How this study was built

What you are judging, how the notes were prepared, how the comparison stays blind, and who is behind it.

What you are judging

The study has eight cardiology cases. Each case is one visit, and three AI scribes (NovaScribe, Freed and Heidi) each produced a note for that same visit in SOAP format: Subjective, Objective, and Assessment & Plan. The three notes are labeled only A, B and C. You can read them one at a time or, on a wide screen, side by side, and you pick the one you think is best. You can add a short reason if you like.

All eight visits are fictional. No real patient's information is used.

Above each case is a one-line description of the patient and visit type (for example, "64-year-old woman, 6-month cardiology follow-up"). The study team wrote these lines using only details that none of the three notes contradict.

How the notes were prepared

The clinical text is shown as each scribe generated it. Apart from the three changes below, the study team did not edit, shorten or correct any wording:

• Names replaced with fictional stand-ins. In the source notes, names of people and named places or facilities were masked with neutral labels such as [Patient], [Clinician 1] and [Location 1]. So the notes read naturally, each label is shown as a made-up name. The names you see were chosen by the study team, not written by the scribes, and they do not refer to any real person, place or facility. Within a case, a label always gets the same stand-in, in all three notes, and two different labels never share one. Stand-ins for clinicians are plain names with no title, and stand-ins for facilities carry no specialty, so they add nothing the label did not say.
• Product names removed. Any non-clinical text naming the product that wrote a note was removed.
• Formatting standardized. The scribes format their notes differently, and those differences could reveal which product wrote which note. So every note is shown in the same typeface and size, with the same three section headings and the same style of bullet. Formatting symbols such as bold markers were removed. Apart from the stand-in names above, no words were added, removed or reordered.

Dates, ages, punctuation, omissions and any discrepancies between notes were left exactly as generated.

How the comparison stays blind

• Each clinician gets four of the eight cases, with the option to do the other four. Cases are handed out least-assigned first, so each case goes to a similar number of clinicians. Completed reviews per case can still differ if someone stops partway.
• For every clinician and every case, the A, B and C labels are shuffled at random. Note A for one clinician may be note C for the next, so a product's position tells you nothing.
• Your browser does not receive any information about which product wrote which note until you submit your final answer. You can change any answer until then.
• After you finish, we show you which product wrote each note in the cases you reviewed. Because other clinicians may still be taking part, please do not share which product wrote which note.

Who takes part

Participation is by personal invitation link, one link per clinician, and each link can be completed only once. The study team checks for anyone invited twice before counting results. We ask for your credential, specialty, years in practice and the AI scribe you use now, if any, so results can be broken down by group. Some participants are colleagues of Dr. Proctor and some are recruited separately. We record which group each invitation came from, so results can be reported for each group.

How results are reported

Results are reported as the share of picks each product received, with the number of clinicians and the number of case votes behind every figure. We also report a per-clinician view, showing which product each clinician picked most often. This is the more conservative measure, because one clinician's votes across several cases are not independent of each other.

Your information

We store the profile details you enter, your picks and any reasons you write. Results may be published in aggregate. We will not publish your name, or attribute a comment to you, without your separate permission. Questions about the study can go to support@novascribeai.com.

Source material: "Blinded Clinical Note Comparison", 8 cases with 3 notes each, prepared 19 September 2026.