AI-Generated Transcripts in Canadian Litigation: Authenticity and Evidentiary Weight

As artificial intelligence (“AI”) becomes increasingly integrated into legal practice, lawyers must assess not only its utility in their own work but also the evidentiary significance of material generated through its use. AI generated transcripts/records of discussions illustrate this challenge. When offered to prove a matter in issue, they raise three distinct questions:

1. whether the document is authentic,
2. whether it accurately reproduces the conversation, and
3. whether that conversation establishes the legal point in issue.

A faithfully preserved software output may contain transcription errors; an accurate transcript may disclose only unfinished negotiations. Ontario authorities provide a starting point for addressing these inquiries, although the law in this area remains underdeveloped and by no means provides definitive answers to these questions.

For the first inquiry, Ontario’s Evidence Act imposes distinct requirements. Under s. 34.1(4), the party tendering an electronic transcript must produce evidence capable of supporting a finding that it is what the party claims it to be.¹ Where the best-evidence rule applies, s. 34.1(5), subject to the printout provision in s. 34.1(6), requires proof of the integrity of the electronic records; s. 34.1(5.1) permits proof through the integrity of the recording or storage system, or reliable encryption techniques.²

If relying on the business-record exception, the proponent must establish that the record was made in the usual and ordinary course of business and that making such records contemporaneously, or within a reasonable time, was itself in the usual and ordinary course of business.³ The proponent must also give all other parties at least seven days’ notice, and those parties are entitled to production for inspection within five days after giving notice to produce.⁴

In R. v. C.B., 2019 ONCA 380, the Court of Appeal for Ontario (“ONCA”) described authentication as requiring “some evidence that the item is what it purports to be”.⁵ Since establishing that a document is the unmodified output of a transcription system differs from establishing that it faithfully reproduces the participants’ words, the proponent’s characterization of the evidence is therefore important. The ONCA also distinguished the admissibility of the evidence from ultimate evaluation of its weight. The ONCA stated:

Satisfaction of the evidentiary threshold for authentication under s. 31.1 of the CEA or at common law renders the evidence admissible; in other words, available to the trier of fact for ultimate evaluation. It does not follow from admissibility that the trier of fact must find that the evidence is in fact what it claims to be. What remains of the dispute after admissibility has been [page16] established relates to the weight to be assigned to the evidence. And that issue is left to the trier of fact to decide.⁶

Although this case concerned electronic communications in criminal proceedings, its analysis helps identify the limited proposition established by authentication. Admission does not compel acceptance of the transcript’s account.

However, transcription is not necessarily the same as interpretation. Simply because the transcript is authentic does not mean it accurately reproduces the conversation. The boundary between transcription and interpretation is addressed in HSCAS v. M.S., C.D.S., and C.A.D., 2026 ONSC 3038.⁷ The decision records counsel’s concern that the AI tool “had added its own commentary, such as opining that it sounded like the children were playing.”⁸ The court subsequently stated: “I did not admit any of these recordings into evidence for various reasons, orally delivered.”⁹ Although the court did not specify those various reasons in the written decision, its relevance is the possibility that material presented as reproduced speech contains additional interpretation. Accordingly, in civil litigation, an annotation that a participant accepted a proposal cannot substitute for identifying what that participant communicated. An omitted qualification or incorrect speaker attribution may likewise alter an exchange’s apparent legal significance.

Even if an AI generated transcript/record accurately records a conversation, the court must still determine whether the contents of the transcript establish the legal point in issue. An example where parties may wish to rely on transcripts of discussions is when trying to prove the formation of a binding contract. In Ruparell v. J.H. Cochrane Investments Inc., 2021 ONCA 880, the ONCA stated “it is not unusual for contracts to be made by agreement on the essential terms, which are later incorporated into a formal written document. Whether the parties reached a binding contract depends on the circumstances of the case, and in particular on the intention of the parties.”¹⁰ Accordingly, where the exchange concerns contractual formation or execution of formal documents, an accurately recorded qualification may distinguish a concluded bargain from a conditional proposal. For example, a statement that no agreement would become binding until the approval of the Office of the Children’s Lawyer could be decisive. If the transcribed negotiation omits that qualification, the court must still determine the circumstances of the case and the intention of the parties.

The parties’ actions after receiving the transcript/record may also impact the analysis. In Owners, Strata Plan LMS 3905 v. Crystal Square Parking Corp., 2020 SCC 29, the Supreme Court of Canada held that:

In sum, an “outward manifestation of assent by each party such as to induce a reasonable expectation in the other” is required in order to find that a binding post-incorporation contract exists: Waddams, at §25. The test is objective. It requires an examination of how each party’s conduct would appear to a reasonable person in the position of the other party: P. Benson, Justice in Transactions: A Theory of Contract Law (2019), at pp. 112-13. Thus, a court should determine whether a reasonable person in the position of one party would consider that the other party’s conduct constituted an offer: Grant v. Province of New Brunswick (1973), 1973 CanLII 1765 (NB CA), 6 N.B.R. (2d) 95 (S.C. (App. Div.)), at para. 12. And, conversely, whether a reasonable person in the position of the latter would consider that the former’s conduct constituted an acceptance: Saint‑John Tug Boat, at pp. 621-22. The pre-incorporation contract is merely one aspect of the objective circumstances that can be used to interpret the parties’ conduct and from which the terms of a post-incorporation contract may be inferred.¹¹ [Emphasis added]

Applied by analogy, a response confirming the transcript’s accuracy may help establish what occurred during the meeting. A response expressly agreeing to the recorded terms and authorizing performance may also support a finding that an agreement arose afterward—even if no agreement was reached during the meeting itself. These responses serve different evidentiary purposes, although their significance may overlap. Merely receiving the transcript does not automatically establish either its accuracy or acceptance of its terms; the court would likely examine the parties’ communications and conduct in context.

In conclusion, AI generated transcripts/records tend to require the court to distinguish authenticity, accuracy, and contractual effect. Establishing that a transcript is genuine does not necessarily establish that it faithfully reproduces the conversation, and an accurate record does not itself establish a legal point in issue. The underlying recording, any revisions, and the parties’ subsequent communications and conduct may each inform these distinct inquiries.

Although existing contract and evidence principles provide a starting point, their application to AI generated transcripts/records remains underdeveloped. Future judicial decisions will need to clarify how transcription errors, automated interpretation, and subsequent endorsement affect admissibility, evidentiary weight, and proof of a legal point in issue.