If you’re curious how much time an AI scribe could save you, yet wary of the deluge of AI-powered features flooding the market, you're in good company. That split is showing up everywhere in behavioral health right now, and it's worth taking seriously before you flip the switch on any AI tool in your EHR.

The data backs up the ambivalence among clinicians. According to the APA's 2025 Practitioner Pulse Survey, only 44% of psychologists said they'd never used an AI tool in their practice, down sharply from 71% the year before, a real jump in adoption. At the same time, about two-thirds of practitioners in that same survey said they were concerned about data breaches, biased outputs, inaccurate results, and tools that haven't been adequately tested. Practices aren't rejecting AI outright, but trying to figure out which parts are useful and secure. 

In this article, we review what the evidence actually shows about AI scribes and AI charting, the pros and cons of AI in your EHR, and what to look for before you adopt anything.

Why practices are exploring AI scribes for therapists

The appeal is straightforward: documentation is one of the biggest drivers of clinician burnout, and an AI scribe for therapists promises to hand that work back. Clinicians using ambient AI scribes for mental health documentation commonly report saving significant chunks of their week on notes. A Duke University evaluation of AI-generated clinical notes found they cut after-hours note-writing by roughly 30%, with notes rated "generally clear and acceptable" by reviewing clinicians. 

For a solo or small group practice, that kind of time back can mean more sessions, more intake capacity, or reclaiming personal time. It's a real, measurable benefit, and exactly why interest in AI scribes has grown so quickly, even among practitioners who are wary of AI in other contexts.

Why other practitioners are pushing back

The resistance isn't about technophobia. Most behavioral health clinicians already rely on software to run their practices. Rather, three concerns come up consistently in surveys and conversations among private practice clinicians. 

1. Trusting in the output: Clinicians want to feel confident that a note is an accurate reflection of what happened in session before they sign off on it. 
2. Data handling: Clinicians bound by HIPPA and their professional ethics have real and ethical questions around session confidentiality, such as: Where does the audio go? Is it used to train a model? Who can access it?
3. Feeling sold: Clinicians feel frustration with EHR vendors bundling AI features into pricing tiers or marketing them aggressively. That feeling, of a tool being pushed rather than chosen, tends to breed distrust of the EHR relationship generally, not just the AI feature itself.

Even clinicians who've already embraced AI scribes often share the same concerns as those who haven't. That concern is less about the tool itself and more about how much control and information the practice has going in.

Is AI charting accurate? What the research actually shows

This is the question worth sitting with before you adopt anything, and the answer is genuinely encouraging: "accurate" isn't all-or-nothing, and the tools are advancing considerably. 

Independent evaluations of AI-generated clinical documentation show early systems capturing 60-80% of standard note content accurately right out of the gate, and when a note falls short, it's usually a missing detail rather than a made-up one. Purpose-built psychiatric and mental health scribes report even higher accuracy, since they're trained on therapeutic language rather than general medical transcription. 

The clearest pattern in the research: tools built specifically for behavioral health handle clinical nuance and multi-session context far better than general-purpose transcription apps, so choosing the right tool matters more than whether you use AI at all. The one habit worth keeping regardless of which tool you pick is a quick clinician read-through before a note is finalized. Think of AI charting as a strong first draft that gets you most of the way there, not a finished product. 

AI in EHR: pros and cons to weigh

No AI feature is universally good or bad for your practice. It comes down to what it actually gives you against what it costs you, financially and clinically.

The case for: less time on documentation, faster turnaround on notes and treatment plans, and potentially better session presence since you're not splitting attention between the client and your keyboard.

The case for caution: accuracy still varies by tool and use case, clinical oversight remains non-negotiable, data privacy and model training practices differ enormously between vendors, and there's a real cost, financial and in trust, to adopting a feature your practice didn't necessarily need. 

Neither list should be the deciding factor on its own. What matters is whether a specific tool solves a specific problem your practice actually has.

What to look for before adopting an AI feature in your EHR

A few questions worth asking any EHR vendor before you turn on an AI feature:

• Is it opt-in, or is it bundled into your subscription whether you use it or not? Can you turn it off without losing other functionality?
• What happens to session audio and transcripts: are they stored, for how long, and are they used to train the vendor's models?
• Has the tool been evaluated for accuracy specifically in behavioral health contexts, not just general medical documentation?
• Does the vendor undergo independent security audits, and can they show you documentation of that?
• Is pricing transparent, or does the AI feature show up as a surprise line item?

This is the same trust gap showing up in the survey data, translated into practical due diligence, and it's the thinking behind Healthie's approach to AI. Healthie's native tools, including AI Scribe and Ambient AI, are held to real scrutiny on clinical accuracy and risk before they reach a practice, and they run on the same infrastructure behind Healthie's HITRUST R2 and SOC 2 Type 2 certifications, the kind of independent audits worth asking any vendor to show you.

On the data side, session audio is processed solely to generate the chart note and transcript, PHI stays within the client's profile in Healthie rather than being stored or reused by a third party, and transcripts are retained for two years before being automatically and permanently deleted. Encryption in transit and at rest protects that data throughout, though providers are still responsible for getting client consent before recording any session.

AI Scribe is also an add-on rather than something bundled into every plan, priced by usage, and can be turned on or off per provider, so a practice can offer it to the clinicians who want it without changing anything for the ones who don't.

The bottom line

AI in EHRs isn't a single decision, but rather a series of smaller decisions about specific tools, specific use cases, and how much oversight your practice wants to keep in place. The practices getting the most value aren't necessarily the ones that adopted everything or the ones that rejected everything. They're the ones asking good questions about accuracy, data handling, and control before they commit.

This scrutiny is only going to pay off more from here, as accuracy keeps climbing, and vendors that take data handling and clinical oversight seriously are raising the bar for everyone else in the space. The practices asking sharp questions today are the ones best positioned to take advantage as these tools keep improving.