AI Scribes for Telehealth Therapy: How They Hear the Session and How to Set One Up

In an office, an AI scribe listens through one microphone. On a video call, the client's voice arrives through software, and how the scribe reaches it decides whether the note captures both sides of the session, which vendors need a business associate agreement, and what the client should be told. Here is how the main approaches work, where they fail, and a setup checklist.

Five ways a scribe hears a video session

Built into the telehealth platform. Some practice platforms put the scribe inside their own video room. SimplePractice's Note Taker adds a Record audio button to the telehealth toolbar that starts live transcription, then produces a draft note after the session ^[1]. Upheal lists built-in telehealth alongside its notes ^[2], and JotPsych sells a video add-on called JotMeet ^[3].

Browser tab capture. Freed's Chrome extension records your microphone plus the audio of the browser tab running the call. It works in Chrome or Edge with Zoom on the web, Google Meet, Teams on the web, Doxy.me and browser-based EHR video, but not with the Zoom or Teams desktop apps or Safari ^[4]. Heidi's browser Telehealth mode works the same way through a Share tab audio prompt, also Chrome and Edge only ^[5]. One trap: some EHRs open the visit in a pop-up window, and the capture has to be moved to that window by hand or patient audio is missed ^[4].

Desktop app capturing system audio. Heidi's desktop app for Windows and Mac captures system audio rather than a single tab, and Heidi calls it its most stable telehealth option ^[5]. On Windows it can take audio straight from a chosen app such as Zoom when the patient side goes missing ^[5].

A meeting bot. Freed's Zoom marketplace app joins meetings you host as a separate participant. It only joins meetings hosted from your connected account, and you have to admit it from the waiting room if you use one ^[4].

A second device. As a fallback, Freed and Heidi both suggest playing the call out loud and recording on a phone placed nearby ^[4][5]. This depends on a quiet, private room on your side. HHS expects telehealth to happen in private settings where feasible, and suggests lowered voices and no speakerphone when it is not ^[6].

For which ranked vendors document telehealth capture at all, see the telehealth row in our therapist comparison. SimplePractice users can also start with our SimplePractice guide.

Headphones, echo, and one-sided transcripts

Headphones help but change the setup. Freed reads the client's side from the call tab and still recommends headphones, because otherwise your microphone also picks up the client through your speakers, doubling their voice and causing echo ^[4]. Heidi says browser echo suppression can block parts of the patient's audio, and that if the call runs on a phone with headphones in, it cannot hear the patient at all ^[5].

The most common failure is a note built only from the clinician's half of the session. Heidi lists it as a known issue ^[5], and Freed shows separate level meters for your microphone and the client's audio so you can confirm both are live ^[4]. Check both in the first minute.

Two speakers, one transcript

The scribe also has to decide who said what, a step called diarization. Accuracy research in therapy is thin. A 2020 study of 100 psychotherapy sessions found a 25% word error rate overall and 34% for sentences about harm to self or others, and it did not measure speaker attribution ^[7]. A 2025 study of nine clips from psychiatric sessions found low word error but diarization error from 2% to 32%, higher when speakers were similar in gender or age ^[8]. Neither tested a commercial scribe, and the second sample is small. When reviewing a draft, check who is credited with risk statements, symptoms and plans.

Two vendors, two BAAs

A scribed telehealth session usually involves two vendors. HHS says telehealth providers must use video vendors that comply with HIPAA and will sign a BAA ^[9], and that a tool storing recordings or transcripts in the cloud is more than a conduit and needs its own BAA ^[6]. Video platforms handle this differently. Zoom signs BAAs on paid plans, through online checkout for Pro, and notes that some AI features may not be available to healthcare customers with a BAA ^[10]. Doxy.me offers an individual BAA on its free plan and a clinic-level BAA on paid plans ^[11]. What to look for in the scribe's agreement is in our HIPAA and BAA guide.

Consent when the client is somewhere else

The client may be in a state with stricter recording rules. Our client consent guide covers all-party consent states, interstate calls and what the consent form should say. Two telehealth details to add. Freed describes its Zoom bot as joining your meeting as a separate participant, which you may have to admit from the waiting room, so the client may see it on the call ^[4]. Tab and desktop capture run on your own computer and may not show anything to the client. Either way, tell clients how the session is being captured. And platform vendors leave the law to you: SimplePractice states that knowing your state's requirements for recording telehealth audio is the practice's responsibility ^[1]. This is general information, not legal or medical advice; check your state's rules and your licensing board.

Couples, family, and group sessions

SimplePractice says Note Taker works for one-on-one, couples and group telehealth ^[1]. Upheal lists couples and family note types on paid plans, while its free plan covers individual therapy only ^[2]. More voices means more attribution to check, and the diarization research above found more errors when voices are similar ^[8]. Every participant needs to agree, wherever they join from. See our group practice comparison.

Setup checklist

• Match the capture method to how you run video. Browser call: tab capture works. Desktop Zoom or Teams: you need a desktop app, a bot, or a second device ^[4][5].
• Confirm both BAAs before the first session, one for video and one for the scribe ^[6][9].
• Use headphones where the tool supports them, and know which modes do not ^[4][5].
• Run a test call before clinical use; Heidi recommends it ^[5].
• Check both audio meters at the start of every session ^[4].
• Confirm the client's location and consent each time, per the consent guide.
• Review every draft before signing. SimplePractice requires it and keeps the transcript only until the note is signed or seven days pass, after which it is de-identified ^[1]. Know your vendor's retention window.
• Bring telehealth questions to demos with our buyer's checklist.

Where the named vendors score

Our research scores rate public documentation, not telehealth audio quality: Heidi Health (88/100, #1), Freed (76/100, #3), JotPsych (75/100, #4) and Upheal (71/100, #6). SimplePractice Note Taker is not scored. See the scorecard and methodology.

*This post is based on public vendor documentation, regulator guidance and published research. We do not test products hands-on. Some links may earn a referral fee, which never affects scores or rankings, and we accept no paid placement. General information only, not legal, compliance or medical advice. Corrections: corrections.*

Sources

1. Using Note Taker with telehealth (help article, updated September 9, 2026). SimplePractice Support. https://support.simplepractice.com/hc/en-us/articles/42030852915341-Using-Note-Taker-with-telehealth (opened 2026-10-22)
2. Upheal review: built-in telehealth, Chrome extension, Mac app and Zoom capture; couples and family note types; free plan limited to individual therapy. ClinicalAIWeekly. https://www.clinicalaiweekly.com/reviews/upheal (opened 2026-10-22)
3. JotPsych review: works with any telehealth platform; JotMeet video add-on. ClinicalAIWeekly. https://www.clinicalaiweekly.com/reviews/jotpsych (opened 2026-10-22)
4. Telehealth Visit Capture (Virtual Visit Recording) (help article). Freed Scribe Help Center. https://help.getfreed.ai/en/articles/13901913-telehealth-visit-capture-virtual-visit-recording (opened 2026-10-22)
5. Using Heidi in Telehealth (help article). Heidi Help Center. https://support.heidihealth.com/en/articles/9765598-using-heidi-in-telehealth (opened 2026-10-22)
6. Guidance on How the HIPAA Rules Permit Covered Health Care Providers and Health Plans to Use Remote Communication Technologies for Audio-Only Telehealth (content last reviewed June 23, 2026). U.S. Department of Health and Human Services, Office for Civil Rights. https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/hipaa-audio-telehealth/index.html (opened 2026-10-22)
7. Miner AS, Haque A, et al. "Assessing the accuracy of automatic speech recognition for psychotherapy." npj Digital Medicine (2020). https://www.nature.com/articles/s41746-020-0285-8 (opened 2026-10-22)
8. Zeeshan R, Bogue J, Asghar MN. "Relative Applicability of Diverse Automatic Speech Recognition Platforms for Transcription of Psychiatric Treatment Sessions." IEEE Access (2025). https://doi.org/10.1109/ACCESS.2025.3585454 (abstract opened via R Discovery 2026-10-22)
9. HIPAA Rules for telehealth technology. Telehealth.HHS.gov, U.S. Department of Health and Human Services. https://telehealth.hhs.gov/providers/telehealth-policy/hipaa-for-telehealth-technology (opened 2026-10-22)
10. HIPAA Business Associate Agreement (BAA) (support article). Zoom Support. https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0067751 (opened 2026-10-22)
11. How do I know if I need a BAA? (help article, September 15, 2026). Doxy.me Help Center. https://helpcenter.doxy.me/en/articles/8272944-how-do-i-know-if-i-need-a-baa (opened 2026-10-22)