What generative AI can and can't do for founders' legal documents

What AI is good at

Used well, generative AI saves founders time and money on legal work:

• Translating legalese. Asking a model to explain what an indemnity or a limitation of liability clause does, in plain language.
• Preparing for a lawyer. Turning messy notes into a clean list of questions, a timeline, or a summary of the deal terms you agreed.
• Spotting what to ask about. A checklist of topics a SaaS agreement usually covers, so you can see what your draft is missing.
• Internal documents. First drafts of policies, onboarding materials or process docs that a person will review.

Coming to a consultation with organized facts and specific questions makes it more productive, whoever drafted the notes.

Where it goes wrong

It writes for an average company, not yours

A model doesn’t know that your product stores data in the US, that one customer is in Quebec, or that your co-founder never signed an IP assignment unless you tell it, and it won’t always ask. The result reads well and fits nobody in particular.

It mixes jurisdictions

Much of the legal text online is American. Drafts often borrow concepts, statutes or terms that don’t apply in Ontario, or miss Canadian requirements such as CASL consent rules or Quebec’s privacy law.

It invents things, confidently

Models can produce cases, statutory references or “standard” clauses that don’t exist. Canadian courts have dealt with lawyers who filed material citing non-existent cases generated by AI, for example in Zhang v. Chen, 2024 BCSC 285 and Ko v. Li, 2025 ONSC 2766. Ontario’s Rules of Civil Procedure (rule 4.06.1(2.1)) now require a factum to include a signed statement that every authority cited is authentic and every quotation from an authority is accurate. If trained lawyers get caught out, a founder reviewing a contract alone can too.

It can’t weigh trade-offs

Whether to accept an uncapped indemnity from an enterprise customer is a business and risk question. AI can describe the clause. It can’t tell you whether this deal, with this customer, is worth that risk.

Pasting contracts or customer data into a chatbot

Treat it as a disclosure decision. Before pasting anything in, check the tool’s terms and settings: whether inputs are stored, who can access them, and whether they’re used to train models. Business and enterprise versions of many tools have different terms from consumer versions.

Be especially careful with agreements that have a confidentiality clause, customer or employee personal information, and unreleased product plans, pricing, source code or investor terms. See NDAs for startups.

Canada’s federal, provincial and territorial privacy regulators published joint principles for generative AI in December 2023. They ask organizations to consider whether using a generative AI system is necessary and proportionate, and to use de-identified information in prompts where possible. Communications with your lawyer can be protected by privilege; text you paste into a chatbot doesn’t get that protection automatically. An internal AI acceptable use policy helps your team know where the lines are.

How lawyers are expected to use AI

The Law Society of Ontario published a white paper on licensees’ use of generative AI in April 2024, with related resources in its Technology Resource Centre. Its message is that existing professional duties apply, including competence, confidentiality and supervision. When a lawyer uses AI, the lawyer stays responsible for what goes out. When you use AI alone, nobody is. It’s fair to ask any lawyer you hire which tools they use and how they protect your information.

A sensible way to combine the two

• Use AI to prepare. Summarize the deal, list your questions, flag clauses you don’t understand.
• Use a lawyer to decide. Anything you’ll sign, publish to users, or rely on with investors, regulators or customers.
• Keep people in the loop in your own product. If your product uses AI to make decisions about people, plan for review and notice. Quebec requires notice when a decision is based exclusively on automated processing of personal information (section 12.1 of its private sector privacy law).
• Document your own AI use. Customers increasingly ask how you use AI. Clear AI product terms help.

Founders who use AI well tend to spend less on legal work, because the lawyer’s time goes to judgment rather than transcription. A defined review of an AI-drafted document can usually be scoped at a fixed fee.