Do you still need chart review if you use an AI scribe?
Yes. An AI scribe writes the note from the visit, and nobody checks the codes against that note until a claim comes back denied. Here's what slips through, and which tool a small practice should start with.
By Jake Schwartz, CEO & Founder
Yes. An AI scribe turns the visit conversation into a note, and it never checks the diagnosis codes on the claim against what that note says. Those codes still need a check before the claim goes out, by your biller or by software that reads the finished chart.
Most small practices end up needing both. Which one comes first depends on where your money and time are going.
What does an AI scribe miss that ends up on the claim?
A scribe writes down what it hears. It can't write a side nobody said out loud, and it doesn't see the old codes sitting on the problem list. The codes on the claim usually come from that problem list or from a quick pick at the end of the visit, and nobody compares them to the draft.
Here are four lines a scribe could draft at a primary care visit, the codes that went on the claim, and what a check before signing finds.
| What the scribe wrote | What went on the claim | What a check before signing finds |
|---|---|---|
| "Right knee pain for three weeks, worse on stairs." | M25.569, pain in unspecified knee | The note says right. The code with the side is M25.561. |
| "Ear wax removed with irrigation. Canal clear afterward." | H61.20, impacted ear wax, unspecified ear | Nobody said which ear, so the note doesn't either. The doctor adds it, and the code follows: H61.21 right, H61.22 left, H61.23 both. |
| "A1c 7.1 today and 6.9 in August. Type 2 diabetes. Start metformin." | E11.9, type 2 diabetes, with R73.03 (prediabetes) still on the problem list | Prediabetes and type 2 diabetes can't be billed together. R73.03 comes off the list. |
| "Diabetes and stage 3a kidney disease, both stable on current meds." | E11.9 and N18.31, coded separately | The coding rules presume diabetes caused the kidney disease. If the doctor confirms that, the codes are E11.22 and N18.31. |
In the first row, the note already names the right knee, so only the code needs fixing. Many insurance companies deny an unspecified code when a code with a side exists, and Anthem told its California Medicare Advantage providers in 2023 that it would.
The second row happens because a scribe can only write what it hears. The doctor looked in the ear and knew which one it was. The recording only caught "let's get that wax out." The official coding guidelines say to use the unspecified code when the record doesn't name a side, so a biller can't just guess. The fix has to go in the note, and the cheapest time to make it is before the doctor signs.
The third row comes from a problem list that never got cleaned up. The scribe wrote today's diagnosis correctly. Prediabetes stayed on the list from last year and rode along onto the claim. The ICD-10-CM code book has a rule for pairs like this, called Excludes1, meaning the two codes can't be billed together. The code book has more than 5,000 of these rules. See how Excludes1 works.
The fourth row won't always get denied, but it's coded wrong. ICD-10-CM has combination codes that tell the whole story in one code, and E11.22 is type 2 diabetes with kidney disease. Billing the pieces separately sends a claim that doesn't match what the chart supports.
What does one missed code a day cost?
Say one claim a day goes out with a missing side or two codes that can't be billed together. That's about 20 denials a month. An MGMA study put the staff time to fix and resend a denied claim at about $25, so that's $500 a month, or $6,000 a year, in rework alone.
The fixed claims also pay late. A billing lead at a family medicine practice told us a corrected claim takes 30 to 45 days to pay, compared with 7 to 14 days for a clean one. And any denial nobody gets to before the insurance company's deadline is money you earned and never collect.
Why doesn't the scribe catch it?
A scribe's job ends when the draft is done. It listens to the visit and writes the history, exam, assessment and plan. Some scribe products now suggest codes too, but a code suggested from the draft can only be as complete as the draft. If nobody said "left," the suggestion has no side either.
The scribe also hears only today. It doesn't know prediabetes has been on the problem list since 2024, or that diabetes and kidney disease were coded separately three visits ago. Old codes like those ride onto today's claim unless someone checks.
Chart review starts where the scribe stops. It reads the finished note next to the codes and asks three questions. Do any two codes contradict each other? Does every code that needs a side, a stage or a link have one in the note? Does the note support the visit level, up or down? Here's the full checklist before signing.
Should a small practice start with a scribe or chart review?
Start with whichever problem is costing you the most this month. Here's how the common tools line up.
| If your biggest problem is | Start with | What it does |
|---|---|---|
| Doctors finishing notes at 10 p.m. | An AI scribe | Writes the first draft of the note from the visit conversation. |
| Claims denied for a missing side, codes that can't be billed together, or codes that should be one | A chart check before signing | Reads the note and the codes together while the chart is still open. |
| Insurance paying a 99214 as a 99213, or nobody sure the note supports the level | A visit level check | Shows the level the note's medical decision making supports, higher or lower. |
| Claims rejected for a wrong ID, a missing modifier or eligibility | A claim scrubber | Checks each claim against insurance rules before it's sent. It never sees the note. |
A claim scrubber is software, usually from your clearinghouse or billing system, that checks the claim form against insurance rules. It can't tell you the note says right knee when the claim says unspecified, because it only reads the claim.
If evenings are the pain, a scribe is a fair place to start. Spring 2026 research found scribes save some documentation time in clinic, while after-hours time barely moved and the drafts still needed a careful read. Here's what the studies found.
If denials are the pain, start with the check. It works on the claim you're about to send, and it reads any note, whether the doctor typed it, dictated it or had a scribe draft it.
Can you use a scribe and chart review together?
Yes. The tool that wrote the note shouldn't be the only thing checking it, so give each job to a different tool. The scribe drafts the note during the visit. The doctor reads it and adds anything the recording missed. Then the chart check reads the note and the codes together before the doctor signs.
Two habits make it work. Have the doctor say the side out loud in the room ("left ear," "right knee") so the scribe can write it. And clean up the problem list once, so old codes like prediabetes stop riding onto new claims.
Before you buy either tool, ask the vendor what leaves your office and who makes the final call. Here are the questions to ask.
Where Cortex Lens fits
Cortex Lens is the check after the scribe. It reads each Practice Fusion chart before it's signed, whoever wrote the note, and flags codes that can't be billed together, a missing left or right, and type 2 diabetes and kidney disease coded apart. It also shows the visit level the note supports, up or down. It's read-only, and your team makes every change in Practice Fusion. Book a demo.
Sources: Cost to fix a denied claim: an MGMA study cited in Family Practice Management, March/April 2015. Excludes1 pairs and combination codes: the FY2027 ICD-10-CM code set. Using an unspecified code when the record names no side: the FY2027 ICD-10-CM Official Guidelines. Anthem's unspecified-side edit: Anthem provider newsletter, California, July 2023. Payment times come from a billing lead at a family medicine practice.