Ambient AI Scribe Versus Typing Directly in the EHR

What typing actually costs during a visit

Typing in the EHR is not free just because there is no vendor invoice. It costs eye contact, it costs the extra question you did not ask, and it costs the minutes after clinic when you finish what the room did not allow. Some clinicians type so fluently that the patient barely notices. Many do not. They hunt for the right box, click a required field, and look up to find the patient halfway through a sentence they missed. If that is your day, the competitor to ambient capture is not a romantic paper chart. It is divided attention.

Be honest about where the typing happens. If you already finish most of the note in the room, ambient tools have to beat a workflow you have already optimized. If you batch chart at night, the comparison is different: can a draft from the conversation remove that second shift? An AI medical scribe is worth the most when the visit already contains the information and the keyboard is only a delay. It is worth less when you currently document from memory two hours later and never said the findings out loud. The tool cannot type an exam you only thought.

Ambient capture as a different kind of attention

Ambient capture asks you to talk like a clinician who intends to be understood, not like a person clicking boxes. You still examine. You still decide. You narrate enough that the draft has raw material. That can feel theatrical for a week. Then it becomes a habit, the way dictating once felt theatrical. The gain is that the patient's story can be told to you instead of to the side of your face. The risk is that you stop looking at the problem list because you trust the draft to assemble it. Trust is not a workflow. Review is.

Ambient systems also change the social script. Some patients talk more when the keyboard is gone. Some glance at the phone and ask whether they are being recorded. You need a one-sentence explanation ready, and a way to pause capture. That is not an argument against the tool. Typing has its own social script: the clinician who never looks up. Choose the friction you would rather explain. For groups rolling this out across a pod, put the explanation on a card until it is muscle memory. Awkward consent is still better than silent recording.

The edit pass is the real work

Neither typing nor ambient capture eliminates editing. Typing edits as you go, which hides the time inside the visit. Ambient capture moves editing to a two-minute draft you read before you sign. Sunrise aims for that two-minute turnaround on typical clinic visits so the pass can happen before the next knock. If your edits take twenty minutes because the template is wrong, you did not fail at speaking. You failed at setup. Fix the template before you declare that typing was faster. A bad draft of a bad template is not a fair trial.

Measure edits with a simple score for two weeks. Count visits where you only tweaked the plan, visits where you rewrote the history, and visits where you discarded the draft. If most fall in the first bucket, ambient capture is doing its job. If most fall in the third, stop the rollout and look at audio, specialty fit, and whether you are verbalizing the exam. Typing will win on those days because a blank screen does not invent a review of systems. An honest loss is useful. A vague feeling that the AI is 'off' is not.

Hybrid days are more common than pure ones

Real clinics mix methods. You may type orders because that is how your EHR wants them, speak the narrative, and still click through health-maintenance alerts. Ambient capture does not have to own every keystroke to be worth running. It has to own the story. Let the EHR keep structured data entry if that is required for quality programs. Let the scribe draft the prose. Fighting the EHR on discrete fields is a different project from documentation burden, and mixing those fights makes both look impossible.

Delivery still has to land in the encounter you already have open. If staff paste a draft into the wrong visit, you will prefer typing because at least the cursor was in the right chart. This is an operations problem, not a model problem. Use the same encounter-matching rules you would use for transcription. EHR integrations should specify identifiers, document type, and what happens to an addendum after sign. If your vendor cannot say, keep the draft in a holding queue that a trained person files. Do not invent a paste ritual in the hallway.

What to measure before you switch the whole panel

Run ambient capture on a subset of visit types for ten clinic days. Exclude the visit types you already know are visual or silent. Track time to signed note, after-hours EHR time if you have it, and patient comments only if they are volunteered. Do not survey people into a statistic you then treat as evidence. Compare against the same clinicians' typing weeks, not against a national average you cannot verify. If signed-note time drops and the discarded-draft rate stays low, expand. If not, keep typing for those clinics without a morality play.

Cost belongs in the same worksheet. A seat license that idle partners still pay for can erase the time you saved. Per-line transcription may still be the right companion for letters and procedures while ambient handles progress notes. Look at services as a mix, not a conversion campaign. Keep the signed BAA and U.S.-based processing requirements next to the time study so compliance is not a later surprise. Switching the whole panel because a demo was smooth is how you buy a second documentation burden and call it innovation.