Most AI scribes were designed around a fifteen-minute outpatient visit: one patient, one room, a beginning and an end, then the note. That's a fine model of a clinic. It is not a model of an emergency department, and EM physicians who trial clinic-shaped scribes discover the mismatch within one shift. Here's what the ED actually demands, written by people who work there.
The encounter is not an interval
An ED "visit" is a set of fragments spread across hours: ninety seconds at triage, four minutes at the bedside, a hallway update from the nurse, a callback from radiology, a family conversation, a re-examination after the fluids. A scribe built for continuous ambient capture of a closed-door visit has no natural way to represent this. The tool has to let you return to an encounter — add the re-exam, capture the consultant call, append the disposition conversation — and assemble one coherent note from a shift's worth of fragments, in whatever order they arrived.
Four charts open at once
Clinic is serial; the ED is concurrent. Whatever the software's data model, the workflow reality is: you're documenting the chest pain while the ankle films come back and bed 12 starts vomiting. A usable ED scribe has to make switching between active charts instant and safe — the classic failure is dictation intended for one patient landing in another patient's note, which is not an inconvenience but an incident.
The sickest patients generate the least documentation
The resuscitation that consumed forty minutes of your undivided attention produces, too often, three lines written from memory an hour later — while the ankle sprain gets a tidy complete note. This is exactly backwards from both a clinical and a billing standpoint, and it's structural: nobody types during a code. Capture that works hands-free (a watch on your wrist, a phone in a pocket) and reconstruction that works after the event, from what was actually said in the room, is worth more in the ED than any amount of polish on routine notes.
The money leaks in specific, findable places
EM documentation leaves revenue behind in patterns every director knows: critical-care time that was delivered but never attested with the required elements; medical decision-making that supported a higher level than the note's thin MDM section could defend; comorbidities managed but not named. And the environment has gotten less forgiving — payers have moved toward automatic downcoding of high-level claims whose documentation doesn't clearly support them (a policy shift covered in an npj Digital Medicine brief on AI scribes and coding).
That same brief flags the opposite risk: AI scribes that infer billing levels can drift toward upcoding, which is a compliance problem wearing a revenue costume. The defensible position for EM is deterministic — rules applied to what the note actually documents, so the level is supported by the chart rather than suggested by a model. Flag what was documented but not captured; never inflate what wasn't done.
What we built, because this is where we work
tebIQ's founding use case is the ED — it's built by an ER physician who uses it on shift. Concretely, for EM: an emergency note family designed around fragmented encounters; a full ED note with dedicated sections rather than a clinic SOAP wearing a trauma sticker; hands-free capture including Apple Watch and Wear OS; local audio recovery so a crashed phone doesn't erase a resuscitation narrative; critical-care time capture with a propose-and-confirm attestation flow; a sepsis screen run on every chart; and rules-based E/M leveling of the deterministic kind described above.
None of that makes ED documentation pleasant. The bar we actually aim for is narrower and more honest: the sick patient's chart should be the best note of your shift instead of the worst, and it shouldn't cost you your drive home to make it so.
Built on shift, for shifts. See the emergency note at tebiq.com.