Coding errors now drive nearly 7 in 10 completed payer-audit denials, MDaudit reports

Payer audits are hitting coding harder, and there are more of them. MDaudit, a company that audits coding for hospitals and physician groups, reported on July 15, 2026 that payer audit volume and dollars at risk climbed 30% year over year, and that coding errors now drive nearly seven in 10 completed denials across its customer base. The figures come from the vendor's own analysis of the claims its clients run, not a government dataset, so read them as a benchmark from one large book of business, not a national rate.

What the benchmark numbers say

MDaudit's read on 2026 payer activity is blunt: more audits, more dollars at risk, and coding is where the denials stick. The vendor reports audit volume and total dollars at risk up 30% over the prior year, with coding-related denials up as much as 26%.

Its 2025 Annual Benchmark Report, published November 18, 2025 from the first three quarters of 2025 across more than 1.2 million providers, shows where the pressure lands. It describes a near fivefold jump in Request for Information (RFI) and medical-necessity denials, with the average denied amount on those up 70% across all settings. Telehealth denials rose 84%.

Why the audit itself is changing

The same announcement launched an AI tool, Auditor Assist, that reads the medical record against the coded claim to flag accuracy problems. MDaudit is candid about the limit: in its words the tool "does not replace the auditor's judgment; it sharpens it," and "the auditor still makes the call." It surfaces the charts worth a closer look and lines up the evidence, and a person still decides whether the code holds.

Payers are moving the same way from the other side. A machine that compares thousands of notes against submitted codes doesn't get tired at chart 400, and it flags the vague medical-necessity statement or the missing comorbidity every time. That's part of why RFI and medical-necessity denials are climbing: the reviewer no longer samples, it screens the whole batch. For a small practice, the practical shift is that a weak note is now more likely to be seen.

Where a human coder is still required

None of these tools read intent, and none of them build the record for you. An engine can tell a payer that the note doesn't clearly support the level billed. It can't tell whether the physician actually did the work and documented it thinly, which is a coding-and-documentation fix rather than a billing one. Naming the diagnosis, tying it to the encounter, and confirming the note carries MEAT-level support is still coder work, and the audits above are built entirely on whether that work got done.

The gap the automation leaves is judgment about the record. Is this "history of" or active. Does the comorbidity actually get addressed at this visit, or is it riding along from the problem list. Does the medical-necessity language match what the payer's policy asks for. Those are the calls a coder makes and a model only flags.

A pre-audit self-check for a small team

You don't need audit software to sample your own charts the way a payer's engine will. The point is to catch the weak note before it becomes a denied claim, working from the categories MDaudit says are moving.

Run a small monthly sample of your highest-volume claim types against that middle column. If a note wouldn't survive a stranger reading it cold, it won't survive an automated screen either.

How this fits the trend we've been tracking

This is the collection side of a story we've covered from the documentation side. Federal data already shows insufficient documentation as the top driver of Medicare improper payments, and separate analysis has flagged AI scribing pushing coding intensity up in ways payers are watching. MDaudit's numbers are the downstream result: the denials and audits that land when the documentation doesn't hold. The action is the same on all of them, which is to make the note carry the code before anyone, human or machine, comes looking.