MW Should you trust AI for Medicare, Social Security or long-term-care advice?

By Richard Eisenberg

The stakes are high. Here's a guide to using AI for assistance in making complicated financial decisions.

Choosing the right Medicare coverage, deciding whether to buy a long-term-care insurance policy and determining when to claim Social Security retirement benefits are among the most complicated financial decisions older adults need to make.

Can AI help?

Based on my tests of seven popular artificial-intelligence bots and new specialized AI services, as well as interviews with experts, the answer is: Possibly, but exercise caution.

In just the past three years, AI has catapulted to a must-use technology that many Americans are deploying frequently for nearly every imaginable purpose. So, it might seem like AI's vast knowledge and speed could be a terrific way to help grapple with complex financial concerns like claiming Social Security, enrolling in Medicare and buying long-term-care insurance.

But there's also a potential danger: Mess up, and you could wind up making big mistakes that could endanger your financial security.

AI's large language models will likely do a pretty good job answering your basic questions about Medicare, long-term-care insurance and Social Security. But AI chatbots probably won't know enough about you to provide reliable, personalized advice and strategies.

You might also receive inaccurate information - "hallucinations," in AI jargon - or encounter errors of omission. As Microsoft's $(MSFT.UK)$ Copilot bot notes on its site: "You may see responses that sound convincing but are incomplete, inaccurate or inappropriate."

Andrew Lo, director of the Massachusetts Institute of Technology's Laboratory for Financial Engineering, estimates that "we're probably in the second or third inning of effectively dealing with AI hallucinations."

Worries about privacy

Privacy protection and sources used to provide your AI results vary from chatbot to chatbot, too. A survey from the venture-capital firm Menlo Ventures found that 71% of Americans who aren't using AI say it's because they're worried about the privacy of their data, while 58% don't trust AI information.

The beauty of AI is that it can find and synthesize massive amounts of information, often in seconds. That's why financial-services companies and advisers have been using AI to help people invest for retirement and create retirement spending plans, as MarketWatch's Robert Powell wrote in a recent piece.

In the past few years, AI has steamrolled into Americans' lives dramatically and is being adopted by people of all ages. According to Menlo Ventures, 59% of Generation X and 45% of baby boomers used it in the past six months - many of them daily.

But AI hasn't received much attention for what it can do to assist with other tricky financial decisions and complex strategies that retirees and preretirees must contemplate beyond investing and spending.

That's likely to begin changing now that bots like ChatGPT, Claude, Perplexity and Google Gemini $(GOOGL)$ $(GOOG)$ are getting better at reasoning. Companies are also adopting "agentic AI" to make decisions and take actions for people.

"We're at the dawn of a really new era," said John Cutler, a consultant with the Society of Actuaries.

Today's AI landscape also includes specialized financial decision-making platforms.

For instance, WaterLily can tell you whether to consider buying a long-term-care insurance policy. The company compares your situation with others in its AI database - reviewing half-a-billion data points - and can also say how likely a long-term-care policy you own will provide the benefits you may need. It's also free; WaterLily gets paid a flat fee by professionals offering the service. Individuals can get on a waiting list for a free consultation with WaterLily founder and Chief Executive Lily Vittayarukskal assessing the AI results.

The personalized AI service HeyMOE, from the Medicare advisory firm 65 Incorporated (MOE stands for Medicare Open Enrollment), reviews and recommends Medicare Part D prescription-drug plans for people who are looking for or already have them. It costs $30 a year and is a kind of souped-up version of Medicare.gov's Plan Finder tool - but here, AI does all the work and you don't.

The free Healthpilot AI tool helps people choose Medicare Part D plans, Medigap plans and Medicare Advantage plans sold by private health insurers. If you then want to enroll in one, you can do so with a Healthpilot independent agent.

Here's a guide to using AI for assistance in making decisions about Medicare, long-term-care insurance and Social Security.

Choosing Medicare coverage

Deciding on the right Medicare coverage means first figuring out whether to go with original Medicare plus a Medigap supplemental policy and a Part D plan, or with a Medicare Advantage and Part D plan. After that, you can determine which offerings to enroll in.

"Understanding Medicare is a perfect application for AI because it can help you navigate that system," said Dan Andersen, a guest editor of the recent Generations Journal issue on artificial intelligence and aging, published by the American Society of Aging.

But MIT's Lo said he would never recommend using only AI to make a final Medicare decision, because it won't know enough about your circumstances.

As Melinda Caughill, co-founder of 65 Incorporated, said: "The only thing in Medicare that is true for everyone is that nothing in Medicare is true for everyone."

To see how AI would do in helping people decide between original Medicare and Medicare Advantage, I asked seven chatbots - ChatGPT, Google Gemini, Microsoft Copilot, Perplexity, Claude, DeepSeek and Meta AI $(META)$ - this question: Should I go Medicare Advantage, or original Medicare with a Medigap policy and Part D drug plan?

All the bots did well conveying the pros and cons of original Medicare and Medicare Advantage in simple language.

They noted that original Medicare lets you see any doctor who takes Medicare, but generally doesn't cover vision, dental or hearing and, when combined with Medigap, typically costs more than Medicare Advantage. They also explained that Medicare Advantage plans restrict your choice of doctors and hospitals and require prior authorization to see specialists.

The bots were almost always accurate, too. One exception: Anthropic's Claude said "you have eight months from when you first enroll in Medicare Part B to buy Medigap without medical questions." In reality, it's six months.

That kind of an error could spell trouble, since some Medicare decisions are either permanent or come with stiff penalties if you mess up. Medicare won't let you reverse a blunder you made because you relied on information or advice from AI. The possibility of AI making a mistake is reason enough not to make a financial move only because of what you learn from a chatbot.

Google Gemini was the best overall for delivering a quick, detailed, useful and well-sourced answer.

ChatGPT did a fine job comparing original Medicare and Medicare Advantage, making it a good choice for the TL;DR set. But it stopped short of offering help choosing plans and didn't supply sources for its answer.

One intriguing standout: Microsoft's Copilot, which let me receive its AI answers in one of three ways:

-- Quick Response, which arrives in two to three seconds and is conversational. It offered to help compare plans where I live if I wanted. But Quick Response didn't cite its sources.

-- Think Deeper, with answers in 30 seconds. It delivered a side-by-side comparison, additional insights and offered to zero in on plans based on my healthcare use, budget tolerance and preferred doctors or hospitals. It also didn't cite sources.

-- Deep Research, which provides a detailed report in 10 minutes. My nine-page analysis was comprehensive and cited its sources, which were reputable.

However, ChatGPT, Copilot, Deepseek and Meta AI failed to mention that if you enroll in a Medicare Advantage plan and later want to switch to original Medicare, you can be rejected for a Medigap plan or face a steep annual premium.

DeepSeek - the free, open-source chatbot from China - spit out a decent answer, but lacked specifics and was laden with jargon. Privacy experts have raised concerns about DeepSeek's ability to share user data.

Meta AI's response to my question was adequate but skimpy and cited just three sources - including an insurance agents' site and a site that partners with an insurance broker.

Read: Here's how the senior bonus in the megabill will affect your taxes this year

Buying a long-term-care policy

Some 70% of Americans over 65 will need long-term care at some point, the U.S. Department of Health and Human Services estimates.

But with a semiprivate room in a nursing home costing around $100,000 a year and home care at roughly $80,000, according to Genworth Financial, some people in their 50s and 60s may wonder whether to buy a long-term-care insurance policy to defray those potential expenses.

Cathy Sikorski, a Pottstown, Pa.-based caregiving consultant, said she was "pleasantly surprised" at the answer she received when asking ChatGPT: "How can AI help me decide about buying long-term-care insurance?"

The chatbot explained that AI can estimate your long-term-care insurance costs to see if a policy would be affordable, run "what-if" scenarios, compare policies and explain complex terms.

ChatGPT also listed AI's downsides: It doesn't know your family's long-term-care plans or personal values, might be biased or outdated and may have privacy risks.

AI, of course, can't tell you whether you'll ever need long-term care or what care will cost when you reach your 80s and 90s. But the digital tool can scour long-term-care policies to compare premiums and coverage.

To get useful answers about cost, though, you'll need to get specific about the benefits you'd like.

MW Should you trust AI for Medicare, Social Security or long-term-care advice?

By Richard Eisenberg

The stakes are high. Here's a guide to using AI for assistance in making complicated financial decisions.

Choosing the right Medicare coverage, deciding whether to buy a long-term-care insurance policy and determining when to claim Social Security retirement benefits are among the most complicated financial decisions older adults need to make.

Can AI help?

Based on my tests of seven popular artificial-intelligence bots and new specialized AI services, as well as interviews with experts, the answer is: Possibly, but exercise caution.

In just the past three years, AI has catapulted to a must-use technology that many Americans are deploying frequently for nearly every imaginable purpose. So, it might seem like AI's vast knowledge and speed could be a terrific way to help grapple with complex financial concerns like claiming Social Security, enrolling in Medicare and buying long-term-care insurance.

But there's also a potential danger: Mess up, and you could wind up making big mistakes that could endanger your financial security.

AI's large language models will likely do a pretty good job answering your basic questions about Medicare, long-term-care insurance and Social Security. But AI chatbots probably won't know enough about you to provide reliable, personalized advice and strategies.

You might also receive inaccurate information - "hallucinations," in AI jargon - or encounter errors of omission. As Microsoft's (MSFT) Copilot bot notes on its site: "You may see responses that sound convincing but are incomplete, inaccurate or inappropriate."

Andrew Lo, director of the Massachusetts Institute of Technology's Laboratory for Financial Engineering, estimates that "we're probably in the second or third inning of effectively dealing with AI hallucinations."

Worries about privacy

Privacy protection and sources used to provide your AI results vary from chatbot to chatbot, too. A survey from the venture-capital firm Menlo Ventures found that 71% of Americans who aren't using AI say it's because they're worried about the privacy of their data, while 58% don't trust AI information.

The beauty of AI is that it can find and synthesize massive amounts of information, often in seconds. That's why financial-services companies and advisers have been using AI to help people invest for retirement and create retirement spending plans, as MarketWatch's Robert Powell wrote in a recent piece.

In the past few years, AI has steamrolled into Americans' lives dramatically and is being adopted by people of all ages. According to Menlo Ventures, 59% of Generation X and 45% of baby boomers used it in the past six months - many of them daily.

But AI hasn't received much attention for what it can do to assist with other tricky financial decisions and complex strategies that retirees and preretirees must contemplate beyond investing and spending.

That's likely to begin changing now that bots like ChatGPT, Claude, Perplexity and Google Gemini (GOOGL) $(GOOG.UK)$ are getting better at reasoning. Companies are also adopting "agentic AI" to make decisions and take actions for people.

"We're at the dawn of a really new era," said John Cutler, a consultant with the Society of Actuaries.

Today's AI landscape also includes specialized financial decision-making platforms.

For instance, WaterLily can tell you whether to consider buying a long-term-care insurance policy. The company compares your situation with others in its AI database - reviewing half-a-billion data points - and can also say how likely a long-term-care policy you own will provide the benefits you may need. It's also free; WaterLily gets paid a flat fee by professionals offering the service. Individuals can get on a waiting list for a free consultation with WaterLily founder and Chief Executive Lily Vittayarukskal assessing the AI results.

The personalized AI service HeyMOE, from the Medicare advisory firm 65 Incorporated (MOE stands for Medicare Open Enrollment), reviews and recommends Medicare Part D prescription-drug plans for people who are looking for or already have them. It costs $30 a year and is a kind of souped-up version of Medicare.gov's Plan Finder tool - but here, AI does all the work and you don't.

The free Healthpilot AI tool helps people choose Medicare Part D plans, Medigap plans and Medicare Advantage plans sold by private health insurers. If you then want to enroll in one, you can do so with a Healthpilot independent agent.

Here's a guide to using AI for assistance in making decisions about Medicare, long-term-care insurance and Social Security.

Choosing Medicare coverage

Deciding on the right Medicare coverage means first figuring out whether to go with original Medicare plus a Medigap supplemental policy and a Part D plan, or with a Medicare Advantage and Part D plan. After that, you can determine which offerings to enroll in.

"Understanding Medicare is a perfect application for AI because it can help you navigate that system," said Dan Andersen, a guest editor of the recent Generations Journal issue on artificial intelligence and aging, published by the American Society of Aging.

But MIT's Lo said he would never recommend using only AI to make a final Medicare decision, because it won't know enough about your circumstances.

As Melinda Caughill, co-founder of 65 Incorporated, said: "The only thing in Medicare that is true for everyone is that nothing in Medicare is true for everyone."

To see how AI would do in helping people decide between original Medicare and Medicare Advantage, I asked seven chatbots - ChatGPT, Google Gemini, Microsoft Copilot, Perplexity, Claude, DeepSeek and Meta AI (META) - this question: Should I go Medicare Advantage, or original Medicare with a Medigap policy and Part D drug plan?

All the bots did well conveying the pros and cons of original Medicare and Medicare Advantage in simple language.

They noted that original Medicare lets you see any doctor who takes Medicare, but generally doesn't cover vision, dental or hearing and, when combined with Medigap, typically costs more than Medicare Advantage. They also explained that Medicare Advantage plans restrict your choice of doctors and hospitals and require prior authorization to see specialists.

The bots were almost always accurate, too. One exception: Anthropic's Claude said "you have eight months from when you first enroll in Medicare Part B to buy Medigap without medical questions." In reality, it's six months.

That kind of an error could spell trouble, since some Medicare decisions are either permanent or come with stiff penalties if you mess up. Medicare won't let you reverse a blunder you made because you relied on information or advice from AI. The possibility of AI making a mistake is reason enough not to make a financial move only because of what you learn from a chatbot.

Google Gemini was the best overall for delivering a quick, detailed, useful and well-sourced answer.

ChatGPT did a fine job comparing original Medicare and Medicare Advantage, making it a good choice for the TL;DR set. But it stopped short of offering help choosing plans and didn't supply sources for its answer.

One intriguing standout: Microsoft's Copilot, which let me receive its AI answers in one of three ways:

-- Quick Response, which arrives in two to three seconds and is conversational. It offered to help compare plans where I live if I wanted. But Quick Response didn't cite its sources.

-- Think Deeper, with answers in 30 seconds. It delivered a side-by-side comparison, additional insights and offered to zero in on plans based on my healthcare use, budget tolerance and preferred doctors or hospitals. It also didn't cite sources.

-- Deep Research, which provides a detailed report in 10 minutes. My nine-page analysis was comprehensive and cited its sources, which were reputable.

However, ChatGPT, Copilot, Deepseek and Meta AI failed to mention that if you enroll in a Medicare Advantage plan and later want to switch to original Medicare, you can be rejected for a Medigap plan or face a steep annual premium.

DeepSeek - the free, open-source chatbot from China - spit out a decent answer, but lacked specifics and was laden with jargon. Privacy experts have raised concerns about DeepSeek's ability to share user data.

Meta AI's response to my question was adequate but skimpy and cited just three sources - including an insurance agents' site and a site that partners with an insurance broker.

Read: Here's how the senior bonus in the megabill will affect your taxes this year

Buying a long-term-care policy

Some 70% of Americans over 65 will need long-term care at some point, the U.S. Department of Health and Human Services estimates.

But with a semiprivate room in a nursing home costing around $100,000 a year and home care at roughly $80,000, according to Genworth Financial, some people in their 50s and 60s may wonder whether to buy a long-term-care insurance policy to defray those potential expenses.

Cathy Sikorski, a Pottstown, Pa.-based caregiving consultant, said she was "pleasantly surprised" at the answer she received when asking ChatGPT: "How can AI help me decide about buying long-term-care insurance?"

The chatbot explained that AI can estimate your long-term-care insurance costs to see if a policy would be affordable, run "what-if" scenarios, compare policies and explain complex terms.

ChatGPT also listed AI's downsides: It doesn't know your family's long-term-care plans or personal values, might be biased or outdated and may have privacy risks.

AI, of course, can't tell you whether you'll ever need long-term care or what care will cost when you reach your 80s and 90s. But the digital tool can scour long-term-care policies to compare premiums and coverage.

To get useful answers about cost, though, you'll need to get specific about the benefits you'd like.

(MORE TO FOLLOW) Dow Jones Newswires

July 11, 2025 13:56 ET (17:56 GMT)

MW Should you trust AI for Medicare, Social -2-

A 60-year-old man might pay between $1,200 and $3,800 a year for a policy, depending on whether he wants benefits to grow annually with inflation and at what inflation rate, according to the American Association for Long-Term Care Insurance.

The WaterLily AI service, Vittayarukskal said on the "Friends Talk Money" podcast I co-host, "allows you to change your long-term-care trajectory" to see how that could affect your decision about buying a policy.

"We ask: 'Do you want to change where you will live? Do you want to change how your family members might step in?' All of that has an associated cost with it," she said. "From there, we help you navigate, 'How do I solve for that future cost as efficiently as possible?'"

Within a year, Vittayarukskal hopes to launch a WaterLily AI service that consumers can use directly.

Read: Eldercare is overtaking childcare. Governments and employers need to step up.

Claiming Social Security

AI has strengths and weaknesses in helping you choose when to start taking Social Security benefits.

A chatbot can explain Social Security's complex claiming rules and provide the pros and cons of claiming at different ages. But it won't know enough about your work history, income and projected income sources (nor those of your spouse or partner if you have one) to deliver credible, personalized advice.

"You could use AI for a list of Social Security questions to consider asking yourself, recent changes in the law related to Social Security, and how much of Social Security benefits could be taxable," said Robert Laura, CEO of the Retirement Coaches Association. "But AI may not be up to par for a claiming strategy because there are so many factors to consider."

MIT's Lo urges extreme caution before relying on AI to choose when to claim Social Security.

This decision, he said, is "based on some fairly involved numerical calculations. Those kinds of calculations are not necessarily the strengths of large language models."

Two Social Security experts who quizzed AI about claiming Social Security received numerous errors.

The Social Security Advisory Service of the AMAC Foundation (a nonprofit affiliated with the conservative-leaning Association of Mature American Citizens) said ChatGPT provided the wrong answer about a spouse claiming retirement benefits. And Laurence Kotlikoff, co-author of "Get What's Yours: The Revised Secrets to Maxing Out Your Social Security" said Google's AI "returned five horribly wrong answers" when he asked five Social Security questions.

I contacted the companies behind these AI services to see if they had anything to add about using them in this way. A Meta AI spokesperson said it's important to consult a professional or your state health insurance assistance program for personalized guidance. Google declined to comment, and the rest didn't immediately respond in time for publication.

Advice for using AI

If you want to use AI for decisions about Medicare, long-term-care insurance or Social Security, experts offer this advice:

Be sure you're using the latest version of a chatbot. This will help ensure the answers you receive are as up-to-date as possible. ChatGPT is expected to release its new ChatGPT-5 model sometime this summer.

Don't be shy about asking follow-up prompts. You may not learn exactly what you're looking for from your initial prompt. Subsequent prompts can let you refine the question or get the chatbot to delve more deeply and provide more useful guidance.

When the chatbot doesn't reveal its sources or say how current its information is, ask. Your follow-up prompt can let you find out whether the AI answer you received is trustworthy and current.

Read an AI tool's privacy-protection language before signing up. "Look for updated privacy policies," said Jake Cavin, CEO of Healthpilot. You'll then see whether you'd be comfortable with its rules and whether you can opt out of any.

If a chatbot offers a free and a paid version, consider going for the paid one. Not only will it offer more features, but you might get more privacy protection. Paid chatbots typically cost about $20 a month.

Avoid sharing your personal health and financial data. That's because you can't always be sure where the information you give AI will end up. Keep yourself out of prompts; for help picking a Medicare Part D plan, for instance, ask which plans in a particular ZIP code cover certain prescriptions without saying you take them.

If you're worried the chatbot may spread information you provided, see if you can delete your prompts. Some AI bots will let you get them deleted immediately or after 30 days.

Review AI's responses with a trusted financial adviser for a second opinion. "Information from AI large language models needs to be vetted by human experts right now," said Lo. "The hope is that in a few years, we may get to the point where they're accurate and, in some cases, more accurate than humans. But that's not the case today."

Sikorski said that after hearing from your adviser about what AI said, you might even go back to AI for the chatbot's view of what your adviser told you.

"We have this amazing tool that we can use to say, 'Well, what do you think?'" she noted.

-Richard Eisenberg

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

July 11, 2025 13:56 ET (17:56 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.