Picture two law firms in a billing dispute, each hiring the other’s former partner. Nothing gets resolved, the paperwork triples, and the client pays for all of it. That’s roughly where American healthcare has landed with AI, except the client is you, and the invoice shows up as a premium increase.
Insurers are now claiming that AI is actively driving up healthcare costs. The mechanism they point to is upcoding: AI documentation and coding tools produce richer, more detailed clinical notes, those notes support higher complexity codes, higher codes mean bigger bills, and bigger bills eventually mean higher premiums. Blue Cross Blue Shield says its data backs this up, per STAT. Industry forecasts on 2026 cost trends name AI documentation and coding tools as an accelerant, alongside a more complex patient population. Payers say they’re already seeing billing amounts climb, and they’re adjusting policies to manage the spend.
Reviewing the tool by what it promises
I spend most of my time evaluating AI products, and there’s a rule that holds up well: read the sales page before you argue about the outcome. Ambient scribes and coding assistants are not sold to hospitals as cost reducers for the system. They’re sold as tools that capture what clinicians were previously failing to document, and documentation is what determines payment. “Stop leaving money on the table” is the pitch, more or less. So when payers report that billing amounts went up after adoption, that isn’t a malfunction. That’s the product working.
Which is what makes this whole fight so slippery. There are two readings of the exact same number, and from the outside they look identical:
- Doctors were always doing complicated work and under-recording it. AI closed the gap. Bills went up because bills were wrong before.
- AI learned that verbose, complexity-flagged notes pay better, and it now produces them by default. Bills went up because the paperwork got better at arguing.
Both stories predict higher costs. Neither is easy to disprove with claims data alone. And the party publishing the analysis is the party writing the checks, which does not make it wrong, but does mean the interpretation is not neutral. An insurer’s incentive is to classify increased documentation as inflation. A hospital’s incentive is to classify it as accuracy. The chart looks the same either way.
Everyone brought a model
The part that gets undersold is that this is not providers deploying AI against a human-staffed payer. CBS News reported in March 2026 that U.S. hospitals and insurers are both turning to AI in fights over claims and payments. On the payer side, a 2026 legislative briefing on AI in health insurance notes prior authorization as one of the main places insurers have started using it, deciding whether care gets approved before it happens.
So the shape of the thing: an AI writes a note designed to justify payment, and an AI evaluates whether payment is justified. Volume goes up on both ends. Speed goes up on both ends. The number of humans who understand any individual decision goes down. This is an arms race where the ammunition is documents, and the only guaranteed growth sector is administrative overhead.
If you’ve been following AI adoption in any industry, you know how this usually goes. Automation lands first on whichever task is easiest to measure and most directly tied to revenue. In healthcare, that task is billing. Not diagnosis, not triage, not the things that would actually justify the hype. Billing.
What I’d want to see before believing anyone
As a reviewer, my honest position is that the insurers’ claim is plausible and unproven, and the providers’ defense is also plausible and unproven. What would move me:
- Coding distributions before and after AI adoption, matched against actual clinical outcomes rather than just against prior billing patterns. If complexity codes rise and nothing about the patients changed, that’s a real signal.
- Independent audits, not payer-authored analyses or vendor case studies. Both current sources have a financial position.
- Denial rates and appeal outcomes on the payer side over the same window. If approvals tighten as documentation improves, the story is an arms race, not upcoding.
None of that changes the practical read for anyone buying these tools. If a vendor is telling you their scribe will raise your reimbursement per encounter, they are telling you the same thing insurers are complaining about, just with friendlier framing. That’s not a reason to avoid the category. Clinical documentation genuinely is miserable work and these tools genuinely help with it. But you should know which outcome you’re actually purchasing, because payers now know, and they’re rewriting policy in response.
The patient, as usual, didn’t buy any software and gets billed for all of it.
🕒 Published: