Why unlimited AI notes is a pricing decision, not a feature
Per-note fees and monthly caps do not just cost money. They insert a purchasing decision into the moment a clinician is deciding how thoroughly to document a session, which is the last place a purchasing decision belongs. The interesting question is why vendors meter, and what choosing not to says about the underlying cost.
When an EHR meters AI documentation (a fee per note, a monthly allowance, a credit bundle that runs out in the third week) it has not made a feature decision. It has made a pricing decision, and the pricing decision changes clinical behaviour, because a clinician facing a per-unit charge for a note will eventually decide that some notes are not worth the charge. Documentation is not the part of the workflow where you want the user asking whether this one counts.
What "unlimited AI notes" actually means
Three models are in circulation and they are not variations on a theme. A per-note fee bills each generated note, sometimes each regeneration. A capped plan includes a fixed number per month, commonly a number close to the median customer's volume, and charges overage or blocks above it. An unmetered plan includes documentation in the subscription regardless of volume. Several competitors meter AI notes per month; the specifics and the dates we verified them are on the Mentalyc and Upheal comparison pages rather than restated here, because vendor pricing moves and a number in a blog post ages badly.
The difference between the three is where the decision sits. Under an unmetered plan the decision "should I document this session properly" was made once, at purchase, by whoever signed up. Under a metered plan it is made again every session, by the clinician, at 7pm, after the last client.
What happens when you meter a compliance tool
Consider the arithmetic for a clinician seeing 25 sessions a week across 46 working weeks: 1,150 sessions a year, roughly 96 a month. A per-note fee of $1 is $96 a month, more than most base subscriptions. At $1.50 it is $144. A cap of 40 notes a month covers 42% of that caseload, which means the remaining 58% either costs extra or gets written by hand at the end of the day. These are illustrative rates, not any particular vendor's, and the point does not depend on the exact figure. Any per-unit price above zero produces the same behaviour at the margin.
The behaviour is rationing, and rationing documentation has a specific failure mode. It does not distribute evenly. Clinicians do not skip a random 58% of notes. They skip the ones that feel routine: the stable client, the fourth session of an uneventful course of CBT, the check-in that ran short. Those are precisely the notes that look thin in a records request two years later, because "nothing much happened" is not a defensible reason for a chart to be missing an intervention and a plan goal.
The asymmetry. The note you skip is cheap today and expensive later, and the gap between the two is enormous. A skipped or thin note saves a dollar and a few minutes. The same note, pulled in a post-payment review, can invalidate the claim it supports, and under extrapolation, claims like it. Clawbacks covers the mechanism.
There is a second-order effect that matters more in group practices. If AI notes are metered per clinician, the practice owner now has a per-clinician variable cost tied to how thoroughly their staff document. That is an incentive structure nobody would design on purpose, and it is the direct product of the pricing model rather than of anyone's intentions.
Why vendors meter AI notes
Not out of malice, and not entirely without reason. Three things are going on, and they are worth separating because only one of them is about cost.
Inference costs money. Transcribing an hour of audio and generating a structured note has a real marginal cost: compute, storage, sometimes a third-party model API. The cost is real but small relative to a monthly software subscription, and it has fallen steadily. A vendor charging per note today is charging on a cost basis that was more defensible two years ago than it is now.
Metering is expansion revenue. This is the real reason and it is a finance reason, not a product one. A flat subscription grows only when you add customers or raise prices, and raising prices on existing customers is visible and unpopular. A usage-linked fee grows automatically as customers grow, without a renegotiation, without a renewal conversation, without anyone approving anything. In a subscription business that is an extremely attractive property, and it explains why usage pricing spread through software long before AI existed.
Caps are price discrimination. Setting an included allowance near the median customer's volume means most customers never hit it and feel they are on an all-inclusive plan, while high-volume customers, who are typically larger and less price-sensitive, pay more. That is a textbook design, it works, and it is why caps cluster at oddly specific numbers rather than round ones.
None of this is hidden. It is simply not framed as a pricing decision when it is presented to clinicians, who tend to read a cap as a technical limitation rather than as a revenue instrument.
What unmetered pricing implies about marginal cost
Here is the part that is genuinely informative to a buyer. A vendor cannot offer unmetered documentation unless two things are true.
First, marginal cost per note has to be far below the average revenue per user, with enough headroom to absorb the tail. Volume distributions in behavioral health are skewed: a small number of high-throughput practices generate many times the notes of a median solo clinician. An unmetered vendor eats that tail. If their per-note cost were anywhere near their subscription price, the tail would be uneconomic and they would cap it. Offering unlimited is a public statement about a private cost structure.
Second, incentives have to line up on note quality. Under a per-note fee, a regenerated note is revenue. Under unmetered pricing, a regenerated note is pure cost, which means the vendor is financially motivated to get the first draft right, to transcribe accurately, and to stop producing output that clinicians rewrite. That is a small thing that compounds. It is also the only version of the arrangement where the vendor's interests and the clinician's interests point the same direction.
The honest version of the unmetered pitch is not "this is free". It is: we set the subscription price assuming you use it for every session, we would rather you did, and we have priced the tail in. Weft charges $39 a month on Starter and $69 on Pro, unlimited AI notes on both, and a BAA on every plan including the trial. The full breakdown is on pricing and how the documentation itself works is on notes.
The strongest argument for metering
Steelmanning it: metering is more honest about cost allocation. A practice generating 400 notes a month genuinely imposes more cost than one generating 40, and a flat price means the small practice subsidises the large one. In enterprise software that argument usually wins, which is why almost everything infrastructural is usage-priced.
It loses here for a specific reason: the unit being metered is not discretionary. Metering API calls is fine because a developer can choose to make fewer of them. Metering notes is different, because the clinician cannot choose to see fewer clients and cannot legitimately choose not to document the ones they saw. When the metered unit is mandatory, usage pricing stops being a fair allocation of cost and becomes a tax on doing the required thing. The subsidy objection is real; it is just smaller than the behavioural cost of putting a price tag on a compliance obligation.
What to check before you believe an unlimited claim
- Is "unlimited" per account or per clinician? In a group practice these are very different numbers.
- Do regenerations count? A metered system that counts generations rather than sessions charges you twice for the note you were not happy with the first time.
- Is there a fair-use clause? Look for a stated threshold, a throttle, or language about "reasonable use" that gives the vendor discretion. Unlimited with an undefined ceiling is a cap you have not been shown.
- Is transcription included, or metered separately? Some plans include the note and bill the audio minutes.
- Does the plan include a BAA at that tier? Ambient documentation touches PHI at the moment of recording. A vendor processing session audio is a business associate, full stop. See HHS on HIPAA and our BAA page.
The part that is not about pricing
None of this touches the separate question of whether clients know a tool is listening. Consent for ambient documentation is a clinical and legal matter with state recording law underneath it, and it does not become simpler because the notes are unlimited. What to say, when to say it, and how to document that you said it are on AI scribe consent. Substance-use programmes have an additional confidentiality layer under 42 CFR Part 2, published via SAMHSA.
Pricing determines whether a clinician uses the tool. Consent determines whether they should. Both have to be right, and the first one is easier to fix: it is a number in a table, and somebody chose it.
Common questions
Does unlimited AI notes mean there is no cost to the vendor?
No. Transcription and generation have a real marginal cost. Unlimited means the vendor has priced that cost into the subscription and accepted the high-volume tail, which is only viable if per-note cost sits well below average revenue per user. It is a statement about cost structure, not about cost being zero.
Are monthly note caps common?
Common enough to be worth checking before you sign. Several products in this category include an allowance and charge overage or block above it, and allowances tend to sit near the median customer's volume. Our comparison pages list the current published terms with the date each was verified.
What does Weft charge for AI notes?
Nothing beyond the plan price. Unlimited AI notes are included on Starter at $39 a month and Pro at $69 a month, and on the Group plan at $69 plus $49 per clinician. No per-note fee, no monthly allowance, no separate transcription charge. The 30-day trial does not ask for a card.
Do AI-generated notes create extra audit risk?
Not inherently. The standard is the same regardless of how the draft was produced: the note must be accurate, must reflect the session that occurred, and must be reviewed and signed by the treating clinician. The risk that matters is a thin or missing note, which is exactly what metering encourages.