A controlled test

We gave ChatGPT a real medical bill. It found 1 error. We found 4.

Same bill, same EOB, same insurance plan. Here is everything the general-purpose model missed, and why it could not have caught it.

ChatGPT

1

error found

$660 flagged

rekupr

4

errors found

$954.40 flagged

$294.40 in real errors that ChatGPT missed

The bill we put to the test

One real, de-identified childbirth bill, run through both with identical documents.

Scenario

Vaginal delivery

Bill total

$19,065

Documents

Bill, EOB, insurance

Billed to patient

$3,640.80

Same documents, different outcomes

ChatGPT caught one charge. The rules engine, the code databases, and the policy lookup found three more.

ChatGPT

$660 flagged

Balance billing on CPT 59400

Coinsurance on billed vs allowed amount

$660

Also noted: CPT 36415 missing its description, but dismissed it as "not a billing issue."

"All other line items match the EOB exactly."

rekupr

$954.40 flagged

Balance billing on CPT 59400

Coinsurance taken on the billed amount instead of the allowed amount.

$660

NCCI bundling violation

Postpartum care (59430) is already inside the global OB package (59400).

$176

Copay misapplication

$150 charged where the plan specifies $60 for a specialist visit.

$90

Duplicate charge on CPT 96372

The same Pitocin injection billed twice on one date of service.

$28.40

Three errors. Three blind spots.

Each one needs a different layer of detection. A language model only has the one.

  1. 01

    Missed by ChatGPT

    NCCI bundling violation

    $176

    Postpartum care (CPT 59430) was billed on its own, but it is already included in the global OB package (CPT 59400). The hospital charged for the same care twice.

    Requires: NCCI CPT-pair database1.7M coded pairs, not something a language model memorizes.
  2. 02

    Missed by ChatGPT

    Copay misapplication

    $90

    The neonatology consultation (CPT 99252) carried a $150 copay. The patient’s Aetna Choice POS II plan specifies $60 for a specialist visit. That is $90 too much.

    Requires: Your insurance policyImpossible to catch without the specific plan details.
  3. 03

    Missed by ChatGPT

    Duplicate charge

    $28.40

    CPT 96372 (Pitocin injection) appears twice on the same date. ChatGPT saw both lines and still concluded "all other line items match the EOB exactly."

    Requires: Systematic line matchingA model could catch this. It just did not.

Purpose-built beats general-purpose

ChatGPT makes one general pass at your bill. rekupr runs it through five layers of verified analysis.

rekupr

5 layers
1

Rules engine

35+ specialized billing rules

2

Code databases

1.7M NCCI pairs, 17,600 CPT codes

3

Insurance policy

Your plan’s copays, deductibles, limits

4

EOB reconciliation

Line-by-line bill against EOB

5

Output

Structured, verified, actionable

ChatGPT

1 pass
1

General-purpose pass

A best guess from a chat model

No rules engine

No licensed code databases

No plan awareness

No EOB reconciliation

Your bill deserves better than a best guess

Upload your bill and get verified, structured results in minutes.

Analyze my bill free

No credit card required. HIPAA compliant.