WSJ Tech News Briefing: Chatbot Confidential – Using AI for Tax Preparation

Episode Release Date: March 30, 2025

1. Introduction to "Chatbot Confidential" Series

In this special series of the Wall Street Journal's Tech News Briefing, host Nicole Nguyen delves into the burgeoning role of AI-powered chatbots in everyday tasks. Kicking off with the episode titled "Chatbot Confidential: Using AI for Tax Preparation," Nguyen sets the stage for an in-depth exploration of how artificial intelligence is reshaping the way individuals approach tax season.

2. AI Chatbots Enter the Tax Preparation Arena

a. Rising Popularity and Usage Statistics

With the tax deadline looming in 15 days from the episode's release, Nguyen highlights the immense volume of tax filings in the United States—over 161 million individual income tax returns were filed in 2022, according to the Internal Revenue Service (IRS). Amid this annual rush, AI chatbots like OpenAI's ChatGPT, Anthropic's Claude, and Microsoft's Copilot are gaining traction as tools to assist filers.

"Generative AI powered chatbots to get help with everything from writing code to planning vacations," Nguyen explains (00:33).

A recent Harris poll cited in the episode reveals that 17% of tax filers have already utilized AI for tax preparation, with 45% considering it in the future.

b. Testing the AI Contenders

To assess the efficacy of these AI tools, Nguyen and Wall Street Journal’s tax columnist Laura Saunders conducted a comparative analysis of three prominent chatbots:

1. OpenAI's ChatGPT
2. Anthropic's Claude
3. Microsoft's Copilot

They posed the same query to each: “I am employed, work from home, married, have a mortgage, and a child. What kind of deductions do I qualify for? How can I reduce my tax bill?”

c. Evaluating the Responses

ChatGPT emerged as the most detailed, providing comprehensive information but limited to the 2023 tax code. Saunders notes that while ChatGPT covers benefits like the dependent care credit and flexible spending accounts, it fails to mention that these are typically mutually exclusive, potentially misleading users (04:10).

Claude offered less numerical detail but introduced inaccuracies, such as incorrectly suggesting eligibility for the home office deduction for employees working from home—a deduction not applicable under current IRS rules for employees (04:49).

Copilot was found to be the vaguest, lacking specific figures and providing minimal guidance. It only broadly stated eligibility criteria without delving into the nuances essential for accurate tax preparation (05:20).

Laura Saunders concludes, “ChatGPT was the best of the three. It had the most detail” (06:35), yet she cautions that the IRS website remains the most reliable resource.

3. Insights on AI Performance in Tax Preparation

Saunders emphasizes the complexity of the U.S. tax system as a significant barrier for AI to navigate accurately. She highlights that:

• Complexity of Tax Laws: The intricate nature of tax codes, compounded by individual life circumstances, makes it challenging for AI to provide fully accurate advice.

• Dynamic Legislative Changes: Tax regulations evolve, and AI models may lag in updating these changes, leading to outdated or incorrect information.

• Human Oversight: While AI can offer a foundational understanding, Saunders advocates for consulting official IRS publications or professional tax advisors to ensure accuracy.

4. Privacy Concerns with AI Chatbots

a. Interview with Privacy Expert Jennifer King

Transitioning from functionality to privacy, Nguyen interviews Jennifer King, Privacy and Data Policy Fellow at the Stanford Institute for Human Centered Artificial Intelligence. King sheds light on the ownership and handling of data input into AI chatbots.

"I would argue that you lose possession of it," King states (08:53). She explains that data submitted to chatbots typically becomes the property of the AI provider and may be used for training purposes.

b. Data Ownership and Privacy Policies

King critically examines privacy policies, noting that companies like OpenAI, Anthropic, and Microsoft have varying approaches:

• OpenAI’s ChatGPT: Offers a "temporary chat" feature, ensuring that conversations aren’t used to train models, thereby enhancing user privacy (10:18).

• Anthropic’s Claude: Markets itself as "privacy first," focusing on safeguarding user data.

• Microsoft’s Copilot: States that it doesn’t use customer data for training without explicit consent and doesn’t share data with third parties without permission (10:57).

King emphasizes the importance of regulatory oversight to ensure these companies adhere to their privacy commitments.

c. Tips to Protect Personal Data

Nguyen and King provide actionable advice for users:

• Use Temporary Chats: Utilize features that prevent chat history from being stored or used for AI training.

• Avoid Sensitive Information: Refrain from entering personally identifiable information (e.g., Social Security numbers) or uploading sensitive documents like tax returns.

• Verify Responses: Always cross-check AI-generated information with official sources or trusted professionals.

5. Conclusions and Recommendations

The episode wraps up with practical recommendations for taxpayers considering AI assistance:

• Start with AI, Confirm with Official Sources: AI chatbots can offer a starting point but should not be solely relied upon for critical financial decisions.

• Consult Professionals: Engaging with certified tax professionals or established tax software providers like TurboTax and H&R Block can provide more reliable guidance.

• Maintain Data Privacy: Users must be vigilant about the information they share with AI tools to protect their privacy.

Nguyen reinforces the age-old computing adage, “Garbage in, garbage out,” stressing the importance of inputting accurate and non-sensitive information into chatbots to minimize errors (07:28).

6. Teaser for Future Episodes

Looking ahead, Nguyen hints at upcoming discussions on the use of chatbots in workplace settings, addressing the associated risks and best practices. She invites listeners to engage by submitting questions about AI and privacy for future episodes (12:19).

Notable Quotes with Timestamps

• Nicole Nguyen (00:33): “Generative AI powered chatbots to get help with everything from writing code to planning vacations.”

• Laura Saunders (02:18): “The main issue is that you'll think that you know something that you don't really know if you use the bot.”

• Laura Saunders (03:42): “Taxes are complicated because life is complicated.”

• Laura Saunders (06:35): “ChatGPT was the best of the three. It had the most detail.”

• Jennifer King (08:53): “I would argue that you lose possession of it.”

• Nicole Nguyen (07:28): “Garbage in, garbage out. Isn't that one of the first rules of computing?”

Final Thoughts

Chatbot Confidential: Using AI for Tax Preparation provides a nuanced examination of the advantages and pitfalls of integrating AI into tax season. While AI chatbots like ChatGPT offer valuable assistance, users must approach them with caution, ensuring they verify information and safeguard their personal data. As AI continues to evolve, the Wall Street Journal's series promises to keep listeners informed about the intersection of technology, privacy, and everyday life.