AI Scribe for Ophthalmology: The Structured-Data Gap

An ophthalmology scribe can write the conversation. It cannot fill the rest of the eye chart by magic. The history, the ophthalmologist’s spoken findings, the assessment, and the plan can become a draft. Visual acuity, refraction, pressure, drawings, OCT, fields, and imaging still belong in the systems that measured or stored them.

That dividing line matters more here than in a mostly narrative specialty. Buy a scribe for the words. Keep the ophthalmology record for the numbers and pictures.

Key takeaways

  • A scribe drafts only what it hears; it does not inspect the eye or read a diagnostic image.
  • Laterality and device measurements need a deliberate spoken-and-review step if they appear in the note.
  • An eye-specific EHR remains the better home for refraction grids, drawings, image links, and longitudinal test data.
  • The signing clinician checks the draft. CMS treats that signature as the practitioner’s authentication of the record.

The eye chart has two different jobs

The first job is narrative. Why did the patient come in? What changed since the last visit? What did the ophthalmologist see, conclude, discuss, and plan? An ambient scribe is useful here because those facts pass through the room as speech.

The second job is structured and visual. The American Academy of Ophthalmology’s data-standards work identifies visual acuity, intraocular pressure, pachymetry, visual-field results, and retinal nerve fiber layer measurements as distinct observations that an EHR may need to exchange or track (AAO EyeWiki). An OCT image is not a sentence. Neither is a refraction grid.

Mixing those jobs creates bad expectations. A tool can produce a fluent note while leaving the measurements stranded somewhere else. Worse, it can make a dictated number look authoritative even when the clinician misspoke. The useful question is not, “Does the scribe know ophthalmology?” It is, “Which parts of this visit are actually spoken?”

A spoken-versus-structured map

Use this map before a demo. It stops a polished sample note from hiding a poor fit.

Visit materialA standalone scribe can draft it when spokenKeep it in the eye system
Symptom history and interval changeYesPrior values still need chart review
Ophthalmologist’s narrated examYes, then verify lateralityStructured exam fields and drawings
Visual acuity, refraction, IOPOnly when someone says each valueInstrument or technician-entered fields
OCT, fundus image, visual fieldOnly the clinician’s spoken interpretationOriginal image, report, and serial comparison
Assessment and counselingYesFinal clinical judgment remains with the clinician
Medication discussionA prescription draft can reflect the discussionOrdering, safety checks, and transmission stay in the clinical workflow

This is the original artifact worth carrying into procurement: circle every row that currently costs the ophthalmologist time. If most circles land in the right column, buy or fix the eye system first. If they land in the left column, a scribe trial is sensible.

Laterality deserves its own test

Eye notes punish casual review. OD and OS are tiny strings with a large clinical difference. A pressure of 18 in the right eye and 28 in the left eye cannot trade places because the prose around them sounds plausible.

The AAO’s 2026 registry specifications assume that laterality is recorded in the medical record, and its data standards treat right-eye and left-eye measurements as separate observations (AAO EyeWiki). For a scribe trial, build a simple laterality check into every note: compare each dictated side and value against the technician sheet or device report before signing.

Don’t ask the scribe to infer the side from context. Say it. “Right eye pressure 18, left eye 28” is safer input than “pressure is 18 and 28.” The draft still needs checking, but the source sentence is no longer ambiguous.

Patient Square sits beside the ophthalmology EHR

Patient Square is an AI clinical platform. Practice Copilot brings the whole practice under one AI copilot: an ambient AI Medical Scribe that hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft minutes after the visit, plus a bundled AI EHR, scheduling, and messaging as you move up the plan. Hospitals get Hospital Copilot.

For a US ophthalmology practice, the conservative setup is to keep the eye-specific EHR and run the scribe beside it. There is no claim here of an OCT feed, autorefractor connection, image interpretation, or named-EHR write-back. Review the draft, then move the finished narrative into the record yourself.

Practice Copilot does offer a bundled AI EHR from the Copilot tier, according to the published US plans. We don’t present it as an ophthalmology system. A clinic that needs structured refraction, imaging links, or a drawing surface should preserve the specialty software that already does those jobs.

The scribe module returns ICD-10 suggestions and a prescription draft. It does not choose CPT or E/M codes, submit a charge, send a prescription, or decide what an image means. Those boundaries are not missing features tucked into a future promise. They are the line between drafting and practicing medicine.

The sign-off is the control

CMS tells practitioners who use a scribe to sign the entry and thereby authenticate the document and the care provided; the scribe does not need to sign or date it (CMS MLN). AMA policy is equally plain that AI-generated medical-record content requires the physician’s consent and final review (AMA).

In ophthalmology, review should be mechanical before it becomes editorial. Check patient, eye, measurement, source, and plan. Five boxes. Only then fix the prose.

Review boxOne fast check
PatientCorrect encounter and speaker
EyeOD, OS, or both matches what was examined
MeasurementNumber and unit match the source
SourceDictated finding is not presented as a device import
PlanOrders and prescriptions reflect the clinician’s decision

We think this is a better trial score than asking whether the note “looks good.” A handsome note with one swapped eye is a failed note.

A four-visit trial exposes the fit

Run a short follow-up first. Then use a new consult, a visit built around test review, and a procedure discussion. Those four shapes reveal whether the scribe handles brevity, a long history, dense measurements, and consent language without flattening them into the same SOAP block.

For each visit, count corrections rather than trusting an impression. Mark laterality errors, omitted numbers, invented facts, and minutes spent editing. Compare those minutes with the time the same clinician normally spends writing. No vendor benchmark can replace that local result; the AMA’s 2026 evaluation guide likewise puts clinical use, risk, effectiveness, workflow, and monitoring at the center of an AI assessment (AMA).

Security belongs in the same trial. Patient Square processes audio in memory and discards it once the draft is ready. Notes are encrypted in transit and at rest, BAAs are available, and the SOC 2 Type II audit is in progress, as documented on the security page. Ask every vendor the same questions and put the answers in writing.

If narrative charting is the burden,

Book a demo for US clinics

Prefer a separate page? Open booking in a new tab.

and bring one de-identified visit shape from each of the four categories. If device data is the burden, spend that hour with the ophthalmology-EHR team instead.

FAQ

Common questions

Can an AI scribe document an ophthalmology visit?

Yes, for the spoken part. It can draft the history, the ophthalmologist's narrated findings, the assessment, and the plan. It cannot inspect the eye, read an OCT or visual-field report, or collect measurements from an instrument. Those values enter the draft only when the clinician says them aloud.

Can Patient Square pull data from an OCT, autorefractor, or tonometer?

No. The scribe has no device feed and makes no imaging interpretation. You may speak a value or your reading of a report, then verify the resulting draft. Keep an ophthalmology EHR or device system when structured measurements and image links are part of the required workflow.

Does an ophthalmology practice still need a specialty EHR?

Often, yes. A scribe handles narrative documentation, while an ophthalmology EHR may hold refraction fields, laterality, drawings, image links, and longitudinal measurements. A practice that relies on those fields should keep its specialty system and judge the scribe as a companion, not a replacement.

Who is responsible for checking an AI-generated eye note?

The signing clinician is. CMS says a treating practitioner authenticates a scribed entry by signing it, and AMA policy calls for final physician review of AI-generated medical-record content. Check laterality, measurements, medication instructions, and the assessment before the note enters the chart.

Does the scribe suggest ophthalmology billing codes?

It can return ICD-10 suggestions from the assessment you stated. They remain suggestions. It does not choose CPT or E/M codes, determine medical necessity, post a charge, or read test results to infer a diagnosis. The practice keeps its existing coding and claim workflow.

How should an ophthalmologist test a scribe?

Use four real visit shapes: a short follow-up, a new consult, a test-review visit, and a procedure discussion. Compare the draft with the chart and count laterality errors, missing measurements, invented findings, and minutes spent editing. Keep the tool only if review is faster than writing.

Sources

  1. American Academy of Ophthalmology EyeWiki: glaucoma data standards for visual acuity, intraocular pressure, visual fields, and OCT measurements (fetched September 2026).
  2. CMS MLN: Complying with Medicare Signature Requirements, including entries prepared by scribes (April 2024).
  3. American Medical Association: AI systems that create medical-record content require physician consent and final review (fetched September 2026).
  4. Patient Square: US pricing and included product scope (fetched September 2026).
  5. Patient Square: US security, BAA, encryption, audio handling, and SOC 2 status (fetched September 2026).