As more small businesses turn to general purpose AI tools to increase productivity and save time, accounting professionals warn that using ChatGPT and other public AI tools for tax advice should not replace professional judgment.
AI-powered bookkeeping software, Dext, commissioned a recent study of 500 Canadian accountants and bookkeepers and 76 per cent reported that their clients are using public AI tools, such as ChatGPT and other general purpose large language models, for tax advice. But that surge has come with a cost — 50 per cent of businesses have suffered direct financial losses due to incorrect AI-generated advice.
When using general purpose AI tools, Dext found the most common mistakes affecting business owners are misinterpreting business expenses, making incorrect tax claims or charges, and flawed personal tax planning.
Three accountants share their stories from this tax season — and how small businesses can avoid costly AI tax errors.
Tax Advice Based on Wrong Jurisdiction
“During tax season, we had a business owner come in convinced that a certain tax planning technique would work for them. They had done some research using an AI tool and came in fairly confident in the answer. When we asked where the idea came from, they showed us the prompt and response from the AI model. The issue was that the guidance clearly related to U.S. tax rules, not Canadian tax law. The client understandably didn’t realize the distinction, but the strategy would not have applied to a Canadian resident and could have created filing issues if implemented. It was a good example of how AI tools can generate convincing explanations, but tax advice is extremely jurisdiction-specific, and the models don’t always recognize the difference unless the user is very precise.” — Darren Lewis, Partner and CPA at Douglas Spence & Associates based in Oakville, ON
AI Suggests Tax Strategies, But Misses Nuances
“While AI is a powerful tool for generating tax insights, it often generalizes or provides incomplete answers. Tax planning and interpretation require nuance and a full understanding of the client’s situation—something AI doesn’t always capture. For example, AI may suggest that incorporating is beneficial due to lower corporate tax rates. However, without considering key factors—such as how corporate and personal tax integration works, how and when cash will be extracted, the nature of the business activities, and how the owner will be compensated—the recommendation can vary widely. In practice, the right answer could range from not incorporating at all to implementing a multi-entity structure.
As users refine their prompts, AI responses may become more accurate, but often still lack full clarity. This can create a gap in expectations between business owners and their accountants. Ultimately, AI outputs do not necessarily translate to fully replacing professional judgment—effective tax interpretation depends on analyzing specific facts, understanding context, and applying experience to real-world situations.” — Tom Son, VP Operations at ConnectCPA based in Toronto, ON
Tax Advice That’s Technically Accurate but Not Optimal
“We onboarded a client doing $1 million in revenue, newly incorporated, where the owner previously ran personal items through the business bank account. He used an AI tool to categorize what he stated were his business revenues and expenses based on bank statements, which it did, but it neglected to question whether any of the expenses were personal. It would have only known this based on a relationship with the client. It also assumes business owners know what is meant by ‘business revenues and expenses’, when often a discussion is warranted to educate, understand and then recommend. None of the expenses looked unusual in the data (payments to a leasing company, property management company, etc.). The AI categorized them exactly the way anyone would if all you had was the bank feed, and it underestimated the tax liability by thousands of dollars. The only way to know some of these expenses were personal is to know the client, and ask them the questions.
That context comes from the relationship, not the data. An AI tool processing bank transactions based on specific prompts will do what it is told. Our value-add as experts is in the judgement we exercise to ask the right questions and provide personalized advice that keeps our clients out of trouble.” — Moez Bawania, Partner and CPA at AMLB LLP, Toronto ON
Dext Launches AI Assist in Canada
Dext launched Dext AI Assist in Canada, a specialized AI agent for bookkeepers and accountants. A core principle behind AI Assist is ensuring automation remains transparent, controlled, and reviewable, ensuring accountants retain full control over how automation is applied. The result is less time spent manually applying judgment across repetitive tasks and more time delivering insight.
Source: Dext
