Personal injury firms use AI on the paperwork around a file — medical chronologies, production summaries, routine correspondence — not on the parts a law society holds a licensee accountable for. The gain is in review time, not in judgment.
Key takeaways
A personal injury practice is not short of judgment. It is short of hours, and most of those hours go into paperwork that nobody would call legal work if it were happening anywhere else: assembling clinical records from six providers, reading them in date order, chasing the seventh provider, reconciling a disbursement ledger, and writing the same status update to forty clients who all want to know the same thing.
That is the part AI is good at. It reads, it extracts, it orders, it drafts. It does not decide whether a claim is worth advancing, it does not advise a client, and it does not sign anything.
Medical chronologies. Given a set of clinical records, an AI tool can produce a dated chronology with a page reference against each entry, which a clerk or lawyer then checks against the source. The value is not that the machine is right; it is that checking a draft chronology against the records is a faster job than building one from nothing.
Production and disclosure summaries. Large productions can be summarised by theme, party or date so counsel reads a map before reading the documents. Every summary should carry the document identifier it came from, so the summary is verifiable rather than merely plausible.
First drafts of routine correspondence. Requests for records, follow-ups to an adjuster, status letters to a client, undertakings charts — all of these are structured, repetitive and reviewable.
Intake triage. A first call and an accident report can be turned into a structured file summary — date of loss, injuries alleged, treatment to date, other parties, insurers involved — so the lawyer reads one page instead of a folder before deciding whether to take the matter.
Ontario licensees do not get a technology exemption, and they do not get a technology prohibition either. The Rules of Professional Conduct say a lawyer must perform legal services to the standard of a competent lawyer, and the commentary added in June 2022 is explicit that a lawyer should develop an understanding of, and ability to use, technology relevant to the nature and area of the lawyer’s practice, while recognising the duty to protect confidential information. The same commentary notes that the required level of technological competence depends on practice area, geography and client requirements.
Delegation is where most AI plans quietly go wrong. Rule 6.1-1 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, with the commentary adding that the lawyer must review that work at frequent intervals to ensure its proper and timely completion. A drafting tool does not change who is answerable; it changes how much material arrives for review, which means the review step has to be designed, not assumed.
Diarising limitation dates is the one part of a personal injury practice where an automation error is unrecoverable. Use software to prompt, never to decide. Ontario’s basic limitation period and the discoverability rules that sit under it are legal questions, not calendar arithmetic — our sister firm sets out the basic limitation period in Ontario and how it applies to a personal injury claim. If your intake tool proposes a date, treat it as a prompt for a licensee to confirm, and keep the confirmation in the file.
The same caution applies to liability analysis. Whether a driver is liable, and on what basis, is a legal conclusion — see motor vehicle negligence liability in Ontario. An AI summary of an accident report is evidence handling, not an opinion.
Ontario auto claims generate a second stream of paperwork alongside the tort claim, because the accident benefits system runs on forms and deadlines with an insurer on the other side. The Financial Services Regulatory Authority of Ontario is the regulator here and publishes consumer guidance on auto insurance, including what to do after an accident and how complaints against an insurer are reviewed. Nothing in that process is decided by a firm’s software, but a great deal of it is correspondence a tool can draft and a person can approve.
Contingency retainers make the fee conversation harder, not easier, because the client is being asked to understand a percentage, a disbursement float and a costs risk at the same time. A model is useful for drafting a plain-language explanation of an account and for reconciling a disbursement ledger into a readable schedule. It is not useful for deciding what to charge. Start from the actual framework — how a contingency fee agreement works in Ontario — and let the tool draft inside it.
A personal injury file is one of the most sensitive record sets a small business will ever hold: clinical history, income, family circumstances. Before any of it is pasted into a general-purpose chatbot, settle three contractual questions — where the data is processed, how long it is retained, and whether it is used to improve the vendor’s models. The Office of the Privacy Commissioner of Canada has published guidance on privacy and artificial intelligence, including principles for responsible, trustworthy and privacy-protective generative AI, and its summary of the fair information principles in Schedule 1 of PIPEDA is the shortest useful checklist for a firm that has never run this analysis.
A three-lawyer Ontario practice with about ninety open files starts with one step: medical chronologies. Records already arrive as PDFs into a matter folder. The tool produces a dated chronology with a page citation per line; a law clerk verifies every line against the source and corrects it; the lawyer reads the verified chronology.
The firm measures two things for one quarter: the elapsed days from records-complete to chronology-approved, and the number of corrections the clerk makes per hundred lines. The first tells them whether it is faster. The second tells them whether it is safe to widen. If corrections stay high on a particular record type — handwritten clinical notes, say — that record type stays manual.
Nothing about intake, limitation diarising, advice or settlement authority changes during that quarter. One step, measured, before the second.
Can AI value a claim?
No. Quantum is a judgment built on the evidence, the medical picture and the law, and a licensee is accountable for the advice given to the client. A tool can assemble the inputs that judgment rests on.
Does using AI need to be disclosed to the client?
Treat it the way you treat any other person or system doing work on the file: your retainer should be accurate about who and what handles client information, and your obligations to protect that information do not change. The Law Society’s rules on competence and confidentiality are the starting point, and firm-specific wording is worth getting reviewed.
What is the single worst use case?
Asking a general chatbot for the law and relying on the answer. Legal research output must be verified against the source; unverified citations have caused real problems in Canadian courtrooms, and the accountability for a filed document sits with the licensee who filed it.
A 30-minute call is enough to tell you whether AI pays for itself here.
