With about a month left until the April 15 deadline for filing taxes, you may be tempted to pawn off the task onto an AI chatbot. After all, companies keep saying they are more capable than ever of autonomously handling tasks like coding apps or, you know, bombing military targets. Surely, they can handle navigating the American tax code…right?
James Burnham, xAI’s General Counsel and head of law and government, sure thinks so. In a post on X, he told his followers, “Doublecheck your taxes with @Grok. A friend had Grok doublecheck TurboTax and it increased her tax refund by $1400. That covers nearly four years of Grok Premium!” (He later edited the post to add, “Disclaimer: This/Grok is not tax advice so always confirm yourself too.”)
Setting aside the fact that you can absolutely find better things to do with your money than getting nearly a half-decade’s worth of access to the MechaHitler machine, and the fact that Burnham isn’t even citing his own experience, just passing on the account of a friend, there’s a logic to assuming a chatbot can help you with taxes. Even tax giants like H&R Block have introduced AI assistants designed to tackle tax-related questions. But that doesn’t mean you should just pass your W2 or 1099s over to Grok or your chatbot of choice and let them fill out the forms for you.
In fact, tax experts would really rather you not do this—after all, they’re probably the ones who are going to have to help you roll back all the mistakes. “I haven’t personally used Grok to prepare or review tax returns, and I wouldn’t advise taxpayers to [try it], especially using a general-purpose chatbot as a tax reviewer,” Joel Salas, owner of Elevated Tax Strategies, told Gizmodo. “Practically speaking, that’s just not a good idea.”
There are several reasons not to trust the Groks of the world to roleplay as your Certified Public Accountant. The first is accuracy. The New York Times put multiple chatbots through the paces of filing taxes, handing them tax situations prepared as training materials by tax preparation company TaxSlayer. It found that the chatbots miscalculated refunds and amounts owed to the IRS by an average of more than $2,000. So sure, Grok might have saved Burnham’s friend $1,400 on paper, but who knows if it arrived at that number because it found some hidden exemptions or because it just made stuff up and entered numbers incorrectly.
Those findings are backed up by other tests, too. TaxCalcBench, a benchmarking test designed to evaluate an AI model’s ability to calculate tax returns, found that most options are wildly insufficient, with most failing to even crack 50% accuracy across a full return.
“Tax preparation in particular requires purpose-built systems designed for accuracy, compliance, reliability, and security,” a spokesperson for tax and finance software giant Intuit told Gizmodo. “Consumers are looking for confidence and peace of mind in moments that carry real financial liability. While a general-purpose LLM may provide broad tax information, it is not specifically trained or validated to prepare accurate tax returns across complex federal and state scenarios.”
The second reason not to use a bot to do your taxes, Salas said, is that it’s just irrational to trust your average chatbot with the type of sensitive personal information that is contained in your tax documents. “Honestly, it hasn’t been around long enough for us to trust these companies with that type of data. If you review any of their terms and conditions and fully understand them, you’ll realize you’re playing a bit of Russian roulette with your data,” he said. “Your data can be used for other purposes if you do not opt out or take certain actions.”
Last year, researchers at Stanford analyzed privacy policies of major AI chatbots and found that many of the leading US companies feed user inputs back into their models to improve their capabilities. Companies, including Anthropic, OpenAI, and xAI, require users to manually opt out of allowing their conversations to be used for training purposes. Meta previously allowed users to browse other people’s prompts and conversations with its chatbot, revealing medical, legal, and other sensitive information. A similar situation happened with xAI, which temporarily made user conversations with Grok publicly visible and searchable.
It’s also no secret that xAI and its founder, Elon Musk, have been after taxpayer data for a while now. The Musk-led Department of Government Efficiency reportedly sought to access sensitive taxpayer data at the IRS last year. It’d certainly make their lives much easier if people just willingly handed that information over, which Musk encouraged them to do, stating on Twitter, “Grok can help with your taxes.”
If you do want to use AI to help with filing your taxes, there are tools out there that are better suited than Grok. Salas pointed to the tool Taxbox, which he said can be used as a “tax assistant or tutor” to better understand what different documents mean and how to navigate the filing system. He even said that you can use a chatbot to do things like build a checklist of common tax mistakes or ask some tax basics as you go through the filing process. Just don’t give them your actual documents.
“The risk arises when copying and pasting W-2s, 1099s, or draft returns directly into the internet bot and asking it to check them,” Salas said. “The bot can not only be confidently wrong, but you are also relinquishing personal data.” When it comes to actually filing, just stick to the software designed specifically for the task. “The actual form logic, calculation, and e-file validation are built into the tax software. They’re not built into the chatbot,” Salas said. And if you end up getting audited, your chatbot isn’t going to be able to help you.
Intuit, which obviously has its own motivations for driving consumers to its services, offered similar advice. “General large language models can be helpful tools for general education and guidance, including answering high-level financial or tax questions,” a spokesperson said. But, they specified, “AI can absolutely play a valuable role in financial decision-making, but when it comes to high-stakes matters like taxes, consumers should look for solutions that are purpose-built, secure, and designed to deliver accuracy at scale.”
And if you don’t believe the experts, you can always ask Grok. When asked on Twitter about its ability to file taxes, the chatbot told users, “I’m not a licensed tax pro or official software, so no—don’t ‘do’ your taxes solely with me.” You heard the bot.
