No — AI drafts the documents around an audit, not the opinion inside the auditor’s report. The opinion is a conclusion about evidence, and a licensed practitioner forms and signs it.
Key takeaways
Two different documents get called the audit report. One is the auditor’s report itself: a short, highly prescribed document in which the practitioner expresses an opinion on the financial statements. The other is everything around it — the engagement summary, the memo to the file, the disclosure checklist, the management letter, the report to those charged with governance.
AI has almost nothing to offer on the first and quite a lot to offer on the second. Confusing them is how firms end up with an efficiency project that stalls at the review stage.
The auditor’s report is the visible end of a body of evidence gathered under the Canadian Auditing Standards. Those standards are set here: the Auditing and Assurance Standards Board establishes standards for assurance and related services engagements and adopts the International Standards on Auditing as Canadian Auditing Standards. The opinion is a conclusion about whether sufficient appropriate evidence supports the statements. A model has not gathered that evidence; it has read your file about it. Those are different things, and only one of them can be signed.
Disclosure checking. Comparing a set of financial statements against a disclosure checklist is comparison work with a defined reference. Caseware describes exactly this in its Canadian AI documentation: a Disclosure Checklist Agent that automates disclosure checklist review against financial statements and generates citation-backed suggestions auditors can review, refine, accept or override within the engagement workflow. Note the four verbs at the end — the vendor’s own design assumes an auditor decides.
The management letter. Control observations collected during fieldwork have to be turned into readable, non-accusatory prose with a recommendation attached. Drafting that from your own notes is a writing task.
The memo to file. A first draft that pulls the significant judgments, the conclusions reached and the workpaper references into one narrative, for the engagement partner to correct.
Prior-year comparison. Diffing this year’s draft statements and note disclosures against last year’s, and listing what changed, is exactly the kind of mechanical reading humans do badly at 9pm.
The Canadian Public Accountability Board has made this an inspection focus. Its published position is that firms integrating advanced tools, including agentic artificial intelligence, gain opportunities to improve audit quality and consistency but also introduce new risks, and that its inspections include evaluating the system of quality management practices associated with obtaining or developing, approving, implementing and maintaining automated tools and techniques at both firm and engagement level. CPAB has published guidance on this — The use of artificial intelligence in the audit — balancing innovation and risk and, more recently, The evolving use of technology in auditing — and it leads the International Forum of Independent Audit Regulators’ Technology Task Force.
The practical consequence for a mid-sized Canadian firm: the decision to adopt a drafting tool is a quality management decision with documentation attached, not a software purchase.
A firm with fourteen assurance staff picks one narrow use: the management letter. Control observations are already captured in the engagement file in a standard note format. The tool drafts the letter from those notes into the firm’s template — observation, implication, recommendation, management response placeholder.
The manager reviews it against the notes, not against a memory of the fieldwork. The partner reviews as before. Nothing about the opinion, the evidence or the file changes.
What the firm records: which tool, what data it saw, who approved its use, and the review step that sits between the draft and the client. That record is the point. It is what turns "we tried AI" into something a regulator can look at.
Fluent wrongness. Drafting tools produce confident prose regardless of whether the underlying evidence supports it. A reviewer reading for grammar will pass a paragraph that a reviewer reading against the workpaper would stop.
Invented references. A model asked to cite a standard may produce a plausible paragraph number that does not exist. Any citation in a draft is checked against the standard as issued by the AASB before it survives review, which is why citation-backed suggestions traceable to a source are worth more than free-form drafting.
So can AI write an audit report?
It can produce a draft of the surrounding documents. It cannot form or express the opinion, because the opinion is a professional conclusion about evidence and a licensed practitioner signs it.
Does using a drafting tool have to be disclosed to the client?
The audited entity is not entitled to dictate your methodology, but data handling is a contractual matter and worth settling in the engagement letter rather than discovering later.
Where should a small firm start?
With one document type that already has structured inputs in the file — the management letter or the file memo. Not with anything that touches the opinion.
A 30-minute call is enough to tell you whether AI pays for itself here.
