Are small law firms using AI? What the Canadian numbers show

Of 344 Canadian law firms surveyed, 63% said they do not use generative AI at all. Generative AI means tools like ChatGPT that write text on request. The profession's reputation for caution turns out to be earned.

If you run a small firm, or you are a client of one, this matters in two ways. The tools are real and some firms are saving time with them. The rules around them are also stricter than in most trades, and small firms are the ones least likely to have written them down.

What the survey actually measured

The numbers come from a Best Law Firms survey published in December 2025. It covered 344 Canadian firms employing more than 14,800 lawyers, about 11% of the country's active lawyers. That is a good sample, but it is one survey, a year old by the time you read this, and the firms chose to answer.

Among firms with more than 20 lawyers, 80% were either exploring generative AI (46%) or piloting it in select practice areas (37%). Only 7% had rolled it out across several practice areas. About one in ten had no plans to adopt it.

A separate Clio report covered by Canadian Lawyer in May 2025 found the same split by size. Larger firms were further along, with 59% reporting broader adoption and 11% reporting full implementation. Most solo and small-firm lawyers described their own use as minimal, mostly experimenting with ChatGPT. That article does not give a percentage for the small firms, so treat "minimal" as a direction, not a measurement.

Where the time goes: research and marketing, not courtrooms

Among firms that use or plan to use AI, the most common task was legal research, at 63%. Marketing content and client pitches came next at 56%. Knowledge management, contract and document drafting, and routine correspondence each came in just under half. Using AI for litigation risk or strategy was rare, about one firm in five.

That ordering is sensible. Drafting a newsletter or summarizing a long document is low stakes if you read the result. Predicting how a judge will rule is not.

The first uses firms pick are the ones where a human reads every word before it leaves the building.

The billing surprise

Of the 37% of firms that do use generative AI, the billing results were modest. 17% saw no change in billable hours. 11% saw fewer billable hours on certain tasks. 9% reported higher efficiency with no change in billables.

For a small firm billing by the hour, this is the uncomfortable part. A tool that halves the time to draft a letter also halves the hours you can bill for it. The same survey notes that firms have automated about a fifth of their administrative work, and small firms report more of it (23%) than large or midsize firms (15%). The gain shows up in admin, where nobody bills a client.

The Law Society of Ontario (LSO) is clear on the billing side. Its white paper says that if you bill by the hour, you can only charge for the time actually spent, and it points to alternative fee arrangements as an option if they are fair and reasonable.

The gap that matters: policies, training and consent

This is where the size of the firm shows. The survey asked whether firms have a clear AI policy for client matters, whether they train lawyers on AI's limits, and whether they get client consent.

About two-thirds of firms with 100 or more lawyers have a policy for client matters. Half of midsize firms do. A third of small firms do.

Training is closer across sizes than policy is. Small firms are close to the others on teaching people the risks, but far behind on writing the rules down.

Consent is lower everywhere:

Only 11% of firms actively tell clients about their AI use as part of their pitch. 17% disclose it if asked, and 7% never disclose it. The survey does not say how many of the rest simply never come up.

What Ontario's rules ask of a small firm

The LSO has not written a separate AI rule. Its April 2024 white paper applies the existing conduct rules for lawyers and paralegals.

The points a small firm needs to know:

• Competence. The licensee must understand the tool they use: its terms of service, its limits, how it handles data. The paper suggests taking a training module before adopting a product.
• Verification. The licensee is responsible for the work product. The paper says it is the licensee's "express obligation" to make sure the output meets the standard of a competent licensee.
• Confidentiality. Some AI providers use what you type to train their tools. The paper says not to enter confidential or privileged information without adequate security, and that anonymizing is not perfect: a tool may be able to piece the facts together.
• Supervision. Treat AI output like work from an assistant you must directly supervise.
• Disclosure. Consider telling the client when they would reasonably expect a person to have done the work, when a court requires it, or when there is a reputational risk to them.

Privacy law may also apply to what you put into a tool, so check with your own counsel or insurer if you are unsure.

What goes wrong: a real Ontario case

In May 2025, in Ko v. Li, an Ontario Superior Court judge, Justice Fred Myers, could not find the cases cited in a family law factum. The hyperlinks did not lead to the cited cases, and at the hearing counsel could not provide proper citations. He noted that tools such as ChatGPT can produce fake citations, though the report does not confirm AI was used. He ordered counsel to show cause why she should not be held in contempt.

The lesson is not that the tool failed. It is that a lawyer who files a document is the one answering for it. Later steps in the case have been reported differently by different outlets, so read the court's decisions rather than a summary if it matters to you.

A five-question check before you start

1. Does the tool's contract let the provider train on what you type? If so, turn it off or pick another tool.
2. Is the first task one where a person reads every word before it leaves the firm?
3. Is there a written, one-page rule on what may never go in: names, file numbers, privileged advice?
4. Who verifies citations, quotes and dates, and where is that recorded?
5. If you bill hourly, how will you bill for time the tool saved?

Where to start

1. Write the one-page policy this week. Even a short one puts you ahead of two in three small firms in that survey.
2. Pick one low-risk job, such as a firm newsletter or a first-pass summary of a public document, and run it for a month.
3. Read the LSO white paper once, start to finish. It is short and written for licensees.

If you want help choosing a tool or setting up the workflow, you can book a call with Trimotech.