Canadian accountants who have been hoping that artificial intelligence will ease their tax season burden are finding it is adding to their workloads instead. 

A recent survey of Canadian accountants by bookkeeping software company Dext found that 44 per cent of those who encountered AI-generated mistakes now spend up to three hours a month correcting errors caused by clients misusing advice they’ve received from AI models. Another 38 per cent reported spending even longer, between four to 10 hours per month. Only seven per cent of those surveyed said they had not encountered a mistake generated by a public-facing AI tool.

Most of the AI-generated errors involved issues like incorrect interpretations of business expenses and payroll errors, the survey found, and about 70 per cent of respondents said clients are using AI-generated outputs to challenge professional advice. Dext and Censuswide surveyed 500 Canadian accountants and bookkeepers in December.

The statistics resonate with B.C.-based financial advisor Bharathi Sandhu, a senior business development specialist at Raymond James’s Nava Wealth. While Sandhu freely chats with ChatGPT or Grok throughout the day on topics like what to feed a Pomeranian, at work she finds that chatbots frequently make mistakes about Canadian financial rules around like capital gains taxes, RRSP withdrawals, dividends and more. 

“I really think that it oversimplifies—and then the worst part is that it gives the user the impression that they’ve gotten a comprehensive view,” she said.

Sandhu said it’s become so common that she can usually spot when a client has a misconception that comes from their use of AI. But she said she understands one impulse to turn to AI for financial advice: to avoid the feeling of being judged by a human. Plus, ChatGPT is free—at least as long as it doesn’t make a mistake that costs you money.

AI has exposed longstanding gaps in Canada’s tax system that make it difficult for those with modest incomes to find reliable tax advice, said Elizabeth Mulholland, CEO of anti-poverty charity Prosper Canada. The charity, which receives funding from the federal government, runs free tax advice clinics and has advocated that the government accelerate the rollout of the automatic tax filing services being piloted for low-income Canadians.

But many low-income Canadians do gig work or are self-employed with micro-businesses, complicating their tax filings, she said.

“We use the tax system to do a million things, and consequently it’s become rather unnavigable for the average person,” Mulholland said.

Paul Craig, an Ottawa-based web developer, sees a place for AI to help people understand the complicated tax code. He created TaxGPT, a free chatbot that relies primarily on information from the Canada Revenue Agency’s website. Because people need to file taxes to access government safety net programs, he thinks it’s better to have technologists creating more sources of tax information, particularly as the CRA has struggled to build an accurate chatbot of its own.

“It doesn’t solve all the problems,” Craig said of AI. “It’s a thing in the landscape in which there’s a lot of bad options. You can wait on hold for a long time with the CRA. You can pay a lot of money for good advice.”

(The CRA toldThe Logic that it updates canada.ca/taxes with the express purpose of making sure generative AI tools can “surface” reliable information.)

Benjamin Alarie, whose AI company Blue J works with accountants and the CRA, defended the technology, saying AI is less likely to make mistakes in the hands of professionals, who recognize the nuances of tax rules by country, province and tax year.

Still, many accountants remain apprehensive about AI, even as some big accounting firms hesitate to hire new staff in favour of AI investments. Much of the technology is a black box, with little transparency for users about how it works or how it arrives at the answers and advice it offers. That conflicts with the culture of a profession that’s taught to verify paper trails. One of the biggest accounting scandals, the Horizon incident, involved a faulty software program that prompted the U.K. postal service to falsely accuse branch leaders of fraud, with some postmasters being wrongly imprisoned, declaring bankruptcy and dying by suicide.

Accountant Philip Maguire sees value in AI for tasks like analyzing large amounts of data and freeing up junior accountants to focus on meaningful work. But, Maguire, principal at Glenidan Consultancy, said he’s also “concerned with the general feeling that AI is so powerful that it will do everything for you.”

Dext CEO Sabby Gill, who commissioned the survey of accountants about AI slop, said the results show that there needs to be more guardrails around AI and accuracy. 

“Provide some professional oversight. If you don’t, these consequences are going to be pretty severe,” said Gill. “Businesses suffered those financial losses from getting that incorrect AI advice, because there was no guardrail.”

Correction: This story has been updated to clarify that Prosper Canada receives funding from the federal government.