New chatbots providing US tax support were discovered to be unhelpful or wrong as much as half of the time, according to the Washington Post.

AI-powered chatbots added by TurboTax and H&R Block to their tax-preparation software are supposed to help people answer questions and better understand their returns. 

But Washington Post journalist Geoffrey A. Fowler and two tax experts received 'random, misleading or inaccurate AI answers' most of the time.

Intuit's chatbot provided inaccurate responses to over half of the 16 test questions it was asked, while H&R Block's chatbot confidently recommended an incorrect filing status and erroneously described IRS guidance on cryptocurrency. 

Intuit updated its Intuit Assist chatbot after feedback from Fowler; however, the subsequent version proved unhelpful on a quarter of questions.

The findings raised questions about the accuracy and reliability of the two systems.

 Users should be 'especially wary of generative AI when there are real-life consequences to it being wrong', including the possibility of tax audits, Fowler warned. 

'We can’t necessarily trust companies experimenting with AI to make the right decisions to protect our interests,' he added.