AI Issue Spotting for Ontario Litigators: 2026 Guide

AI can help Ontario litigators spot candidate issues by organizing pleadings, productions, transcripts, chronologies, correspondence, and research notes into a reviewable map. It cannot decide which issues govern the case, resolve conflicting evidence, assess credibility, or replace counsel's legal and strategic judgment. A reliable workflow starts with a defined question and controlled record, separates extracted facts from inferences and legal research, shows supporting and contrary material, ties every material point to its source, and records the lawyer's decision before the map is used in discovery, motion work, settlement, or trial preparation.

Key facts and limits at a glance

Define the issue question before reviewing the record

“Spot every issue” is usually too broad to produce a dependable result. It leaves the system to infer the forum, causes of action, defences, procedural stage, material dates, burden of proof, and purpose of the review. Those are legal and strategic choices. Start instead with a short review charter prepared or approved by counsel. Identify the proceeding, the current pleadings or live questions, the relevant date range, the materials included, and the decision the map is meant to support.

The question should also state what the exercise will not decide. A map for discovery planning may organize allegations, denials, supporting records, contrary records, admissions, and open questions. It should not silently decide whether a witness is truthful or whether a legal test is met. A map for a motion may catalogue facts relevant to counsel's approved elements and identify where evidence is disputed. It should not invent the legal test or predict an outcome.

A useful review charter

• Purpose: the concrete work product or decision the map will support.
• Issue frame: counsel-approved issues, elements, allegations, defences, or factual questions.
• Population: the authoritative collection, date cut-off, exclusions, and known processing limits.
• Source convention: the file, page, paragraph, transcript line, exhibit, or stable locator required for each point.
• Output fields: facts, contrary material, gaps, uncertainty, research questions, and review decisions.
• Escalation: what the system must leave unresolved for counsel.

Control the record before asking AI to map it

An issue map cannot be more complete than the collection it received. Build an inventory of the materials included and separate distinct source types: pleadings, affidavits, productions, transcripts, correspondence, expert material, orders, client instructions, legal research, and lawyer work product. Keep assumptions and instructions out of the fact lane. If a document is missing attachments, password-protected, image-only, illegible, duplicated, or outside the date range, record that limitation rather than allowing silence to look like absence.

Use stable identifiers. A proposition linked only to “the contract” or “the examination” will be difficult to verify later. Prefer a controlled document name and the best available page, paragraph, line, exhibit, or event reference. If OCR is used, retain access to the original image and mark low-confidence text. Reconcile the processed set against the inventory so that files that failed extraction are visible to the reviewer.

Confidentiality review comes before upload. The appropriate process depends on the tool, matter, client obligations, firm policy, contractual terms, access controls, retention, and the minimum information actually needed. The LSO guide tells licensees to review the tool's terms, understand the risks and how it uses inputs, avoid entering protected information where safeguards are inadequate, and carefully consider and redact the information provided. It does not turn one vendor configuration into a universal answer for every file.

A six-step AI issue-spotting workflow

1. Freeze the review set and instructions. Record which files and versions are in scope, which issue frame counsel approved, and what output is required. Preserve a run note so a later reviewer can reproduce the boundaries of the exercise.
2. Extract before interpreting. Ask first for people, dates, events, quoted or paraphrased passages, documents, and locators. Do not combine this pass with a request for legal conclusions. Extraction errors are easier to see when they are not wrapped in analysis.
3. Group facts under counsel-defined issues. For each issue, collect material that appears supportive, contrary, neutral, or incomplete. Preserve facts that do not fit the initial theory in an “unmapped” lane rather than discarding them.
4. Challenge the first map. Ask what evidence cuts against each proposed point, what context is missing, which terms are ambiguous, where dates conflict, and which propositions depend on assumptions. This is a search for review work, not an automated credibility assessment.
5. Verify in the sources and law. Open each material source. Separately research the governing law using current primary authority. Confirm that the issue labels match counsel's analysis rather than a model's generic taxonomy.
6. Record lawyer disposition. Mark each point accepted, revised, rejected, unresolved, or sent for follow-up. Date the map, identify the reviewer, and preserve the source links and open questions when the approved content moves into later work product.

Build a map that keeps fact, law, and judgment separate

A good issue map is not a long narrative summary. It is a control surface that lets a reviewer move from an issue to the supporting and contrary record, see uncertainty, and record a decision. The exact fields will vary by matter, but separation matters. A fact extracted from a transcript, a proposition drawn from a case, a client instruction, and counsel's strategic inference should not occupy one undifferentiated cell.

Worked example: testing a notice issue without deciding it

Assume counsel is assessing whether the record contains evidence relevant to when a party received notice of a project delay. The legal significance of notice depends on the governing law, pleadings, contract, and facts; the AI task should not decide it. Counsel defines the factual review question: identify communications that mention the delay, when they were sent or received, who participated, what attachments were included, and whether later evidence conflicts with the contemporaneous record.

The controlled set contains a contract version selected by counsel, project correspondence, meeting minutes, a chronology, and two transcripts. The extraction pass lists candidate events with document and page references. The grouping pass places a dated email and meeting entry in the supporting lane, a missing attachment and ambiguous distribution list in the gap lane, and a later transcript answer in the contrary-or-context lane. Nothing is labelled “effective notice.”

Counsel then opens the email, confirms the timestamp and recipients, checks the surrounding thread, determines that the referenced attachment is absent, and reads the transcript passage in context. Separately, counsel researches the governing legal question and contract terms. The final map says what the sources establish, what remains uncertain, and what follow-up is needed. That is more useful—and safer—than asking a model for a binary legal answer from an uncontrolled upload.

Verify the map and challenge what looks settled

Review begins with the propositions most likely to affect the next decision, not with a quick scan of the formatting. For each material point, open the cited source and ask whether the wording, date, speaker, document version, and surrounding context match. Check whether the map converted an allegation into a fact, blended two witnesses, treated silence as proof, or described an inference as an event. Then test coverage: were contrary terms used, were all custodians or source lanes included, and did any file fail processing?

Legal verification is a separate pass. Open the statute, regulation, rule, practice direction, or decision rather than relying on a model's citation. Confirm jurisdiction, court, date, amendments, paragraph, context, and subsequent treatment as applicable. Keep legal propositions linked to their authorities and factual propositions linked to the record. A source reference can still be wrong; the point of the reference is to make checking possible.

Keep Ontario court work inside the current filing boundary

If an issue map informs material for the Ontario Superior Court of Justice, review the currentConsolidated Civil Provincial Practice Directionand the rules and directions governing the proceeding. Part J.12 of the direction addresses AI use in court submissions and emphasizes accuracy, authoritative-source verification, oversight, and the fact-dependent response to misuse. The current Rules of Civil Procedurealso govern filed documents, including Rule 4.06.1 where it applies. An internal map does not excuse source checking or change counsel's responsibility for the material ultimately used.

A verified Curia workflow for issue mapping

Curia's public matter-workspace page describes organizing documents, timelines, research, and drafting around a matter. Its research pagesays surfaced authorities link to source passages, while the drafting pagedescribes drafting grounded in matter context and version history. Those verified capabilities support a practical sequence: organize the controlled record in the matter, build a source-linked factual map, conduct and verify legal research separately, then carry only lawyer-approved points into a draft with a visible review trail.

Common errors and exceptions

• Starting with a universal prompt: the model invents a generic issue taxonomy instead of following the pleaded and live questions.
• Mixing source types: allegations, evidence, instructions, assumptions, and legal propositions become indistinguishable.
• Seeking only support: the map ignores contrary evidence, qualifications, and alternative explanations.
• Using weak locators: reviewers cannot reproduce a point from “the email chain” or “the transcript.”
• Ignoring failed files: unreadable scans or missing attachments disappear from the apparent record.
• Automating credibility: an apparent inconsistency is treated as a finding about truthfulness.
• Letting labels harden: tentative AI categories become headings in advice or court material without legal review.
• Reusing a stale map: pleadings, productions, orders, or authorities change while the working map does not.

Some matters also require a different design. A small, clean record may not justify an AI pass. A multilingual, handwritten, image-heavy, audio, or technically complex collection may require specialized processing and validation. Privilege review, sensitive personal information, protective orders, client terms, and cross-border data concerns may narrow or rule out a proposed tool. Record why the chosen process is appropriate instead of assuming that more automation is always better.

Practical issue-spotting checklist

• Define the proceeding, purpose, live issue frame, and decision the map will support.
• Inventory the authoritative collection, versions, date cut-off, exclusions, and processing failures.
• Review confidentiality, privilege, security, tool terms, firm policy, and client constraints before upload.
• Set a stable source-reference convention before extraction begins.
• Extract facts and events separately from inferences and legal propositions.
• Map supporting, contrary, neutral, and unmapped material.
• Keep legal research in a separate authority lane and verify primary sources.
• Open every source behind a material finding and read it in context.
• Reconcile processed files against the collection and investigate gaps.
• Record counsel's accepted, revised, rejected, or unresolved disposition.
• Recheck current court rules and practice directions before downstream filing work.
• Date and version the approved map, then update it when the record or issues change.

Key takeaways

• Use AI to organize candidate issues and evidence, not to decide the case.
• Define the legal and factual frame before processing the record.
• Preserve supporting and contrary material, gaps, uncertainty, and stable source links.
• Separate fact extraction, legal research, inference, and lawyer judgment.
• Verify the original record and current primary authority before relying on any material point.
• Carry only lawyer-approved content into discovery, advice, negotiation, or court work.

Primary sources and authoritative guidance

• Law Society of Ontario, Generative AI: Your Professional Obligations.
• Law Society of Ontario, Rules of Professional Conduct.
• Ontario Superior Court of Justice, Consolidated Civil Provincial Practice Direction.
• Ontario, Rules of Civil Procedure, R.R.O. 1990, Reg. 194.

Related Curia resources

Continue with the Ontario AI document-review guide, litigation-timeline workflow, discovery-question workflow, and legal-memo drafting guide.