AI Affidavit Drafting for Ontario Lawyers: 2026 Guide

AI can help Ontario lawyers organize an affidavit source record, prepare a chronology, map proposed paragraphs to supporting material, and create a working draft. It should not decide whether evidence is admissible, what the deponent knows, which facts should be sworn, or whether the affidavit is ready for service or filing. A reliable workflow begins with the purpose and governing requirements, separates facts from inference and advocacy, links every material paragraph to a reviewed source or instruction, records gaps, and gives the deponent and responsible lawyer meaningful control over the final text.

Key facts and limits

An affidavit is not merely a polished narrative. Its content depends on the deponent, the proceeding, the purpose for which it is offered, the governing rules, and the source record. The same sentence may be a directly remembered event, a summary of a business record, a statement attributed to another person, an inference, or argument. An AI system can classify candidates, but the label is only a review aid.

Keep drafting assistance separate from legal conclusions. A model can identify that a proposed paragraph contains a date and a quoted email. It cannot establish that the email is authentic, that the proposed deponent has the required knowledge, that the quotation is complete, or that the paragraph is admissible. Fluency should never substitute for those decisions.

Define the affidavit mandate before drafting

Start with a short mandate approved by the lawyer responsible for the matter. Identify the proceeding, court or tribunal, proposed deponent, purpose, issues in scope, procedural step, current due date, source package, requested output, reviewer, and stopping rule. Link the current rules, practice directions, orders, and forms that counsel has selected. Do not ask a model to infer the test, relief, deadline, or evidence needed from a folder of documents.

The mandate should also state what the first pass is not. It may be a chronology and paragraph map rather than a finished affidavit. It may exclude privileged communications, settlement material, expert opinions, or documents awaiting instructions. It may route procedural questions to counsel without proposing answers. Those boundaries make the resulting draft easier to supervise and reduce accidental overreach.

Useful mandate fields

• Purpose and audience: the specific step the draft supports and who is expected to read it.
• Deponent: identity, role, relevant time period, and the proposed basis of knowledge—subject to confirmation.
• Issues in and out: the factual topics to cover, exclusions, and points awaiting a lawyer decision.
• Authorities: current rules, forms, orders, and practice directions selected by counsel.
• Source boundary: included files, expected but missing materials, privileged sets, and unreadable exceptions.
• Deliverable: chronology, paragraph map, exhibit register, working draft, questions, or comparison report.

Build and reconcile the source set

Inventory the materials before asking for prose. Depending on the mandate, the approved set may contain pleadings, orders, correspondence, contracts, business records, photographs, transcripts, interview notes, prior affidavits, and candidate exhibits. Give each file a stable identifier. Record its date, version, custodian or origin where known, readability, relationship to other files, and whether it is included, excluded, privileged, duplicated, superseded, or missing.

Preserve document relationships. An email may refer to an attachment; a chronology entry may rely on a later correction; a signed agreement may differ from the draft in counsel's notes; a transcript answer may require the surrounding questions. Keep families and context together. If pages, tables, handwriting, annotations, or embedded objects cannot be processed reliably, create an exception instead of allowing silent omission.

An eight-step AI affidavit drafting workflow

1. Confirm the mandate and environment. Record the proceeding, purpose, approved sources, applicable materials selected by counsel, reviewer, and authorized workspace.
2. Inventory and reconcile. Count files and pages, connect document families, preserve versions, and route missing or unreadable material to an exception queue.
3. Extract observable facts. Capture people, dates, events, communications, amounts, record entries, quotations, and document references without turning them into conclusions.
4. Create a chronology and issue map. Sort candidates by time and issue while keeping conflicting accounts, corrections, and uncertainty visible.
5. Classify proposed knowledge basis. Ask what source and proposed knowledge basis support each point. Treat the result as a question for deponent and lawyer review.
6. Draft through a paragraph map. Give each proposed paragraph an identifier, neutral proposition, source locator, candidate exhibit, gap, and review status before polishing prose.
7. Run deponent and lawyer review. The deponent corrects voice, knowledge, and facts; counsel checks scope, procedure, evidence, argument, privilege, and exhibits.
8. Reconcile the final package. Compare the approved text with the paragraph map and exhibit register, resolve or preserve every exception, and follow the applicable execution and commissioning process.

Staging matters. A single prompt that asks for a finished affidavit hides whether an error came from a missing document, poor extraction, a chronology conflict, an unsupported inference, or invented connecting language. Intermediate records let the team test one layer at a time and rerun only the affected stage when instructions or evidence change.

Use a paragraph-to-source map

The map is the control centre for the working draft. It should remain usable even if the prose changes. Each row identifies one candidate proposition, its proposed source, the passage or page to check, the proposed knowledge basis, related exhibit, uncertainty, and decision. Do not collapse several events into one paragraph if they rely on different sources or require different review.

Worked example: a source-linked event chronology

Assume counsel is preparing a working draft for a motion. The approved package contains an agreement, an email thread, a delivery record, internal business records, interview notes, and a later correction from the proposed deponent. Counsel has already identified the governing materials and issues. The system's assignment is limited to organizing the record and preparing candidates for review.

1. Inventory before ordering events

Record each file and version, connect the email attachments, and flag that one referenced attachment is absent. Keep the later correction beside the earlier interview note. The chronology must not silently use the earliest account as final or treat the missing attachment as if it supports the event.

2. Create event rows, not conclusions

For each date, record the observable event, participants, source locator, competing source, and open question. If a delivery record shows a timestamp but not what was inside the package, state only what the record supports. Do not convert temporal sequence into causation or intent.

3. Route candidates through the paragraph map

The output is a reviewable draft plus an exception list: the missing attachment, any disputed chronology point, proposed exhibit questions, and paragraphs whose knowledge basis is not confirmed. That is more useful than a seamless narrative that conceals the record's limits.

Make deponent review substantive

The deponent should be able to see where material language came from and identify what is wrong, incomplete, or outside their knowledge. Review in manageable sections and keep the source passages available. Ask the deponent to verify names, dates, roles, sequence, quotations, record descriptions, and the basis for each material point. A stylistic approval of fluent text is not the same as adopting its facts.

Preserve the deponent's own language where it matters. AI can normalize tone too aggressively, merge separate memories, or introduce transitions that imply certainty or causation. Review every material connective phrase, attribution, absolute statement, and summary. Record corrections as source updates so later drafting passes do not restore rejected language.

Control exhibits, execution, and commissioning separately

Maintain an exhibit register beside the draft. It should identify the proposed document, stable file and version, date, source, related paragraph, completeness check, confidentiality or redaction question, and status. Confirm that every in-text reference matches the correct document and that every proposed exhibit is included once in the intended order. Never let generated exhibit labels or numbering become authoritative without manual reconciliation.

The Rules of Civil Procedure contain specific provisions concerning affidavit contents and exhibits, and different proceedings or courts may add requirements. The Commissioners for Taking Affidavits Act is part of the current Ontario framework for taking affidavits. Counsel should check the current official text and the actual commissioning method before execution. AI should not be used to infer that a package has been validly sworn, affirmed, commissioned, filed, served, or accepted.

Quality control must test omissions and drift

Reconcile the processed file and page counts with the inventory. Sample paragraphs the system drafted and source passages it did not use. Test difficult material separately: scans, tables, marginal notes, handwriting, long email threads, attachments, transcript formatting, photographs, spreadsheets, and duplicate versions. Track wrong dates, merged speakers, incomplete quotations, lost qualifications, unsupported pronouns, and chronology changes.

Run a reverse check from source to draft and from draft to source. The first asks whether important approved material was omitted. The second asks whether every material sentence is supported. Then compare the final lawyer-reviewed version with the paragraph map and exhibit register. A late wording edit can break an earlier source link.

The Law Society of Ontario's generative-AI professional-obligations resource tells licensees to understand a tool's capabilities, limitations, standards, and terms of use, independently verify information intended to be relied on through a human process, and protect confidential information. Apply those checks to the actual provider, configuration, matter, users, retention, and proposed source set. Also consult the current Rules of Professional Conduct.

A verified Curia workflow

Curia's current public product pages describe matter-grounded work across documents, research, and drafting. The drafting page describes first-pass output based on matter evidence and researched authorities, with source support visible during review. For an affidavit workflow, that can support keeping approved matter documents and drafting context connected. It does not establish that Curia determines admissibility, the deponent's knowledge, procedural compliance, valid commissioning, filing readiness, or the evidence a lawyer should present.

Common affidavit-drafting errors

• Starting with “draft an affidavit.” A broad request hides the purpose, sources, governing requirements, deponent, and review standard.
• Treating the folder as complete. Missing attachments, later corrections, superseded versions, and unreadable pages must remain visible.
• Blending knowledge bases. A fluent paragraph can merge memory, records, another person's statement, and inference.
• Turning sequence into causation. Events occurring in order do not by themselves establish motive, knowledge, or legal significance.
• Flattening conflicts. Inconsistent documents or accounts should be routed for review, not harmonized automatically.
• Using detached quotations. A short excerpt can omit conditions, replies, attachments, or surrounding questions.
• Automating exhibit labels. Generated numbering can drift from the reviewed package and applicable process.
• Polishing away the deponent's voice. Smooth transitions can introduce facts or certainty the person did not adopt.
• Checking only drafted paragraphs. Quality control must also test relevant approved material the system omitted.
• Overstating the product. Matter context and source-linked drafting do not establish legal or procedural readiness.

Practical checklist

• Write the proceeding, purpose, deponent, issues, current due date, governing materials, reviewer, and stopping rule.
• Confirm the approved workspace, access, data use, retention, client restrictions, and minimum necessary source set.
• Inventory every file and page; preserve versions, families, exclusions, missing items, and unreadable exceptions.
• Extract observable facts before drafting prose, and keep conflicts and uncertainty visible.
• Create a paragraph-to-source map with locators, proposed knowledge basis, exhibits, gaps, and decisions.
• Keep fact, attribution, inference, opinion, argument, and legal conclusion separately reviewable.
• Have the deponent verify material facts and knowledge; have counsel decide procedure, evidence, privilege, scope, and wording.
• Reconcile quotations, dates, names, chronology, pronouns, exhibit references, and document versions.
• Test both positive findings and omissions, with separate checks for difficult formats.
• Follow the current execution, commissioning, filing, and service process selected for the actual matter.

Key takeaways and primary sources

• AI affidavit drafting is strongest as a staged organization, mapping, drafting, and verification process.
• The current governing materials, purpose, deponent, and approved source set control the work.
• Every material paragraph should remain traceable to a reviewed source or confirmed instruction.
• The deponent and responsible lawyer retain control over facts, knowledge, evidence, exhibits, procedure, and final use.
• Quality control must test missing material, contradictory sources, difficult formats, unsupported prose, and late version drift.

Primary and authoritative sources

1. Ontario, Rules of Civil Procedure, R.R.O. 1990, Reg. 194.
2. Ontario, Commissioners for Taking Affidavits Act, R.S.O. 1990, c. C.17.
3. Law Society of Ontario, Generative AI — Your Professional Obligations.
4. Law Society of Ontario, Rules of Professional Conduct.

Related Curia resources

• AI motion materials for Ontario litigators
• AI legal drafting for Ontario lawyers
• AI document review for Ontario litigation
• What lawyers can safely upload to AI tools