What does a 90837 note have to document, and when should it be a 90834?

What has to be in the note?

The clearest published statement of the time requirement comes from a Medicare contractor’s billing article, and it is worth reading in its own words. First Coast Service Options, the Medicare Administrative Contractor for Jurisdiction N, states: “Psychotherapy services (CPT codes 90832-90838) are time-based codes. Start and stop times or total times must be documented for CPT codes 90832, 90834, and 90837.”

Either form satisfies it. A total is enough; you do not need both. What is not enough is a note from which the duration cannot be determined at all, which is the ordinary result of a template that records the date and the modality and moves on.

The same article’s documentation requirements are unglamorous and are the part people skip. The record must be legible, must carry patient identification and the dates of service, must bear the legible signature of the practitioner responsible for the care, and must support the diagnosis codes submitted. And one sentence in it does more work than the rest combined: “The submitted CPT/HCPCS code must describe the service performed.” That is the whole principle. The record says what happened; the code describes the record.

One scope label before you rely on any of that, because it is the kind of thing that gets quoted as national rule and is not. This is one contractor’s policy, covering Florida, Puerto Rico and the Virgin Islands, in the version effective from January 1, 2025. Other Medicare contractors publish their own articles and the related local coverage determination is likewise local. Look up the one for your jurisdiction rather than borrowing this one. We cite it because it states the requirement plainly, not because it governs you.

How is the time counted, and which code does it land on?

These are time-based codes, so the reported code follows the duration of the psychotherapy itself. In broad terms 90832 is the shortest of the three, 90834 is the middle one usually described as the 45-minute session, and 90837 is the longest, usually described as the 60-minute session. The exact minute ranges that make each one reportable are set by the time rule in CPT.

We are not going to reproduce those ranges here, and the reason is worth stating rather than hiding. CPT is copyrighted by the American Medical Association and licensed; the Medicare Coverage Database itself puts the code descriptors behind a licence acceptance. Read the ranges in the current CPT book or in your payer’s licensed materials, and read the current year’s, because that is what governs. Any billing page that hands you the numbers casually is reproducing licensed content, and a page that does that is not a page to trust on the rest of it either.

The practical point does not depend on the exact numbers. The time you are counting is the psychotherapy time, and it has to be a real measurement of a real session rather than the nominal length of a calendar slot. A practice that books in hour blocks and documents every session as an hour has not documented time. It has documented its scheduling software. The same is true in the other direction for a practice whose template auto-fills a fixed duration.

Two adjacent situations have their own rules and are worth knowing exist rather than guessing at. Psychotherapy furnished alongside an evaluation and management service is reported with the add-on codes, and the contractor article above notes that the psychotherapy time in that combination need not be continuous but still has to be documented as its own time. And for a session running to 90 minutes or more without an E/M service, that article addresses when a prolonged service code can be reported. Neither is a place to improvise.

What does the audit evidence actually show?

That the failures are mostly administrative, and that they are more varied than the usual summary of them suggests. The most substantial public look at this is a 2023 HHS Office of Inspector General audit of Medicare psychotherapy payments. Its abstract mentions undocumented time as an example, but the report itself publishes the full breakdown, and it is worth having in front of you rather than the summary.

OIG sorted 128 failing enrollee days by what was wrong with them. The types total 175, because 43 of those days had more than one thing wrong: psychotherapy time not documented, 60; treatment plans incomplete or missing, 43; no psychotherapy provided or the documentation missing, 24; the psychotherapy itself incompletely documented, 20; an incorrect number of services or an incorrect CPT code billed, 16; and incident-to requirements not met, 12. A separate count against Medicare guidance found missing provider signatures 31 times and missing telehealth modifiers 29 times.

Two things follow from that table which do not follow from the summary. Time is the largest single category but it is about a third of the total, so “document the time” is the first fix and not the whole of it.

And the second-largest category is the treatment plan, with a scope label that matters more than the number does. OIG counted that failure only where the claim was processed by one of the three contractors whose policies actually require a treatment plan. A further 35 days had treatment plans missing and were allowable, because the contractor in that jurisdiction did not require one. First Coast, quoted at the top of this page, is one of the four that did not — though not one of those four was indifferent to it: three told OIG they expected a treatment plan anyway, and the fourth said on its own website that psychotherapy documentation should include one.

The time requirement is different in kind. OIG applied it to every sampled day whatever the contractor, because it rests on the statute and on CPT’s own time rule rather than on any one jurisdiction’s policy. The treatment-plan requirement is your contractor’s to impose — which is the same lesson as the rest of this page, arriving from the opposite direction. A treatment plan is in any case a thing you maintain rather than a thing you write at the end of a session.

The headline number from that audit gets quoted a great deal and almost always without its labels, so here it is with them. Medicare Part B paid $1 billion for psychotherapy services from March 2020 through February 2021, and OIG estimated $580 million of that was improper, split $348 million telehealth and $232 million non-telehealth. Now the labels. It is Medicare Part B only and says nothing about commercial payers. The period is the first year of the COVID public health emergency, when telehealth rules and volumes were unlike any year before or since, and most of the estimate is telehealth. And the $580 million is a projection from two stratified random samples of 111 and 105 enrollee days, not a count of anything. It is evidence that time documentation was widely weak in an unusual year. It is not a claim that most psychotherapy billing is wrong, and it says nothing whatever about your practice.

What makes it useful anyway is the direction. An auditor reviewing a psychotherapy claim is not second-guessing the therapy. They are asking whether the record shows what the claim says happened — and on that list, an incorrect number of services or an incorrect code is there too, sixteen times. The record and the claim have to agree in both directions.

Why is this getting more attention now?

Because at least one very large payer has told investors that provider coding intensity is driving its commercial costs. In its second-quarter 2026 published earnings remarks, UnitedHealth Group told investors that in its commercial business it was “not yet seeing evidence of cost trend moderation – in fact, it is the opposite,” and named among the primary drivers “more aggressive billing practices among providers, especially higher service and coding intensity and higher costs per encounter that result from the more fee for service orientation of commercial plans.”

Read that precisely, because it is easy to make it say more than it does. It is one payer, describing its own commercial book, explaining its own cost trend to investors. It is not about mental health specifically. It is not an audit finding. It is not evidence about any individual practice, and nobody should read it as an accusation. What it is, is a very large payer stating on the record that the distribution of service levels across its network is something it is now naming as a cost driver. Providers usually learn that when a records request arrives. Here it is in a public document, months earlier.

The sane response to that is not to change what you bill. It is to make sure the record already supports what you bill, which is what you should be doing anyway. The practices that find this uncomfortable are the ones whose documentation is thin enough that they cannot tell.

What if the note supports a different code than the one billed?

Then the claim and the record disagree, and that is the problem regardless of which direction it runs in. A note that supports less than was billed is the one everyone thinks of. A note that supports more than was billed is the same defect wearing different clothes, and it is the one nobody looks for.

Treat both as documentation accuracy rather than as money. The question is never “which code can this note carry.” It is “what happened in the session, and does the record say so.” If a 70-minute session is documented with no time at all and billed at the shorter code because that felt safer, the record is still wrong and the practice has still made a claim it cannot support. Fixing that means writing down the time — not swapping one habit of coding away from the record for the opposite one.

Where a pattern shows up across a clinician or a practice, the fix is upstream of the claim: the template, the workflow, and whether anyone looks at the note before the claim goes out. That is the same argument as preventing denials rather than appealing them, applied one step earlier, to the accuracy of the record rather than the completeness of the claim.

One thing we will not do on this page, and you should be suspicious of anyone who will: tell you what your distribution of session lengths ought to look like. There is no published benchmark for it that survives contact with a real caseload. A practice doing trauma work and a practice doing brief solution-focused work should not look alike, and a number invented to make them look alike would be worse than no number. The defensible position is that every session is documented for what it was.

Does any of this vary by payer?

Yes, considerably, and that is the part that makes it operationally annoying rather than intellectually hard. The requirement quoted at the top is one Medicare contractor’s. Other Medicare contractors publish their own. Commercial plans set their own documentation and medical-necessity policies in their provider manuals, Medicaid managed care plans add state rules on top, and a behavioral health carve-out administrator may apply different ones again from the medical plan whose card the patient carries.

The practical consequence is that “we document to Medicare’s standard” is a reasonable floor and not a complete answer. What travels across all of them is the principle in the contractor’s own sentence: the code submitted must describe the service performed, and for a time-based code the time is part of the service. A note that carries the time, the necessity and what was done will satisfy most reviewers in most places. A note missing the time satisfies none of them.

This is also why documentation problems show up as denials long after the session. Nothing about a thin note stops a claim from going out. It stops it from surviving a request for records, which arrives weeks or months later, by which point the aging is already working against you — the pattern behind a lot of what sits in an aging report.

Key takeaways

• These are time-based codes. Document start and stop times or a total. Either works; neither is optional.
• The governing principle, in a Medicare contractor’s own words, is that the code submitted must describe the service performed. The record comes first.
• When auditors looked at Medicare psychotherapy claims, the largest single deficiency was time not being documented. Incomplete or missing treatment plans were the next largest, and an incorrect code was on the list too.
• Read the widely quoted $580 million figure with its labels: Medicare Part B only, the first pandemic year, mostly telehealth, and a projection from two stratified random samples of 111 and 105 enrollee days.
• A note supporting less than was billed and a note supporting more than was billed are the same defect. Both mean the claim and the record disagree.
• A scheduling block is not a measurement. An hour slot documented as an hour, every time, is a record of your calendar.
• Requirements vary by contractor, by commercial plan, by state Medicaid and by carve-out administrator. Look up yours rather than borrowing someone else’s.
• The exact minute ranges live in CPT, which is licensed. Read the current year’s book; be wary of pages that hand them over casually.