Tax season 2026 has a new unofficial advisor, and it doesn’t have a CPA license. Roughly half of Americans say they’ve already asked an AI chatbot a tax question, and that number is climbing fast. But here’s the thing: these tools are confidently wrong just often enough to cost you real money. A recent NerdWallet analysis of what AI gets right and wrong about taxes found that chatbots aced the easy stuff but stumbled badly on personalized advice, state-specific rules, and even basic dollar amounts.
Where AI Chatbots Actually Shine on Tax Questions
Credit where it’s due: AI has gotten dramatically better at taxes compared to even two years ago. When NerdWallet’s data team tested ChatGPT, Gemini, and Perplexity with questions pulled from the IRS enrolled agent practice exam (basically the bar exam for tax pros), all three nearly aced them.
Here’s what the chatbots handled well:
- Black-and-white factual questions about federal tax rules, filing statuses, and standard definitions
- General guidance for first-time filers, including step-by-step overviews of the filing process
- Explaining the reasoning behind answers, which makes it easier for you to verify their work
- Basic arithmetic, which used to be embarrassingly bad but has improved significantly in 2026 models
If you need a quick refresher on whether you qualify for a specific credit or want a plain-English explanation of a tax concept, chatbots are genuinely useful. Think of them like a smart study buddy who read the entire IRS publication library overnight.
The $3,000 Mistake You Wouldn’t Catch
Here’s where things get uncomfortable. NerdWallet’s team found that chatbots stated an incorrect standard deduction amount in some responses, off by nearly $3,000. That’s not a rounding error. That’s the difference between owing the IRS money and getting a refund.
The scariest part? The chatbot delivered the wrong number with the same confident tone it used for correct answers. There’s no hesitation, no asterisk, no “I’m not 100% sure about this.” It just states the wrong figure like it’s reading from the tax code.
Other specific errors the testing uncovered:
| Error Type | What the Chatbot Said | What’s Actually True |
|---|---|---|
| EV tax credit | A new Tesla qualifies for the credit | It doesn’t (manufacturer eligibility changes) |
| Overtime pay | $10,000 in overtime isn’t deductible | Overtime pay is taxable income, not a deduction question – the framing was wrong |
| State filing | File in the wrong state | Multiple chatbots focused on federal rules and botched state-specific guidance |
| Standard deduction | Off by ~$3,000 | Used outdated figures from a prior tax year |
Sam Taube, lead writer for NerdWallet’s investing and taxes team, put it bluntly: “Taxes involve both math and law. Even though AI has gotten better at the former, it’s not a reliable source of truth yet.”
The Inconsistency Problem That Should Worry You
Ask a CPA the same question twice, and you’ll get the same answer. Ask a chatbot the same question twice, and you might get two different answers. NerdWallet’s team had three testers ask identical questions with identical fictional taxpayer profiles, and the results were all over the map.
One example: when asked to estimate a tax bill for the same fictional filer, responses ranged from a specific “$32,267” to a vague “tens of thousands.” One chatbot casually suggested saving an extra 20% as a “cushion,” which would mean coming up with an additional $6,000 on short notice. That’s not a minor discrepancy.
The tax software recommendation test was even more revealing. Same prompt, same filer profile, different testers:
- Tester 1 was told FreeTaxUSA was the best option
- Tester 2 was told TurboTax was the best option
- One chatbot recommended a pricier product and then, in its own explanation, admitted a cheaper option had the same features
That last one is particularly telling. A human reviewer with real expertise applies consistent logic. Chatbots can mimic the structure of expert reasoning, but the informed judgment underneath is missing. What fills that gap might be as random as a coin flip.
Your Chat History Is Quietly Poisoning Your Tax Advice
This is the finding from NerdWallet’s analysis that doesn’t get enough attention. Even when testers explicitly told chatbots to forget all previous conversations and make no assumptions, the tools still pulled in biographical details from earlier chats.
Red flags the testers documented:
- The chatbot used personal details from completely unrelated previous questions to inform tax advice
- VPN location appeared to influence state filing recommendations, regardless of what the user actually said about where they lived
- It confused the state where a filer attended college with the state where they worked a summer job
- First-time filer guidance skipped basics like the April 15 deadline and penalties for not filing, apparently because the chatbot’s “picture” of the user suggested more experience
Your tax situation is like a fingerprint that changes over time. New job, new state, marriage, kids, retirement contributions: all of these shift your tax obligations. If the chatbot is building a profile of you from scattered chat fragments, that profile could be outdated or flat-out wrong.
And here’s a scenario worth considering: if someone else uses your account, or you ask a question from a hypothetical perspective, can the chatbot tell the difference? Based on this testing, the answer is no.
Five Rules for Using AI on Your 2026 Taxes Without Getting Burned
Rather than avoiding chatbots entirely (that ship has sailed), here’s how to use them without setting yourself up for an audit.
1. Treat every AI answer as a first draft
Verify specific numbers, deadlines, and eligibility claims against IRS.gov or a qualified tax professional before acting on them. This is especially true for dollar amounts, credit eligibility, and state-specific rules, which are the three areas where chatbots failed most often.
2. Match the tool to the question
General questions about how tax brackets work? Chatbots are fine. Questions involving your specific income, deductions, and filing situation? You want either tax preparation software with built-in validation or a human professional. Many tax software companies now embed AI chatbots trained specifically on tax data, which may perform better than general-purpose tools for personalized questions.
3. Know your safety net
If a chatbot gives you bad advice about which deductions to claim, your tax software or CPA will likely catch the error during the actual filing process. That’s a solid safety net for filing-related questions.
But strategic tax planning advice, like whether to convert a traditional IRA to a Roth, or how to structure freelance income, has consequences that might not surface for years. By then, you can’t always unwind the damage. An amended return squares things up with the IRS, but it doesn’t reverse every financial decision you made based on faulty guidance.
4. Use the best model available to you
NerdWallet tested the default (free) versions of each chatbot. Paid tiers with more advanced models may handle nuanced tax questions better. Check your chatbot’s settings to confirm which version you’re running.
5. Verify your chatbot’s “picture” of you
Start tax-related conversations in a fresh session. State your filing status, state of residence, and key financial details explicitly. Don’t assume the chatbot remembers you correctly, because the testing showed it often doesn’t.
How the Math Actually Works: AI’s Accuracy by Question Type
Here’s a simplified breakdown of how chatbot performance varied based on the type of tax question:
| Question Type | Accuracy Level | Risk to You |
|---|---|---|
| IRS factual questions (single correct answer) | High (nearly perfect) | Low |
| General filing guidance | Moderate to high | Low to moderate |
| Personalized tax estimates | Inconsistent | Moderate |
| State-specific filing rules | Poor | High |
| Product recommendations | Inconsistent and contradictory | Moderate |
| Strategic tax planning | Unreliable | High |
The pattern is clear: the more personal and specific your question, the less reliable the answer.
Warning Signs Your Chatbot Is Giving You Bad Tax Advice
Watch for these red flags during any tax-related AI conversation:
- It references details you didn’t provide in the current chat (it’s pulling from old conversations)
- It gives a specific dollar amount without showing the calculation (verify the math yourself)
- It doesn’t ask follow-up questions about your situation before giving personalized advice
- It recommends a tax strategy without mentioning any limitations or eligibility requirements
- It confidently discusses state tax rules without confirming which state you’re asking about
If you spot any of these, stop and cross-reference the information before acting on it.
Frequently Asked Questions
Can I use ChatGPT or Gemini to file my taxes?
No. AI chatbots cannot file taxes on your behalf. They can answer questions about tax concepts and provide general guidance, but the actual filing process requires tax software (like TurboTax, H&R Block, or FreeTaxUSA) or a tax professional. Think of chatbots as a research tool, not a filing tool.
Are AI chatbots accurate enough for basic tax questions in 2026?
For straightforward factual questions with a single correct answer, yes, they’re quite accurate. NerdWallet’s testing found that all three major chatbots nearly aced IRS-level quiz questions. The problems emerge with personalized, state-specific, or open-ended questions where accuracy drops significantly.
What’s the biggest risk of using AI for tax advice?
The biggest risk is that wrong answers sound exactly like right answers. Chatbots don’t flag uncertainty or qualify their confidence level. A $3,000 error in your standard deduction or incorrect state filing guidance gets delivered with the same authoritative tone as a perfectly accurate response. Always verify specific numbers and rules independently.
Should I pay for a premium AI model for tax questions?
It may help. NerdWallet’s testing used free default versions, and paid tiers with more advanced models could handle complex questions better. That said, even premium AI models shouldn’t replace tax software or professional advice for your actual filing. If your tax situation involves multiple income sources, state residency questions, or significant deductions, spend that money on a CPA consultation instead. A single session with a qualified tax advisor (typically $150 to $400) could save you far more than a chatbot subscription.
