Telehealth after the PHE: five workflow shifts that stuck

When the federal public health emergency ended, a lot of people expected telehealth in behavioral health to roll back to 2019 levels. It didn't. What rolled back were the edges: cross-state waivers, some controlled-substance flexibilities, parts of the prescribing relief. The core of virtual care stayed.

For behavioral health specifically, the staying power has been striking. Depending on the practice and the modality, 40 to 70% of sessions are still virtual. That's not a transition number. That's a new equilibrium.

Five shifts that came out of the PHE era and are now permanent features of how a behavioral health practice runs:

1. The session schedule is fungible

Before 2020, most practices had in-person slots and telehealth slots on separate days. Now, most have a single schedule with a modality flag per appointment. Clients switch modalities week to week. Intake forms have a "preferred modality" field that frequently means "depends on the week."

Documentation implication: Notes need a modality tag. Simple, but enforced consistently across the chart, because the tag affects the CPT code.

2. Session length is more variable

Virtual sessions run shorter on average than in-person, and the distribution is wider. You see a lot more 30-minute and 45-minute virtual sessions than the pre-PHE norm of 50. At the same time, crisis virtual sessions sometimes run longer than the equivalent in-person session because there's no room to triage in a waiting area.

Documentation implication: The accurate session duration is the one computed from what actually happened, not the scheduled slot. Timestamped transcripts handle this automatically. Scheduled-slot inference doesn't.

3. Documentation happens further from the session

In-person practice has a natural documentation window: the 10 minutes between clients in the office. Telehealth practice often doesn't. The next call connects instantly. Charting pushes into evenings more consistently for telehealth-heavy practices than for mixed ones.

Documentation implication: This is exactly where ambient AI shifts the workflow. The charting window closes. The in-session draft opens.

4. Client intake is async-first

Digital intake forms, consent flows, and assessment inventories have largely replaced the pre-session paper packet. By the time the first session happens, the clinician already has PHQ-9, GAD-7, trauma screeners, and history, often filled out the night before.

Documentation implication: The first session note can reference structured data from the intake rather than repeating it. A good note system should pull that in without the clinician having to retype it.

5. The session is more observable

When the session is on camera, the environment is part of the clinical picture. Is the client in their car? Wearing pajamas at 3 p.m.? Is someone walking through the background? Body language is more visible than over the phone, less visible than in person, and different from either.

Documentation implication: Mental status exam language evolved during COVID and hasn't un-evolved. "Appropriate grooming" in a home environment is a different observation than in an office. A well-trained note model knows this. A naive one writes the same MSE it would have written in 2019.

The thread that runs through all of them

Each of these shifts makes documentation harder without changing what documentation is supposed to do. The note still needs to show what happened, what you thought, and what you decided. The environment it has to capture is just messier now.

That's the case we keep making for ambient AI: not that it's faster, although it is, but that it holds up better under a workflow that has more variability and less rest built in. Write from the session, not from memory at 11 p.m., and the accuracy gap closes on its own.