Home › AI second opinion
Written and reviewed by Dick Tracy, licensed health insurance broker (NPN 20414610) · Published September 7, 2026
You asked an AI what to do about your health insurance, it gave you a confident answer, and something about it did not sit right. Paste it below and I will read it myself and tell you what holds up, what does not, and what it did not know to ask you. Free, no obligation, usually back to you within one business day. I am a licensed broker in 25 states, and I do this because half the people who call me were told something by a machine that was going to cost them money.
Let me be fair to the robots first. AI is genuinely good at explaining what a deductible is, what coinsurance means, and why your Explanation of Benefits reads like a ransom note. In the Nayya research below, 98 percent of employees said AI was easier to understand than their own plan documents. That is not a compliment to AI. That is an indictment of how insurance companies write.
Where it falls down is the part that costs you money. Four mistakes, over and over:
Ask an AI where a self-employed person gets health insurance and it will walk you straight to the marketplace, because that is what most of the internet says. In New York there is another door most people have never heard of: pre-established ERISA group plans you can merge into, which get you group rates, lower deductibles, and a nationwide PPO. It is not a loophole, it is a federal law from 1974. AI almost never brings it up, because almost nothing written online explains it.
"The average premium is about X" is a real number and a useless one. Your price is set by your state, your county, your age, your household size, and which door you walk through. I have had two people with nearly identical situations, one street apart, get quotes that were not close. An average is a starting point for an article, not a budget.
This is the one that bites hardest. Ask if your annual physical is covered and AI says yes, preventive care is free. True. Then you mention the sniffles while you are in there, the visit gets coded as a sick visit instead of routine, and it lands on your deductible. Preventive colonoscopy, free, until they remove a polyp and it becomes a surgical code. Mammogram, free, until dense tissue makes it diagnostic imaging. The word "covered" is doing a lot of work in that answer, and AI does not know to warn you.
They are not, and it is not close. Things that are routine in Tennessee are not sold in New York. Advice that is right in Florida can be flatly wrong in New Jersey. If the answer you got did not ask what state you live in before telling you what to do, that is the tell.
In April 2026, Nayya ran two independent surveys, 995 full-time benefits-eligible U.S. employees and 126 HR and benefits professionals, nationally representative by age, gender, household income, and region. What they found:
That is employees reporting on the answers they got, not an audit of the machines. Read it as "about half the time something was off," not "half of AI is a liar."
Of the people who acted on wrong AI guidance, more than a quarter reported unexpected out-of-pocket costs of \ or more. Some reported over \,500. That is the number that made me build this page.
So this is not a fringe habit, and I am not going to tell you to stop. I am telling you to get the answer checked before you act on it.
Source: Nayya, "New Nayya Research: 1 in 4 Employees Who Followed AI Benefits Advice Faced Unexpected Costs, Some Exceeding \,500," published May 19, 2026. Read the release. Survey of employees with employer benefits; I am citing it because the failure pattern is the same one I see with individual and self-employed plans.
Copy the whole answer, not just the part that worried you. Context matters, and the mistake is often three sentences above the part that caught your eye.
I read these myself, so give me a business day. If it turns out the AI was right, I will tell you that too, and you will not hear from me again unless you ask.
In the meantime, if you want to skip the wait: grab 15 minutes on my calendar.
A plain-English email from me, not a form letter and not a robot. It says which parts of the answer were right, which were wrong, and which were technically true but going to mislead you. If the AI was basically right, I say so. That happens, and telling you otherwise would make me the thing I am complaining about.
What you will not get: a quote you did not ask for, a phone call you did not agree to, or your email address on a list. If you want to talk after reading it, the calendar link is right there. If you do not, we are done, and you got a free second opinion out of it.
Two reasons, and one of them is self-interested, so I will say both. The honest one: I spent years on the inside of healthcare, reading billing codes and certificates of medical necessity, and roughly 80 percent of the medical claims I look at are coded wrong. People have been getting bad information about this stuff since long before AI, and it has always cost them. Now the bad information is faster and sounds more confident.
The self-interested one: some of you will read what I send back, realize this is more complicated than a chatbot can handle, and hire me. That is the whole business model. I would rather earn it by being right than by being loud.
Fifteen minutes, no pitch. Bring the AI answer, bring your current plan, bring the bill that made no sense. I educate, you decide.