Should you let AI do your taxes?

OPINION:

When artificial intelligence evangelists posted virally on X recently that Americans could use AI to “help with your taxes,” they tapped into a real hope.

Tax season is stressful, the rules are complicated, and the forms are tedious. If AI can write emails and summarize documents, then why not let it handle a return? Here’s why: Because a tax return is a legal filing built on factual inputs and judgment calls, and AI doesn’t have the precision needed to get the job done.

A recent New York Times test ran eight fictional tax situations through four major chatbots. Despite being given the necessary forms, the systems reportedly miscalculated refunds or tax owed by more than $2,000 on average. The mistakes also could trigger amended returns, IRS correspondence and cleanup costs.

A return turns on dozens of details such as filing status, dependency, timing, basis, phaseouts and credits. A small assumption in one line can ripple through the rest of the filing. In tax preparation, “close enough” isn’t good enough. It’s considered wrong and can lead to audits and penalties.

The deeper issue is that AI mistakes are often hard for nonexperts to spot. A completed return can look organized, fluent and confident while being incorrect in ways a lay reader is unlikely to notice.

These are rarely simple arithmetic errors. They are legal or factual mistakes hidden inside the numbers, such as assuming someone qualifies as a dependent, misreading rental use or treating foreign income too casually.

That risk is compounded by AI hallucinations, where systems generate made-up material while presenting it as real.

The same prompt can yield different answers across runs. This imprecision may be harmless when using AI to brainstorm gift ideas, but it is far from harmless when dealing with a legal filing submitted to the government.

Advertisement Advertisement

Tax law also changes constantly. Congress revises rules, agencies issue new guidance and thresholds move. An AI system using stale knowledge can sound every bit as confident as one using current law.

As Investopedia notes, AI is especially likely to struggle with business ownership, trusts and international tax issues.

Then there is the matter of privacy, which is far from a side issue. Tax returns contain Social Security numbers, bank details, income records and sometimes family and/or medical information. Once users paste that material into a consumer chatbot, they lose control over how it is stored or reused.

None of this means AI has no place at tax time. Used well, it can make people better prepared for a conversation with a human adviser. Make sure you set the context for the chatbot by broadly describing your tax situation before asking questions.

Safe and productive uses for AI include:

Advertisement Advertisement

Jargon translation: Turn tax terms and notices into ordinary English. Ask AI, “What deductions or credits should I investigate?”

Document organization: Build a checklist, summarize what changed from last year and organize the facts a preparer will need. Ask it, “What changed from last year that I should double-check?”

Question generation: Draft sharper questions for an accountant or enrolled agent, especially about dependency, home office use or foreign income. Ask it, “What questions should I ask my accountant?”

Even the companies building these tools frame AI as support, not sign-off. OpenAI’s usage policies distinguish higher-risk uses that require care. When Intuit announced its latest AI offerings, it paired automation with expert help.

Advertisement Advertisement

AI can make tax season less painful. It can organize records, translate jargon and surface better questions.

Still, a tax return is unlike other paperwork because it is a legal filing with the taxpayer owning the filing risk. AI will eventually play a meaningful role in tax preparation, but only when its systems are built inside a controlled workflow designed for tax preparation.

For now, consumer AI is best used to prepare the conversation, not to prepare the return.

• Vidyanand (VC) Choudhary is a professor of information systems and the director of international programs at the University of California, Irvine, Merage School of Business. He is an authority on competitive strategy for technology products and AI.

Advertisement Advertisement