Which AI Calculates Taxes Appropriately?

A number of readers requested me whether or not they might use AI to calculate taxes after studying my tales about AI serving to me with house electrical and water heater issues within the earlier submit AI Provides Higher Solutions Than Google. They need a tax projection to assist with setting tax withholding, paying estimated taxes, or planning for Roth conversions.

My instant response was that calculating taxes isn’t the perfect use of AI, as a result of it falls beneath “verifiable details” and “newest growth” classes. The retirement plan contribution limits, tax brackets, the utmost deduction quantities, and so on., are all verifiable details. The IRS units them yearly, and they’re what they’re. Simply go to the IRS web site for the newest numbers, or Google. In case you ask AI, it had higher get the newest numbers on-line anyway, as a result of AI’s coaching lags.

However, tax guidelines are difficult. After you’ve got all the newest numbers from the IRS, you continue to want to use the complicated guidelines — what counts and what doesn’t, and which charges apply to which revenue. On-line tax calculators normally don’t cowl the present 12 months till late within the 12 months, and they are often both too easy or too difficult. It might be good to have AI calculate taxes particularly for the forms of revenue and deductions now we have: not too easy with solely restricted inputs, and never too difficult with the whole lot beneath the solar.

The Quiz

I believed I might check how effectively AI calculates taxes. I got here up with this quiz query:

Jill, single, age 63, has these incomes in 2026:

• $30,000 from Social Safety
• $10,000 from curiosity and pre-tax IRA withdrawals
• $37,000 from certified dividends and long-term capital positive factors
• $1,000 from muni bond curiosity

Jill contributes $4,400 to her HSA and donates $1,500 in money on to charities. What’s Jill’s federal revenue tax for the 2026 tax 12 months?

I designed this query fastidiously to cowl a number of calculations. Barely lower than 85% of Social Safety is taxable on this case. AI has to know tips on how to calculate the precise taxable quantity, and never simply make a straight 85% of Social Safety taxable. The gross revenue consists of odd revenue, tax-exempt revenue, and funding revenue taxed at preferential charges. The odd revenue part goes throughout two tax brackets. So does the funding revenue. One deduction is above-the-line, and the opposite is below-the-line.

I added these directions to encourage AI to get the newest tax numbers from the IRS:

It’s necessary to calculate it precisely. Please use tax quantities just for 2026 and solely from official IRS sources.

Preliminary Outcomes

First, I ran the check case by my tax calculator in Calculator: How A lot of My Social Safety Advantages Is Taxable?. It gave this consequence:

$25,410 of your Social Safety advantages is taxable, which suggests 15% of your advantages is tax-free.

Your 2026 federal revenue tax is roughly $1,640.

That is calculated from a taxable revenue of $50,910 (AGI $68,010 – Customary Deduction $16,100 – Senior Deduction $0 – charitable contribution deduction $1,000).

Then I despatched the quiz to all 4 main AI chatbots: ChatGPT, Claude, Gemini, and Grok. I solely used the free model in each, with the free Pondering, Professional, or Knowledgeable mode enabled. I in contrast the solutions to the proper consequence from my tax calculator.

My quiz was too arduous! All 4 main AI fashions failed to provide the proper reply. The $1,500 money donation threw them off. It falls squarely within the “newest growth” class. The charity donation deduction for non-itemizers is new. The AI fashions all have “donations are deductible solely while you itemize” imprinted of their coaching. Claude, Gemini, and Grok all would have given the proper reply if I hadn’t included the money donation. ChatGPT was off on how a lot of Social Safety is taxable.

Whenever you see AI not providing you with a deduction, you may elevate a query and ask it to double-check. I adopted up with this assertion:

I heard that money donations are deductible for folks utilizing the usual deduction, beginning in 2026, as much as a restrict.

I solely stated “I heard,” which can or might not be true. AI nonetheless has to confirm. I additionally added this request for ChatGPT:

One other AI mannequin calculated a distinct quantity for a way a lot Social Safety is taxable. Please double-check your calculation to see who’s proper.

Spherical 2

All 4 AI chatbots double-checked and revised their solutions.

ChatGPT and Claude obtained the proper reply. Gemini and Grok mistakenly handled the money donation deduction as above-the-line and used it to cut back the Social Safety taxable quantity.

I requested Gemini and Grok to make clear whether or not the HSA contribution and the charity donation ought to or shouldn’t be included in calculating the Social Safety taxable quantity. Gemini didn’t notice its mistake. It stated each ought to be included. Grok refused to go additional as a result of I reached the message restrict for not having an account. [A reader with a Grok account did the same chat with Grok and told me that Grok also said both should be included.]

Listed here are the total chat transcripts. They make an fascinating learn.

https://chatgpt.com/share/69bf73fb-db44-8004-9905-b77b56fbb6ce

https://claude.ai/share/ef62408a-6502-4d70-82be-6ea46a4a4cc3

https://gemini.google.com/share/795bdad9f5ad

[No sharable link for Grok because I don’t have an account.]

Impressions

I don’t suppose we are able to draw definitive conclusions primarily based on just one check. I name these impressions.

1. Whether or not AI can calculate taxes accurately is determined by the complexity. Three out of the 4 main AI chatbots would’ve given the proper reply in a single go if the query didn’t embrace the brand new and difficult charity donation deduction. The specific request to retrieve the newest data from official IRS sources and turning on the Pondering/Professional/Knowledgeable mode in all probability helped in enhancing accuracy.

2. AI works higher while you make it a dialog. Don’t cease and say it’s ineffective while you spot some errors within the preliminary reply. Assist the instrument aid you. Each ChatGPT and Claude reached the proper reply in Spherical 2 after I requested them to double-check.

3. It takes little effort to ship the identical query to 2 or extra AI fashions. You don’t must know which AI is appropriate while you get completely different solutions. You possibly can inform one AI that one other AI stated one thing completely different. It’ll re-examine.

4. AI was incorrect as a result of human sources had been incorrect. Gemini and Grok handled the donation deduction as above-the-line as a result of many human sources incorrectly referred to as it above-the-line. In case you Googled, you’d encounter these incorrect sources too. ChatGPT and Claude used extra dependable sources and accurately recognized the distinction between above-the-line and never requiring itemizing deductions.

5. A web-based tax calculator is quicker and extra correct if you understand which one to make use of. That’s a giant “if.” Solely saying “Don’t use AI as a result of it’s typically incorrect” doesn’t say the place you will see that extra dependable sources.

As a lot as I’d prefer to see everybody use my tax calculator, let’s face it: 99.99% of individuals don’t know that I exist. They’ve little likelihood of discovering the few sources that do it precisely as a result of Google doesn’t rank tax calculators by timeliness or thoroughness.

Right here’s a small problem for you:

Faux you’re not accustomed to tax calculations. Discover one other technique to reply the quiz query. Don’t say “I’ve a spreadsheet,” or “I take advantage of the super-duper Case Research Spreadsheet or Excel1040.” 99.99% of individuals don’t have these. Please inform me within the feedback part how you might do it in any other case.

In need of discovering the nice sources, the outcomes from AI aren’t that dangerous (see #1).

6. AI’s normal method was appropriate even when it was incorrect in some nuances. All 4 AI chatbots adopted these appropriate steps:

• Calculate how a lot of Social Safety is taxable
• Calculate AGI
• Calculate taxable revenue
• Separate odd revenue and preferential funding revenue, and understand how they stack
• Apply completely different tax brackets

This normal method is “widespread information of an insider.” You possibly can be taught the steps from AI in the event you’re not accustomed to tax calculations. That’s extra precious than simply having a remaining quantity. Once I noticed that ChatGPT didn’t embrace particulars of some interim steps, I requested it to indicate its work. ChatGPT defined the main points sooner and higher than I might (learn the transcript).

A web-based tax calculator provides you a quantity however doesn’t clarify the steps. AI is a greater instrument in educating you tips on how to fish than providing you with a fish.