Free AI in your EHR versus an AI that owns the record

Every major EHR now includes some form of AI at no extra charge. It is already connected to your data, it needs no migration, no new contract, and no second login. For a great many practices that combination is unbeatable, and any post that pretends otherwise is selling something.

So start there honestly: the free thing is often the right answer. What follows is about where its ceiling is, and why that ceiling sits lower for a practice that never files a claim.

What free is actually buying you

The advantages are real and worth stating plainly.

There is no switching cost, which — as anyone who has priced a migration knows — is the largest number in the comparison and the one most often left out. There is no integration work, because the AI is already inside the system holding your data. There is no additional vendor to evaluate, no additional business associate agreement, and no additional attack surface. And the marginal cost is zero, which means trying it costs you an afternoon rather than a budget cycle.

Against that, the honest evidence on what bundled documentation AI delivers is encouraging rather than transformative — and more interesting than the headline suggests. A large multisite study published in JAMA in April 2026, covering 8,581 clinicians across five health systems, found that adopting AI scribes was associated with 16.0 fewer minutes of documentation time per eight scheduled patient hours, 13.4 fewer minutes of total EHR time, and about half an additional visit per week. It also reported that electronic health record time outside work hours did not change significantly.

That last result is the one worth sitting with, because it is the one nobody puts on a slide. The tool did exactly what it promised: writing the note got faster. But documentation time fell by more than total EHR time did, which suggests part of the saving was taken up by other work inside the system rather than leaving the day — and the hours worked in the evening did not move at all.

The study is observational, so it describes an association rather than a guaranteed result for any given practice. But the shape of that result is precisely what you would predict from a tool that can change how a note gets written and cannot change anything else around it. A faster step is not a shorter day.

Thirteen minutes for free is still a good trade. Take it. The question is what happens when you want the day to change rather than the step.

The ceiling is the record underneath

A bundled AI can only act within the workflow its host system already defines.

It can draft text into fields that an older schema laid out. It can suggest a code the billing module was already built to accept. What it cannot do is change what those fields are for, or remove a handoff, or reorganize the sequence of steps the system was designed around fifteen years ago. It is a very capable assistant working inside a building whose walls it did not choose and cannot move.

That is not a criticism of the engineering, which is often excellent. It is a structural limit. If your complaint about your EHR is the typing, a feature that does the typing is precisely the right fix. If your complaint is the workflow — the eleven clicks, the context switching, the fact that the intake form, the note, the order, and the follow-up are four disconnected places — then a feature added to that workflow cannot resolve it, because the workflow is the thing you are objecting to.

Worth being clear about the limit that applies to both architectures: neither removes clinician review. A 2025 commentary in npj Digital Medicine notes that evaluations of modern LLM-based ambient scribes report overall error rates of roughly 1–3%, while cautioning that reported rates vary considerably with evaluation methodology — some studies count only factual inaccuracies, others include omissions and clinical inconsistencies. Any error rate above zero is why review before signing is a matter of clinical safety rather than preference. No architecture changes that.

The cash-pay version of the problem

Here is where the general argument becomes specific to a direct practice, and where it gets sharper.

Bundled AI in a conventional EHR is tuned toward a particular output: a structured, coded, audit-defensible note that supports a level of service. That is the correct target for the system it lives in, and enormous effort has gone into hitting it well.

You have no use for that artifact.

You are not selecting an E/M level. You are not justifying anything to an auditor. Nobody outside the room is going to review this note to decide whether you get paid. What you actually want from documentation is the thing a note was originally for — an accurate, readable account of what happened and what should happen next, useful to whoever sees this patient in six months, including you.

So the free AI keeps optimizing hard toward a goal you do not share. It suggests codes you will not submit. It nudges toward completeness that exists to support billing rather than care. It produces the padded artifact that a fee-for-service environment rewards, in a practice built specifically to escape that environment. It is solving someone else's problem extremely well, at no charge, inside your chart.

This is the trap worth naming, because it is invisible in a demo. The AI looks impressive. It is impressive. It is aimed somewhere else.

What owning the record actually means

The alternative is not "a better scribe." It is a system where the AI's output is the record rather than a draft handed across to something else.

The practical difference is whether intelligence can act across steps or only produce text at one of them. When the same intelligence carries a visit from the patient conversation through the structured note, into orders and prescribing, and out to follow-up, the handoffs stop existing rather than getting automated. There is no paste step, because there is nothing to paste between.

That continuity is only available when the system is built around it. It is also, to be fair, the more disruptive option by a wide margin: it means changing your EHR, with everything that implies.

When free wins

Genuinely often. Take the bundled option when:

• Your EHR is fundamentally fine and documentation time is the specific complaint.
• You are inside a contract term, or the export terms make leaving expensive.
• You have just switched systems and the practice cannot absorb another change.
• You are a solo clinician for whom thirteen minutes a day is a real win and a migration is a genuine risk.

Consider the harder option when:

• The workflow itself is the problem, not the typing.
• You are starting a practice and have no incumbent to defend, which is the cheapest moment this decision will ever be.
• Your documentation needs are shaped by care rather than by coding, and the tooling keeps pulling the other way.
• You are already paying for several systems that exist to compensate for the main one.

How to tell the difference in a demo

Four questions that cut through it:

1. Show me the AI doing something that isn't text. Placing an order, moving work forward, closing a loop. If every demonstration is note generation, you are looking at a scribe with good placement.
2. What happens after the note is signed? If the answer stops there, the AI stops there.
3. Can it produce a note that isn't built around a billing level? Ask to see documentation for a practice that codes nothing.
4. Where does the human sit? Any system that suggests review is optional is overselling, and any vendor comfortable saying "our AI is wrong sometimes, here is where you check it" is more trustworthy than one that isn't.

Where Zenthea fits

Zenthea is built as an AI-native EHR, designed around a clinical AI assistant, Thea, from the workflow foundation rather than as a feature layered onto a legacy record — documentation, orders and prescribing, charting, scheduling, and a patient portal in one workflow. The principle is that the AI prepares and a clinician decides: Thea drafts and prepares, and a human clinician reviews and signs every clinical action.

The concession this post has to make, and means: if your current EHR works and you mainly want the typing to stop, the free AI already sitting inside it is the better-value answer, and the cost of proving us right is one afternoon. The argument for changing the record underneath only holds when the workflow, rather than the transcription, is what is wearing you down.

The broader case for why a direct practice should weigh all of this differently is in what an EHR for a cash-pay or direct primary care practice has to do differently. If you are weighing a move, what switching EHRs actually costs a small practice is the number that belongs on the other side of the scale.