CPT 90834: the code everything defaults to
90834 is the most-billed outpatient psychotherapy code in the United States, and much of that is genuine — the 45-minute session is the backbone of outpatient practice. But some of it is drift, in both directions, and the drift is worth understanding because it costs money in one direction and invites recoupment in the other.
The rule first, since it is short. 90834 is individual psychotherapy with a stated time of 45 minutes, billable for sessions running 38 to 52 minutes of face-to-face time. Below 38 minutes you are in 90832 territory; at 53 minutes you have reached 90837.
Why the 45-minute hour exists
The "therapeutic hour" of 45 to 50 minutes is a scheduling artefact as much as a clinical one. It lets a clinician run sessions on the hour with time to write, reset and manage the transition between clients. Fifty-minute sessions with a ten-minute gap fit a calendar; sixty-minute sessions with adequate documentation time do not, unless you schedule at 75-minute intervals and accept fewer sessions per day.
The consequence is that a very large share of outpatient work genuinely lands in the 38–52 band, and 90834 being the most common code is not in itself suspicious. Payer analytics do not flag 90834 the way they flag 90837, precisely because the distribution is expected.
Failure mode one: quiet upward drift
A practice schedules 50-minute sessions. Sessions frequently run a few minutes over. Someone notices that a session running to 53 minutes qualifies for 90837, which pays more. Over a few months, "ran a bit over" becomes the norm, and the documented duration on every note becomes 53 or 55 minutes.
Nothing here is fraudulent in intent, and each individual session may be accurately recorded. The problem is the aggregate: a caseload where every session crosses the threshold by two minutes is a distribution no clinician actually produces. Real sessions vary. A note pattern that does not vary is the pattern reviewers look for, and it is hard to defend at audit even when each individual session was honestly recorded.
The practical test. Look at your own distribution of documented session durations over the last six months. If it is bimodal — a cluster at 45–50 and another at exactly 53–55, with nothing between — that is worth understanding before someone else asks about it. Genuine clinical variation produces a spread, not two spikes.
Failure mode two: defaulting down
The opposite error is more common and almost never discussed, because it costs the clinician rather than the payer. A practice that has heard 90837 "triggers audits" sets 90834 as the template default and stops recording actual durations. Sessions that genuinely ran 55 or 60 minutes get billed at 90834 because that is what the template says.
Across a full caseload this is a substantial, permanent revenue loss for work that was actually performed and documented. It is also, technically, inaccurate coding — under-coding is still miscoding, and in some contexts it raises its own questions about whether the record reflects the service.
The correct answer to audit anxiety is not to under-bill. It is to record actual session duration on every note, so that whichever code applies is defensible.
What the note needs
90834 is scrutinised less than 90837, but the documentation standard is identical. There is no lighter-touch requirement for the more common code.
- Actual duration, or start and stop times. Not a template constant.
- A current diagnosis supporting medical necessity.
- The treatment plan goal the session advanced, named explicitly. This is the link most often missing — see the golden thread.
- A described intervention, not a summary of what the client talked about. "Explored feelings about work" is content; "used behavioural activation to schedule two mastery activities before next session" is an intervention.
- Response and plan — how the client responded and what happens next.
Adjacent codes and when they displace 90834
- 90832 — the session ran 16–37 minutes.
- 90837 — the session ran 53 minutes or longer.
- 90847 — a family member participated substantively. A partner joining for the last five minutes does not usually convert the session; a genuinely conjoint session does.
- 90791 — this was the diagnostic evaluation, not a treatment session.
90839— this was a crisis presentation requiring urgent assessment, not a scheduled session that became difficult.
Frequency
Weekly 90834 is ordinary and rarely questioned early in an episode of care. What draws attention is duration without direction: two years of weekly sessions with an unchanged treatment plan, no outcome measures, and no documented reassessment. The issue there is not the code. It is that the record does not show the treatment going anywhere, which is a medical-necessity question rather than a coding one.
A periodic documented reassessment — even a short one, tied to an outcome measure — resolves this almost entirely, and it is far cheaper to write at the time than to reconstruct later.
What Weft does with this
Weft records actual session duration rather than defaulting it, and proposes the code that duration supports. If your documented durations start clustering unnaturally, that is visible in your own reporting before it is visible to a payer's analytics. And because the note is written against the active treatment plan, the goal link that most 90834 notes are missing is present by construction rather than by discipline.
Scheduling design determines your code distribution
An underrated point: your calendar decides your coding more than your clinical judgement does. A practice scheduling 50-minute slots produces a 90834-dominant distribution. One scheduling 60-minute slots produces a 90837-dominant one. Neither is wrong, but the choice should be deliberate and defensible on clinical grounds rather than emerging from whatever the EHR defaulted to at setup.
If your sessions genuinely run an hour, schedule an hour and bill accordingly. If they run fifty minutes, schedule fifty and stop worrying about the threshold. The indefensible position is scheduling fifty and billing as though you scheduled sixty.
If you have already drifted
Practices that recognise the pattern in their own data usually ask the same question: what now?
The answer is not to retroactively amend signed notes, which creates a worse problem than the one it solves. It is to change the process going forward — record actual durations, let the distribution normalise, and be able to show when the change happened and why. A documented correction point is a materially better position than an unbroken pattern, and practices that self-correct are in a different posture than practices that are corrected.
If the exposure is large or spans years, that is a conversation with a healthcare attorney and possibly a voluntary disclosure question, not something to resolve from a reference page.
Distribution across a group practice
In group practice the relevant unit of analysis is not the individual clinician but the practice, because payer analytics typically aggregate by tax ID. One clinician doing intensive trauma work with a legitimately 90837-heavy caseload is unremarkable. The same distribution across every clinician in a twelve-person practice is a practice-level pattern.
This cuts both ways for owners. It is worth knowing your practice distribution, and worth understanding the clinical reason when a clinician sits well outside it — which is usually caseload composition rather than coding behaviour. Asking is more productive than assuming, and the answer belongs in a supervision record either way.
Verified 29 July 2026 against AMA CPT descriptors and CMS published guidance. Code definitions are national; coverage, reimbursement and documentation expectations are set by individual payers and vary by plan, state and contract. Primary references: AMA CPT; CMS Physician Fee Schedule; APA Services. This page is billing reference, not legal or coding advice.