Yes — an Ontario licensee can use AI to produce a draft, in the same way they can use a precedent bank or a law clerk, provided the licensee reviews the output and takes responsibility for it. What has not changed is who may hold themselves out as providing legal services, and whether the finished document has been executed in a way Ontario law recognises.
Key takeaways
Nothing in Ontario’s professional rules prohibits a licensee from using software to produce a draft. The rules are written about outcomes and responsibility, not about the tool. Rule 6.1-1 of the Rules of Professional Conduct requires a lawyer to assume complete professional responsibility for their practice of law and to directly supervise non-lawyers to whom particular tasks and functions are assigned.
Alongside that, rule 3.1-1 of chapter 3 defines competence to include writing and drafting among the skills a competent lawyer applies, and rule 3.1-2 requires legal services to be performed to the standard of a competent lawyer. Those obligations attach to the licensee whatever produced the first draft.
So the honest answer to "can AI draft legal documents in Ontario" is: it can produce a draft, and a licensee then has to do the part that makes it a legal document.
Firms often reach for the delegation framework by analogy, and it is worth being precise about why it does not apply. By-Law 7.1 defines a "non-licensee" as an individual engaged by a licensee to provide services, over whose provision of services the licensee has effective control. Software is not an individual, so the by-law’s scheme is not the governing instrument for a drafting tool.
That is not a loophole; it is the opposite. The by-law exists to allocate responsibility when work is done by someone else. When the work is done by a tool, there is no one else, and the responsibility sits entirely on the licensee under rule 6.1-1.
The by-law is still instructive as a standard of care. It requires a licensee to assign only tasks a non-licensee is competent to perform, to review performance at frequent intervals, to assume responsibility for all tasks performed by a non-licensee including all documents they prepared, and to ensure the non-licensee never acts finally in respect of the client’s affairs. A firm that applied those four rules to its drafting tools would have a defensible policy.
There are two very different situations behind the same question. A licensee using AI inside their own practice is using a tool. A person who is not licensed using AI to produce legal documents for other people is doing something else, and the answer there turns on Ontario’s regime for who may provide legal services — a question to put to the Law Society of Ontario directly rather than to a vendor.
There is a third case in the middle: a business using AI to produce documents for itself. That is not the practice of law, but the documents still have to work, and a template that has never been reviewed by anyone can fail at exactly the moment it matters. Which clauses carry the weight is set out in this breakdown of key service-agreement clauses in Ontario.
For a licensee, the practical consequence is scope of retainer. If you use AI to produce a document quickly and cheaply for a client, be clear about what you did and did not review — and if you are limiting the work, how limited-scope retainers work in Ontario is the framework to do it inside.
Drafting speed is wasted if execution fails. In Ontario, whether an electronic signature binds depends on the document and the circumstances; the enforceability of electronic signatures on Ontario contracts is the place to start, and the categories of Ontario documents that are excepted matter more than most people expect — wills and certain other instruments do not follow the general rule.
The distinction between an electronic signature and a digital signature also gets blurred in procurement conversations; the difference between the two in Ontario is worth settling before you choose a platform.
None of this is affected by how the draft was produced. It is simply the part of the workflow that AI does not touch, and the part where a fast process most often falls over.
One: name the tools that may be used and the documents they may be used on. An open policy is not a policy.
Two: require the source material to be the firm’s own precedents wherever a precedent exists, so the output is bounded by work you have already approved.
Three: require a named licensee to review every draft before it leaves the firm, and record that they did.
Four: prohibit putting client-identifying facts into any tool the firm has not contracted with, because rule 3.3-1 requires all information concerning the business and affairs of the client to be held in strict confidence and the Privacy Commissioner’s generative AI principles describe how freely such information can travel.
Five: check the practice directions of any court you file in before AI-assisted material reaches a registry, because that expectation is set by the court rather than by the Law Society. For the professional-obligations side, Ontario’s Technology Resource Centre gathers supports on technological competence including artificial intelligence, and the Law Society of British Columbia publishes guidance on professional responsibility and generative AI that is worth reading whichever province you practise in.
Does the Law Society ban AI drafting?
The rules do not address the tool. They address responsibility: rule 6.1-1 requires the licensee to assume complete professional responsibility for their practice, and competence and quality-of-service obligations apply to the finished work.
Do we have to tell clients we used AI?
The Rules of Professional Conduct do not impose a general disclosure rule, but clients increasingly ask and some courts have their own expectations for filings. Deciding your answer before the question arrives is better than improvising it.
Can a paralegal use these tools?
A licensed paralegal works within their own scope of practice and carries their own professional obligations. The tool does not widen the scope, and the same review and responsibility principles apply.
A 30-minute call is enough to tell you whether AI pays for itself here.
