I Asked ChatGPT To Plan a $100,000 a Year Retirement — Then Had a Financial Planner Review It

People use artificial intelligence (AI) for many things: meal planning, budgeting and interior design. But can you use a platform like ChatGPT for tasks that normally require a professional, like financial planning?

I chose a hypothetical scenario and asked ChatGPT to construct a retirement plan that would allow a retiree to live on $100,000 a year. I assumed an annual income of $125,000 for a 40-year-old who wants to retire in a fully paid-off home in Tennessee at age 65. After setting the parameters (and asking ChatGPT if they were realistic) I asked the generative AI to create a retirement plan.

Then, I asked Eric Franklin, CFP, managing principal at Prospero Wealth to review and assess it.

When I first reached out, Franklin said it sounded interesting; he’d never been enlisted to critique an AI system before. With 23 years of tech experience, he was the perfect choice.

“I had no idea what I was going to get,” he said.

The results were alarming. Not because ChatGPT’s output was immediately, obviously awful or right on the money.  It was scary because it fell into a gray area that artists call “the uncanny valley.”

At first glance, ChatGPT didn’t seem to be completely wrong.

Find Out: Here’s How Much You Need To Retire With a $100K Lifestyle

Read Next: 5 Clever Ways Retirees Are Earning Up To $1K per Month From Home

“I read it through and, at first, at the top level, it passed the sniff test. It looked fairly realistic,” Franklin said. “If I weren’t pushing on it too hard, I’d probably feel like it was somewhat accurate. Those were my first impressions.”

He acknowledged that the plan I presented was only one aspect of what financial advisors at Prospero Wealth offer clients. “It doesn’t include a lot of the things we’d normally include, like stress tests, accounting for changes in your career, changes in your family or possible relocations.”

But as Franklin dove deeper, he realized something that could be dangerous for anyone trying to use ChatGPT as a retirement planning tool without knowledge and experience to back it up.

It started with false premises, failed to adjust for inflation and created an unrealistic scenario. Granted, there were limitations in the exercise because the prompt was designed to be broad and concise. It was written by someone who is not a financial planner and intentionally didn’t flag anything that seemed off throughout the process. This prompt was designed for someone who thinks they can use ChatGPT to help them plan for retirement, plugging in basic information and expecting actionable insights. That probably describes a lot of people using ChatGPT.

But a less-than-perfect prompt doesn’t change the conclusion, which is important for anyone inclined to use AI for financial planning.

“The math is literally wrong,” Franklin said. “I passed it around the office and they all said the same thing.”

Exactly what did ChatGPT get right and wrong?

• Age: 40

• Location: Moderate cost of living

• Income: $125,000

• Target retirement age: 65

• Time horizon: 25 years

• Target retirement spending: $100,000 per year (today’s dollars)

• Emergency fund: Done

• Home: Owned, median-priced, no modeling of future downsizing or equity use

• Current retirement savings: $100,000 in 401(k), growing at 6%

ChatGPT used the 4% rule to determine the future retiree would need $2.5 million in today’s dollars by age 65, which would last for 30 years if they withdraw $100,000 per year.

Franklin pointed out that most professionals wouldn’t use the 4% withdrawal rule to determine how much someone needs to save for retirement. That was the first red flag.

“It’s actually pretty bad when you look at the real-world use case around the 4% rule. It was established in the ’90s, I believe, by a researcher who looked at the best withdrawal rate for the worst possible 30 years. You won’t outlive your money,” he said.

However, in 80% of the scenarios, Franklin shared, retirees died with a hefty amount of cash in their retirement accounts. “That doesn’t necessarily sound bad until you realize 80% of the people would under-live their resources and leave a lot on the table in terms of their possible enjoyment for retirement,” he explained. “More modern techniques are going to look at the way people actually spend in retirement. There are retirement strategies that allow people to start with a lot more happiness rather than austerity.”

ChatGPT’s amount of $2.5 million didn’t take into account inflation. This could have been the fault of a prompt that left out many details, but a financial planner would understand the reality of costs in 25 years compared to today based on historic Consumer Price Index (CPI) data.

“When I talk to clients, I want to use future dollars,” Franklin said. “The client’s going to anchor on that $2.5 million, but that $2.5 million is not going to be worth today’s $2.5 million in the future.”

Franklin explained that $209,000 in 25 years will have the spending power of $100,000 today. “You’re going to need something more like $5.2 million,” he said.

That’s where ChatGPT’s scenario starts to fall apart. But it gets worse.

Without clarification or prompting, ChatGPT assumed an asset allocation of 80% stocks and 20% bonds at age 40, with a gradual shift to 60/40 by retirement, with annual rebalancing. The existing savings of $100,000 at age 30 would grow to roughly $430,000 by 65, which Franklin agreed is “close to accurate.” Then, ChatGPT estimated $2.07 million funded through new contributions.

ChatGPT said: To reach the remaining $2.07 million over 25 years at 6% real:

• Annual savings needed: About $26,000

• Monthly savings needed: About $2,170

• Savings rate: About 21% of gross income

“The truth is the math is just literally wrong in a big way,” Franklin said. He acknowledged that many advisors use a 6% to 7% real return assumption, but only for a portfolio made up of 100% equity. “Six percent is probably a bit ambitious, and that’s probably anchored off historical S&P returns, which is not the diversified approach most advisors would recommend,” he said.

Franklin ran the numbers on a slightly more realistic scenario, still being overly generous on a 6.5% return for the equity portion.

Franklin used ChatGPT’s contributions strategy of 21% of the person’s gross income for 25 years, or $2,170 per month. “The math comes out more than a half a million dollars under. It was significantly off,” he said.

Baffled, as many AI users are when ChatGPT fails at simple tasks a desktop calculator could accomplish, Franklin said, “I was a little bit appalled, that the simple task…it’s literally not even calculating the contribution strategy to get close to what it’s trying to achieve.”

This could be due to ChatGPT’s penchant to please, making users feel like any task is doable with the right plan. Or it could just be bad math.

“I think if you follow this plan, you’re going to fail mathematically and you’ll also fail psychologically,” Franklin said. “No plan is going to persist unabated for most clients. You’re going to have some life changes.”

ChatGPT failed to factor in any Social Security income, although the AI did say healthcare costs could be “partially offset by conservative withdrawal rate and Social Security buffer.”

Overall ChatGPT’s plan lacked human nuance, intuition, discipline (or lake thereof) or empathy.

“I don’t think it builds in any wiggle room for this client,” Franklin said. “Even if the math was right, I can’t think of a single client at this income level, $125,000 right now, who could put their head down for 25 years and put away 21% of their income.”

More From GOBankingRates

This article originally appeared on GOBankingRates.com: I Asked ChatGPT To Plan a $100,000 a Year Retirement — Then Had a Financial Planner Review It