Practical AI for Canadian legal professionals.
Canadian lawyers can capture large, real time savings from AI today — Matthew Peters, partner and National Leader of Transformation at McCarthy Tétrault, says of AI research: "Instead of taking 10 hours to dig around … you maybe do it in two or three hours" (Canadian Affairs, Apr 27, 2026) — but only inside a tight professional-conduct frame: never put unredacted client information into consumer ChatGPT/Claude/Gemini, verify every citation against CanLII/Westlaw/Lexis, and in Ontario certify the authenticity of every authority in any factum (Rule 4.06.1(2.1), in force December 1, 2024).
The hours you get back are billable. That's the whole point.
Everyone sells AI to lawyers as "save time." For a practice that bills by the hour, time is the product, so saved time only matters if you know where it goes. The research says the real constraint isn't your workload, it's your capacity: firms bill barely a third of the workday. Recovered hours are raw revenue waiting to be redirected.
- 74% of hourly billable tasks are automatable (Clio)
- Correspondence, drafting, and document review are the biggest time sinks
- Solo firms bill only 27% of the day — the most to gain
- Bill the recovered capacity, or take on more matters
- 12% of billable work never gets billed today — close that gap
- Move AI-accelerated work to flat fees, so efficiency is yours to keep
- Growing firms use automation ~2× more than stable ones
- Systematized workflows make the practice less dependent on you
- A documented, repeatable practice is worth more when you sell or merge
We won't promise a revenue multiple. What we will show you is the mechanism: the hours AI gives back are the same hours you couldn't bill before, and the firms that grow are the ones that redirect them deliberately instead of just working less. The compliance guardrails in this program exist so those gains never come at the cost of a sanction, a privilege waiver, or a client's trust.
Almost every firm uses AI now, and few are billing more because of it.
Clio puts adoption at 79%, but most solo and small-firm use is experimental dabbling in ChatGPT, and the share using legal-specific tools has actually fallen to 40%. The activity is real and the billable gain usually is not, because the constraint was never a shortage of things to automate. Firms bill barely a third of the workday, and speeding up work you already were not billing does nothing for that.
“I think it is more efficient for me to determine that something is not a priority than for me to automate something that is not a priority.”
That is the discipline we teach: decide what deserves your recovered hours before you automate anything. We start from your real constraint, billable capacity, point AI at the drafting and research that frees it, and drill the verification and confidentiality rules so the gains never cost you a sanction. Fewer tools aimed at billable work beat a firm full of apps nobody bills for.
Three things to hold onto before anything else.
Large, real time savings are available today, but only inside the professional-conduct frame: never put unredacted client information into consumer ChatGPT/Claude/Gemini, verify every citation against CanLII/Westlaw/Lexis, and in Ontario certify the authenticity of every authority in any factum (Rule 4.06.1(2.1), in force December 1, 2024).
The highest-leverage, lowest-risk starting workflows are drafting (correspondence, first drafts, plain-language client explanations), summarizing long documents, and turning plain-English questions into research starting points — not unsupervised legal research, which is where every Canadian sanction has occurred (Zhang v. Chen, Ko v. Li, Hussein v. Canada).
A paid business-tier general assistant (ChatGPT Team/Enterprise or Claude Team) for drafting, free CanLII AI summaries + CanLII Search+ for Canadian case law, and one Canadian-built legal tool (Spellbook for contracts, Alexi for litigation research, Clio Duo if already on Clio, Blue J for tax) — with data-residency and no-training terms confirmed in writing before any client data goes in.
What the research says about lawyers and AI.
Headlines up front; expand any card for the full finding with sources.
01The problem is capacity, not workload33% billed
Clio's 2025 Legal Trends Report finds firms bill only “the industry average of 33% of their workday, or roughly three hours of billable time per day,” and with an average realization rate of 88% “12% of the time a firm puts towards billable work never actually gets billed.” It estimates “up to 74% of hourly billable tasks—such as information gathering and data analysis—could be automated with AI” (81% of legal-secretary tasks vs. 57% of lawyers' tasks). Solo and small firms have the lowest utilization (27% and 32%) because they “wear many hats.” This is the workshop's core value proposition: AI recovers lost non-billable time.
02Adoption is high but shallow among the target audience79% use AI
Clio reports 79% of legal professionals use AI (a figure that debuted in its 2024 report, up from 19% in 2023, and held flat in 2025), but solo/small-firm use is mostly experimental ChatGPT dabbling — and “only 40% of legal professionals are using legal-specific AI solutions, down from 58% in 2024.” Clio's Canada-specific report (Canada's AI-Powered Law Firms Are Pulling Ahead) states 66% of Canadian legal professionals say AI has improved their firm's revenue. Large firms lead adoption (87%); “71% of solo law firms still report using AI.” “More than half of legal professionals (53%) say their firm has no AI policy or that they are unaware of one.”
03Every Canadian AI sanction to date is a verification/candour failure, not a tool failure0 tool failures
Zhang v. Chen (BCSC, 2024), Ko v. Li (ONSC, 2025), and Hussein v. Canada (FC, 2025) all involved fake or mis-cited authorities that a human check would have caught. The lesson for the curriculum: the danger is unverified output reaching a court, not AI itself.
04Ontario's rule is certification, not disclosureCertify, not disclose
Since December 1, 2024 (O. Reg. 384/24), Ontario's Rules of Civil Procedure require a signed statement certifying the authenticity of every authority cited in a factum (Rule 4.06.1(2.1)) and in expert reports — regardless of whether AI was used. Authorities from CanLII, government/court websites, and commercial publishers are presumed authentic. The Federal Court, Manitoba, Yukon, Nova Scotia and others instead require disclosure of AI use. This is the single most important provincial difference.
05Data residency is not data sovereignty; both matter for privilegeResidency ≠ sovereignty
Consumer AI tiers train on inputs and retain data (≥30 days), and a US court (Heppner, Feb 2026) has held consumer-AI exchanges non-privileged. Canadian-hosted or no-training enterprise tiers substantially mitigate this, but US-owned vendors remain subject to the US CLOUD Act regardless of server location; PIPEDA and Quebec's Law 25 raise the bar further for cross-border transfers.
Seven modules, ordered by risk — drafting before research.
Green modules are the confidence-builders. Red modules are where every Canadian sanction has happened, so they come with verification drills attached.
Each task, worked end to end.
Every module runs the same loop: task, AI approach, tools, a real prompt, realistic time saved, and the conduct flag. Expand for full detail.
01Legal research and case-law findingFinding relevant Canadian authorities; orienting in an unfamiliar area.High caution · verify everything
02Contract drafting, review and redliningFirst-draft agreements; redline third-party paper; flag missing/risky clauses.Medium · confidentiality terms first
03Document review, due diligence and discovery summarizationReviewing large document sets; extracting key facts; building chronologies.Medium · privilege on the line
04Correspondence and plain-language client explanationsRoutine letters, emails; translating legal concepts into plain language.Low risk · start here
05Memo and brief draftingFirst-draft research memos; structure for facta.Highest-risk output
06Meeting transcription and summarizationClient-meeting and examination notes/summaries.Highest hidden risk
07Administrative drafting, intake, time entry and billingTime-entry assistance, matter summaries, intake, marketing/admin text.Low risk
The Canadian-viable tool landscape.
Prices are largely third-party estimates — most legal-AI vendors do not publish per-seat pricing. All USD unless noted.
General assistants (ChatGPT, Claude, Gemini, Copilot) — the full pictureCheap and excellent for drafting; unsafe for client data on consumer tiers
Full comparison table — Canadian coverage, data terms, pricingExpand for the tool-by-tool breakdown
| Tool | Canadian coverage | Data residency / training | Pricing (2026) | Learning curve | Clio integration | Best for |
|---|---|---|---|---|---|---|
| CanLII Search+ / AI summaries | Native, authoritative, free; grounded in CanLII text | Canadian non-profit (FLSC); no training on your data | Free (daily query limits) | Very low | No | Everyone — first stop for CA case law |
| Spellbook (Toronto) | Not a research tool (contracts only) | No-training / zero-data-retention; runs in Word | No public price; ~$99–$200/user/mo per third-party estimates; 7-day free trial | Low (in Word) | Not core | Transactional solos/small firms |
| Alexi (Toronto) | All provinces/territories + federal except Quebec | Closed system, encrypted, no training on client data | Quote-based; 14-day trial (2 memos) | Low–medium | Via workflow | Litigators (non-Quebec) |
| Clio Duo / Manage AI (Burnaby, BC) | Works on your matter data, not a case-law DB | Built on Azure OpenAI; firm data isolated | Base Clio ~$49–$149/user/mo; AI in higher tiers (~$49–$59 add-on per third-party estimates) | Low | Native | Firms already on Clio |
| Blue J (Toronto) | Canada + US + UK tax law | No training on inputs | Quote-based; ~US$1,500/user/yr entry (CurateSuite est.); 7-day trial | Medium | No | Tax practitioners |
| CoCounsel Legal Canada (Thomson Reuters) | Best-in-class: Westlaw Advantage Canada + Practical Law Canada | Canadian data residency; no training on user data | Quote-based (1/2/3-yr terms); premium | Medium | Via DMS/M365 | Research/litigation firms with budget |
| Lexis+ with Protégé (LexisNexis Canada) | LexisNexis Canada case law + citator; launched CA July 2025 | Enterprise data controls | Quote-based; ~US$128+/user/mo (third-party est.) | Medium | No | Lexis-ecosystem firms |
- Harvey (BigLaw enterprise, reportedly ~$1,200+/seat — out of scope for solos)
- Generic “legal GPT” wrappers with no Canadian grounding
- Consumer note-takers marketed to lawyers
- Business-tier general assistant for drafting (ChatGPT Team/Enterprise or Claude Team)
- Free CanLII AI summaries + Search+ for Canadian case law
- One Canadian-built legal tool matched to your practice
- No-training and data-residency terms confirmed in writing
Six duties. No law society bans AI; all six duties still apply.
No Canadian law society bans AI. All build on the Federation of Law Societies' Model Code (Competence 3.1, Confidentiality 3.3, Supervision 6.1). Ontario, BC, Alberta, Saskatchewan, Manitoba, Nova Scotia, PEI, NWT, Newfoundland & Labrador, and Quebec have all issued guidance. Ontario's LSO published a white paper (April 2024) and the companion “Generative AI: Your Professional Obligations” (April 10, 2024), mapping six duties.
1Rule 3.1CompetenceTap for the duty in full
2Rule 3.3ConfidentialityTap for the duty in full
3To the clientHonesty and candourTap for the duty in full
4Rules 6.1–6.2SupervisionTap for the duty in full
5Rule 3.6Reasonable feesTap for the duty in full
6Rule 5.1Not misleading the tribunalTap for the duty in full
Confidentiality & privilege deep-dive: Heppner, the CLOUD Act, and the “necessary intermediary” questionThe prudent assumption: self-initiated consumer-AI chats are not privileged
Court obligations by jurisdiction
Verify current — these move fastOntarioCertification, in force Dec 1, 2024
Rules of Civil Procedure require certification of the authenticity of every authority in a factum (Rule 4.06.1(2.1)) and in expert reports/Form 53 (O. Reg. 384/24, in force Dec 1, 2024). CanLII/government/court/commercial-publisher sources presumed authentic. No general AI-disclosure requirement. Tribunals Ontario issued its own AI practice direction (April 2025).
Federal CourtDeclaration required in the first paragraph
Notice to the Parties and the Profession (Dec 20, 2023; updated May 7, 2024) requires a declaration in the first paragraph of any document containing AI-generated content, plus human verification and reliance only on “well-recognized and reliable sources.” Amended Consolidated Practice Guidelines (June 2025) warn of adverse costs/contempt; separate Sept 29, 2025 interim principles govern the Court's own AI use.
ManitobaKing's Bench, June 23, 2023
Materials must “indicate how artificial intelligence was used.”
YukonSupreme Court, June 26, 2023
Must advise the court of the AI tool used and for what purpose.
AlbertaTri-court notice, Oct 6, 2023
Caution; rely exclusively on authoritative sources (CanLII, court sites, commercial publishers); verify with “meaningful human control.”
British ColumbiaCourt of Appeal Filing Directive, 2024/2025
Reminds litigants they are responsible for authenticity/accuracy of all materials (no AI-specific rule).
Nova Scotia, Newfoundland & Labrador, Quebec Superior CourtAI Governance Framework, Fall 2025
Various integrity/disclosure requirements.
The memorable guardrails.
Every Canadian sanction so far is a verification or candour failure that a human check would have caught.
Zhang v. Chen2024 BCSC 285 — the first Canadian caseTwo ChatGPT-hallucinated cases; costs ordered personally.
Ko v. LiONSC 2025 — a three-decision saga (2766 / 2965 / 6785)Candour after an AI error matters more than the error itself.
Hussein v. Canada2025 FC 1060Concealed AI use; special costs, considered payable personally.
Reddy v. Saroya2025 ABCA (Sept 26, 2025)Hallucinated cases in a factum; “should not expect leniency.”
Specter Aviation v. Laprade2025 QCCS 352$5,000 fine for AI-invented “jurisprudence.”
United States v. HeppnerSDNY, Feb 10, 2026 — US transfer taleConsumer AI use = no privilege.
Why it stalls, and what successful adopters do.
- Reliability distrust after hallucination stories (>80% of midsize-firm leaders report lawyer fear centred on reliability, per the 2026 Future-Ready Lawyer survey)
- Confidentiality/data-security fear (46% cite data security as a barrier, per the 2025 Legal Industry Report)
- No training (54% of firms provide none and have no plans to)
- Time (27% of solos and 33% of small firms cite finding time as the top barrier, per Clio)
- Risk aversion and the billable-hour disincentive (hourly billing rewards spending more time, not less)
- Start with one low-risk, high-frequency workflow (drafting/summarizing) rather than research
- Use legal-specific tools with confidentiality terms rather than raw consumer ChatGPT
- Adopt a written AI policy
- Keep a human in the loop with a verification checklist
- (For solos) exploit flat-fee models where efficiency gains flow to the lawyer
Staged, do-this-today.
Safe defaults and a one-page policy
- In any consumer ChatGPT/Claude account, turn off model training (Settings → Data Controls); better, obtain a business/team tier before touching anything client-related.
- Create a free CanLII account; use CanLII Search+ and CanLII AI summaries for Canadian case-law orientation.
- Use a general assistant only for non-confidential drafting, summarizing public documents, and plain-language client explanations, with anonymized inputs. Rule of thumb: if you wouldn't post it publicly, don't type it into a consumer tool.
- Adopt a one-page firm AI policy (approved tools; what never goes in; the verification + Ontario-certification step before any filing). The LSO's model office-policy language (as adopted by Ms. Lee post-sanction) is a usable template.
One Canadian-grounded paid tool
Research tools and flat fees
- New court practice direction or LSO guidance in your jurisdiction → re-verify obligations before your next filing.
- A tool cannot confirm no-training terms and Canadian/enterprise data handling → do not put any client data in it.
- Any hallucination reaches a draft filing → treat as a stop-the-line event; the sanctions above turned on failure to verify and, worse, failure to be candid.
Caveats and sources.
Caveats — read before teaching
- This area moves fast. Court directions and law society guidance change frequently; re-verify every court obligation above against the official source (lso.ca, fct-cf.ca, provincial court sites, FLSC, CBA) before relying on it. Publication dates are noted throughout; the LSO white paper and obligations resource are both April 2024 and may be updated.
- Pricing is largely opaque. Spellbook, Alexi, Blue J, CoCounsel Legal Canada, and Lexis+ with Protégé do not publish per-seat pricing; figures here are third-party estimates and should be confirmed by quote.
- Vendor vs. independent evidence: Alexi's "80%," CoCounsel's "one-third" (Forrester TEI), and Lexis's ROI figures are vendor or vendor-commissioned; the McCarthy Tétrault "10 hours → 2–3" figure (Canadian Affairs) is the strongest independent Canadian data point.
- Privilege law is unsettled in Canada regarding AI; treat consumer-tool inputs as potentially discoverable, and note Heppner is US authority, not binding in Canada.
- Quebec differs materially: Law 25 cross-border-transfer and consent rules, French-language obligations, and Alexi excludes Quebec; CanLII AI enrichment is also capped for Quebec's large corpus.
Sources, with publication dates
- LSO, Generative AI: Your Professional Obligations — April 10, 2024; LSO Futures Committee White Paper on Licensee Use of Generative AI — April 2024.
- Federation of Law Societies, Model Code of Professional Conduct; CBA, Ethics of AI for the Legal Practitioner (toolkit) — Nov 2024.
- Federal Court Notice to the Parties and the Profession — Dec 20, 2023 / updated May 7, 2024; Amended Consolidated Practice Guidelines — June 2025; interim AI principles — Sept 29, 2025.
- Ontario O. Reg. 384/24 (Rule 4.06.1 certification) — in force Dec 1, 2024 (OBA / Court of Appeal Civil Rules Committee).
- Alberta tri-court Notice (LLMs) — Oct 6, 2023; Reddy v. Saroya, 2025 ABCA — Sept 26, 2025.
- Zhang v. Chen, 2024 BCSC 285; Ko v. Li, 2025 ONSC 2766 / 2965 / 6785; Hussein v. Canada, 2025 FC 1060; Specter Aviation c. Laprade, 2025 QCCS 352; United States v. Heppner (SDNY) — Feb 10, 2026.
- Clio 2025 Legal Trends Report (Oct 16, 2025) and Canada-specific Legal Trends for Canadian Law Firms (2026); Clio blog resources on AI use cases (2025–26).
- Osler, Using generative AI to provide legal services and Artificial advocacy (2024–25); McCarthy Tétrault, Bennett Jones, Cassels, WeirFoulds, Fasken commentary (2024–26).
- CanLII/Lexum blog posts on AI case summaries and CanLII Search+ (2023–26); FJA Action Committee overview of CanLII/Lexis+ AI/Westlaw Edge Canada.
- Thomson Reuters CoCounsel Legal Canada launch materials and Forrester TEI (2026); LexisNexis Canada / Protégé (July 2025).
- Spellbook, Alexi (Business Wire, Oct 22, 2024), Blue J, Clio Duo product/pricing pages and third-party reviews (Lawyerist, AI Vortex, Casefleet, 2025–26).
- Canadian Affairs, "AI puts the billable hour on trial" — April 27, 2026 (Matthew Peters / McCarthy Tétrault; Danny Kastner).
- ABA Law Technology Today and GPSolo eReport on AI note-takers and legal writing (2025–26); AI meeting-recorder litigation coverage (Otter/Fireflies, 2025–26).