You can use a general AI medical scribe in an oral and maxillofacial surgery practice. We tried. It gets you a rough draft, and then you spend ten to fifteen minutes per note fixing it.

This post is the unromantic version of why we built Scribbix as an OMFS AI scribe rather than another general medical scribe.

The vocabulary problem

A general medical scribe is trained mostly on internal medicine and primary care encounters. The training distribution is skewed accordingly. Drop a typical OMFS consult into one and watch:

“Pell-Gregory IIB mesioangular” gets transcribed as “Pelegrini II V mesial”
“Wilkes IV TMJ” becomes “Wilkins for TMJ”
“ASA II, Mallampati III” gets one of the two right, never both
“#17” lands as “number 17” with no recognition that it is the lower left third molar in Universal numbering
“lingual nerve paresthesia” frequently degrades to “lingual nerve paralysis”

None of these are typos in transcription. They are the model’s prior pulling the audio toward the closest non-OMFS word it has seen many times. You can fight it with prompt engineering. You can fight it with custom vocab lists. You will never fully win, because the underlying model has not seen enough oral surgery to make those terms its default.

An OMFS-specific AI scribe loads the OMFS vocabulary as the prior. The same audio drops Pell-Gregory IIB mesioangular as Pell-Gregory IIB mesioangular every time, because that is the most common interpretation in the training distribution it is anchored to.

The schema problem

This one is more subtle. A general medical scribe produces a SOAP note, because SOAP is the universal medical-note shape. OMFS notes are not SOAP.

A real third-molar consult note has these sections:

Chief complaint and HPI
Medical history and ASA classification
Airway assessment (Mallampati)
Per-tooth clinical findings (with Universal tooth numbers, impaction type, Pell-Gregory class)
Radiographic exam
Risks discussed and consent
Treatment plan with specific CDT codes
Anesthesia plan

A general scribe will give you a single freeform “Plan” section with “extract teeth, follow up in 1 week.” That is not wrong, but it is also not documentation that survives a payer audit or a legal review.

An oral surgery scribe generates the per-tooth schema directly: each tooth gets its own findings block, its own classification, its own CDT code, its own ICD-10 diagnosis. The note is not just “more detailed”. It is a different shape.

The billing problem

This is where the cost shows up on the P&L.

A general medical scribe will not assign per-tooth CDT codes. At best it gives you a generic “extraction” line. The biller spends 2–4 minutes per patient mapping that line to the actual code set:

D7140 (erupted, simple)
D7210 (surgical, requires elevation/sectioning)
D7220 (soft-tissue impaction)
D7230 (partial bony impaction)
D7240 (complete bony impaction)
D7241 (complete bony with unusual complications)

For a six-quad case the biller is doing this six times. Across a 20-patient day, that is 40 minutes of billing labor that should not exist.

A maxillofacial surgery documentation tool that knows the CDT code map keyed to Universal tooth numbers eliminates that step. Scribbix runs a deterministic billing engine against the per-tooth findings, with no LLM hallucination in the layer that decides what gets billed. Mutual-exclusion rules (D7240 and D7241 cannot both apply) are enforced at the schema level.

Where the time actually goes

We have not published a controlled study, and we are not going to dress up our internal notes as one. What we can describe is the pattern we see when a practice moves off a general scribe.

With a general scribe, the surgeon spends ten to fifteen minutes per note restructuring sections, adding OMFS-specific consent language, and correcting tooth-level detail. There is no per-tooth CDT output at all, so the biller redoes that work downstream. With an OMFS-specific scribe, the per-tooth schema and the deterministic billing engine produce that structure on the first pass, so review is a read-through rather than a rewrite.

Across a practice, that is the difference between charting into the evening and finishing at the end of the block. The sitewide number we use for it is 2 hours per day, 10+ hours per week, 40 hours per month. At $300/hr surgeon time, that is the kind of math where the surgical AI assistant pays for itself in the first week.

What you should ask any scribe vendor

Before you sign with anyone marketing themselves as a medical scribe for your OMFS practice, ask:

What percentage of your training data is oral and maxillofacial surgery?
Does your output schema include per-tooth findings with Universal numbering?
Do you assign CDT codes deterministically, or via the LLM?
Do you handle Pell-Gregory, Winter, ASA, Mallampati, Wilkes, LeFort classifications natively?
Will you commit to the OMFS specialty, or are we one of fifteen?

If they hedge on any of those, you are evaluating a general scribe with OMFS marketing copy.

Scribbix is the first AI scribe built only for OMFS. We do not have a primary care mode. We will not have a cardiology mode. The whole point is to specialize.

If you are evaluating an AI scribe for oral maxillofacial surgery, we would love to put Scribbix in front of you. Book a demo or read the pricing page.