more useful.
Checking AI legal output: a four-step routine
Any legal output from an AI tool is checked before use, and the level of checking depends on the stakes: a rewording gets reread, a binding clause gets validated. Four checks cover most cases.
- Quotes. Every passage presented as taken from the contract must appear there word for word. A search in the document is enough.
- Legal references. Every article number cited by the tool must be checked on Légifrance, the official French law database. Invented or outdated references are among the most frequent AI hallucinations on legal topics.
- Internal consistency. A modified clause can contradict another clause (three months' notice in one article, one month in another). Ask the tool, in a new conversation, to look for contradictions between articles.
- A second pair of eyes. Have the output reviewed by another conversation or another tool, with instructions to look for errors rather than confirm. Asking the AI to spell out its reasoning step by step, using the chain of thought method, also makes its mistakes easier to see.
In every case, the company is bound by the text it signs, not by the analysis that came before.
Building your library of clauses and prompts
The real gain does not come from a single prompt but from a library: your validated templates, your negotiation grid and your tested prompts, all in one place. Once it exists, every new contract starts on a safe basis.
In practice, an SME can organise three things:
- Validated templates: your standard contracts reviewed by a professional, used as the starting point for prompt 1.
- The negotiation grid: one page that sets your acceptable positions, used by prompt 3.
- Tested prompts: the eight templates in this article, adapted to your vocabulary and saved in a shared space.
The project features of AI assistants let you store these documents as permanent context; it is one of the use cases described in our article on Claude for SMEs. When the volume of contracts grows, the question of a search system over your documents comes up: our RAG vs prompting comparison helps you choose. For long contracts, the techniques in our article on document summary prompts also apply.
A final note on ownership of text produced with AI: it raises specific questions, covered in our article on AI and intellectual property.
Frequently asked questions
Can you draft a contract with ChatGPT or Claude?
Yes, to prepare an outline, adapt an existing template or reword a clause. No, to replace the validation of a high-stakes contract: the tool can mix national laws, leave out a required item or cite an inaccurate rule.
Is a contract drafted by AI valid?
A contract's validity does not depend on the tool used to write it but on its content and the parties' consent. A contract drafted with AI is valid or not for the same reasons as any other, which is why its content must be checked.
Can you share a client contract with an AI tool?
With a business account that excludes your data from training, after pseudonymisation, and after checking that the contract contains no confidentiality commitment that forbids it. With a free personal account, the answer is no.
Can AI cite reliable legal references?
Not reliably. Article numbers and rules cited must always be checked against the official source. The safest method is to supply the rule yourself, as in prompt 5.
Which prompt should you use to review a contract before signing?
Prompt 3, with your negotiation grid, then prompt 4 for one-sided clauses. If the stakes are high, finish with prompt 8 to prepare the discussion with your lawyer.
Do you need a specialised legal AI tool?
Not to get started. Business versions of general assistants cover the eight uses in this article. Specialised tools become worthwhile when contract volume or case-law research needs grow.
How GrowthPerf helps
Using AI on contracts saves time as long as you stay in control: start from your templates, set your position, demand quotes and check according to the stakes. GrowthPerf, a Qualiopi-certified training provider, trains owners and teams in SMEs, startups and nonprofits in these practices, using their own documents.
The Operational AI course includes workshops on your real cases: building your negotiation grid, a library of legal prompts, confidentiality rules suited to your business. For a programme focused on prompting techniques, see our page on prompt engineering training. Depending on your situation, these courses may be funded by your OPCO (French training fund).
For the full set of techniques, go back to the complete prompt engineering guide for SMEs, and to the glossary for the basics.
Reviewing contracts or terms of sale without an in-house lawyer? Book a free 30-minute audit: we test these prompts together on one of your documents and define what can be prepared with AI and what needs an adviser's sign-off.