What should an AI scribe capture in a neurology visit?

It should draft the history and the clinician's spoken findings without merging old, patient-reported, and current observations. The exact note structure belongs to the neurologist. Test chronology, attribution, laterality, relevant negatives, medication changes, and whether an unspoken exam element stays absent.

Can an AI scribe perform a neurological examination?

No. A scribe can record findings that the clinician states, but it does not perform the examination or decide what an observation means. NINDS describes the neurological examination as a clinician-led assessment of several functions. The clinician still examines, interprets, corrects, and signs.

Can Patient Square interpret MRI, EEG, or EMG results?

Patient Square does not claim image interpretation, EEG interpretation, EMG interpretation, or diagnostic decision support. If a clinician discusses a verified source result during the visit, the scribe may draft that discussion. The reviewer must keep the source, date, and interpretation correctly attributed.

How do you test chronology in an AI-generated neurology note?

Create a synthetic case with three timepoints: an old event, an interval change, and today's finding. Mention them out of order during the encounter. The draft should restore a readable timeline without changing dates, moving old findings into today, or converting patient history into clinician observation.

Does Patient Square replace a neurology EHR?

Not when the practice needs neurology-specific testing, device, imaging, registry, or structured specialty workflows that Patient Square does not claim. Practice Copilot can work alongside an existing EHR, or a practice can use the bundled Patient Square EHR when the confirmed requirements fit.

AI Scribe for Neurology: Preserve the Timeline

Neurology notes break when timepoints blur. A symptom from last year, a patient-reported change last week, and today’s examination cannot become one timeless paragraph. Test an AI scribe on chronology first. Then test source attribution, laterality, and whether the system leaves an unspoken finding blank.

Build the note around three timepoints

Start a pilot with one synthetic case, not twenty real patients. Give it three facts:

Old: an event documented before today’s visit.
Interval: a change the patient or caregiver reports since the last review.
Today: a finding the clinician observes and states now.

Mention those facts out of order. The draft should produce a readable timeline without silently upgrading an old or reported fact into today’s examination.

Use a small provenance ledger:

Draft element	Timepoint	Source label	Reviewer decision
Prior event	Exact source date	Prior record or patient report	Keep, qualify, or remove
Interval change	Stated period	Patient, caregiver, or source record	Confirm attribution
Current finding	Today	Clinician-spoken examination	Confirm wording and laterality
Test result	Source date	Named report	Confirm source and interpretation
Plan	Today	Clinician decision	Confirm it is a plan, not a completed action

The ledger is not a clinical template. It is a way to catch time travel in generated prose.

Spoken findings are not a performed examination

NINDS explains that a neurological examination may assess movement, sensation, hearing and speech, vision, coordination, balance, and aspects of mental status. It also states that the examination and history help the clinician decide what further testing may be needed (NINDS diagnostic tests and procedures).

An ambient scribe does not perform that work. It can draft the findings a clinician says aloud. If the clinician never states a reflex, gait finding, score, or relevant negative, the system should not supply one from a familiar template. Blank is safer than invented completeness.

The same boundary applies to MRI, EEG, EMG, and outside records. A spoken summary can be documented with the right attribution. The scribe should not claim to have read or interpreted a source it never received.

Try the contradiction round

After the basic timeline test, add one controlled contradiction. The patient says the symptom began in March; a prior record says April. The clinician notes that discrepancy but does not resolve it.

A useful draft preserves both sources and the unresolved state. A bad draft picks one date, averages them into “spring,” or drops the disagreement. That edit may read smoothly, which is exactly why it is dangerous.

Repeat the exercise with laterality and a medication list. Ask the vendor where uncertainty appears, whether reviewers can find changed text, and what the system does when two speakers disagree. HHS explains that covered entities and their business associates have defined HIPAA responsibilities when protected health information is handled (HHS covered entities and business associates). Contract, access, and retention questions belong in the same pilot packet as note quality.

Choose the surrounding record separately

Patient Square is an AI clinical platform. Its AI Medical Scribe drafts a structured SOAP note, ICD-10 suggestions, and a prescription draft for clinician review. Practice Copilot can work alongside an existing EHR or use the bundled Patient Square EHR, named AI Copilot EHR on the pricing page, from the Copilot tier. Annual US pricing is $79 per physician per month for Assist and $119 for Copilot (Patient Square pricing).

Patient Square does not claim MRI, EEG, or EMG interpretation; neurology registries; diagnostic support; procedure coding; or specialty-device integration. A neurology practice that depends on those capabilities should keep the system that owns them. The scribe evaluation can still answer a narrower question: does it reduce drafting work without corrupting chronology?

Our view is blunt. If a vendor demo never includes an old fact, an interval change, and today’s finding in the same visit, the demo is too easy. Pair this timeline test with the data-ownership and export guide and the accuracy evaluation checklist. Book a demo with the contradiction case ready.