I hit a milestone I hope stays in the dark: my health plan’s $10,150 annual cap. Happened in May 2026. Mostly thanks to two major eye surgeries that left my wallet empty.
Good news? I stop paying co-pays now. As long I keep the premiums flowing.
Bad news? Earlier this year I stared at what looked like an infinite waterfall of invoices and wondered: Is someone ripping me off?
I write about money for a living. I know medical billing is a swamp. Mistakes aren’t always obvious typos. Usually they’re hidden behind layers of CPT codes and opaque insurance lingo that most doctors wouldn’t fully parse, let alone patients.
Take my paperwork. Four and a half months yielded 87 separate claims. My policy contract? 149 pages of dense legal speak.
I didn’t want a degree in medical coding. But maybe Generative AI did. After all, large language models are supposed to chew on messy data and spit out the anomalies.
Here’s what actually happened.
The Toolhunt Was Barely That
I assumed the internet would be full of AI bill-checkers. It’s not.
Most of this tech is built for hospital admins, not humans trying to keep their retirement fund intact. Why? Providers want fewer rejected claims. Patients want refunds. The incentives don’t match.
The few apps out there solve tiny problems. One company, Counterforce Health, uses AI to draft appeals when a claim gets denied. That’s niche. What’s missing? A general-purpose auditor for your stack of bills.
So I went analog-ish. I used ChatGPT Plus.
I already pay $20 a month for it. I’d been using it to write scripts when customer service tried to deny me care. It seemed logical. Why not dump the receipts on it?
Here is my methodology, stripped bare:
- I ignored any bill under $150. Too noisy.
- Downloaded my 149-page policy contract and the Explanations of Benefits (EOBs).
- Forced my providers to send itemized statements. Crucial for seeing line-item junk fees.
- Built a spreadsheet of all 87 claims.
- Scraped my name, DOB, and ID numbers. Privacy matters, even when talking to a chatbot.
Then I pasted everything in and hit enter.
The AI Didn’t See the Forest for the Trees
Before uploading even a single invoice, I noticed a massive hole in my logic. How does the bot know if I actually got the care it billed?
My first bill said 31 minutes of operating room time. Did I? Did I bring a stopwatch into anesthesia? Probably not.
Maybe if the bill said 15 minutes, the AI would have blinked. Or if the pre-op eyedrops were listed at $4,000 instead of $12. But accuracy requires truth. AI doesn’t have a retina for my eye surgeries. It has text.
So the bot did what bots do. It focused on the numbers. It told me the insurer paid a pittance compared to the surgeon’s bill. Technically true. Useless contextually.
This says nothing about billing errors. It just proves the US healthcare pricing system is insane.
The AI did flag one denied claim on my spreadsheet. I looked. The surgeon had voluntarily pulled it to resubmit before the insurance company could even reject it.
Pharmacy claims were reversed? Yeah. The auto-refill system worked a little too well. I didn’t need three bottles of the same drops in the same week.
The magic ended fast. I wasn’t going to let the bot fish in an ocean I hadn’t mapped. So I got specific.
I spotted a $250 charge myself.
Breakdown: A $100 specialist co-payment plus a $150 drug co-payment. Total $250. I knew the system sometimes doubles down like this. But an online rep had told me otherwise. He said I’d only pay one or the other.
I uploaded the chat transcript. Asked ChatGPT: Is this an error?
It killed me instantly. Pointed right to section 12.4 of my policy. You owe both. The rep was wrong. The bill was right. I lost that round.
Fine. But then the math broke.
The Ghost in the Numbers
Here is the problem I couldn’t ignore:
I had paid $11,512.
My annual limit was $10,150.
Why the difference?
I asked ChatGpt. It confidently stated I’d paid exactly $10,150 in patient responsibility. I stared at my bank account. Then at the screen.
Then I realized why.
The bot was looking at my EOBs. My EOBs only reflected the amount the insurer considered “applied” toward the maximum at that moment in time. It wasn’t tracking what I actually wrote checks for.
Flashback three weeks earlier.
Same surgery. Different eye. Right this time.
I had already met my deductible from the left eye procedure. The hospital didn’t know this clearly enough to adjust their upfront charge. They charged me $1,552, guessing it would be roughly the same split as before (50% coinsurance, likely).
But since my deductible was gone, my actual responsibility dropped to $999 according to the EOB.
I had paid $1,552 upfront. The bill said I owed $999.
I overpaid $1,512.
I told the bot. Did it catch that discrepancy?
No.
I had to find the error manually. I had to match the credit card receipt to the delayed EOB explanation.
Once I pointed the discrepancy out, ChatGPT was useful. It confirmed: Yes, look, you paid 1,552 but responsibility is 999. They owe you money.
It was like using TurboTax and having to calculate the depreciation schedule yourself. The software helps file, it doesn’t dig through your shoe boxes.
What You Take Away
It is possible other errors hide in my bills. Maybe. Who knows?
Does it matter? Not really. I’ve hit the out-of-pocket ceiling. The insurance company and the hospital can sue each other for the next five years. That is their fight. Not mine.
My $1,512 is stuck in administrative purgatory. I’m waiting.
If you want to try this? Good luck. You need patience. You need data. And you need to stop expecting magic.
Start here.
- Gather everything. Itemized bills only.
- Redact the PII. Protect yourself.
- Prompt heavily. Act like you know nothing about billing codes. Let the bot pretend to be an expert.
- Verify everything the AI says against the PDF text.
The machine helps you read. It does not help you live.
