After dozens of conversations with psychologists and counsellors about AI, the honest pattern is this — they aren't anti-AI, they're protective. Protective of their clinical judgment, of their clients' trust, and of where data lives. The clinicians I talk to want it done responsibly, with them in control.
I want to share what I've heard, plainly. Not because I have it all figured out — I don't — but because these conversations have shaped how I think about building software for clinicians, and I'm grateful for every one of them.
So here's the pattern, as honestly as I can put it.
What worries clinicians most about AI?
What worries them most is losing control — of the work, the relationship, and the data behind both. Almost no one I've spoken with is anti-AI. They're cautious, which is different.
The fear underneath most questions is the same — if I bring this tool into my practice, am I still the one in charge? That's a fair thing to ask, and I've come to think it's the right instinct.
A lot of clinicians also carry a quieter worry — will this sound like me? They've spent years developing how they write, how they frame a client's progress, how they document with care. The idea of a draft that flattens all of that into a generic template is, understandably, a turn-off.
I get it. The goal should always be a draft that reads like yours, not something off a shelf.
How should clinicians handle client consent around AI?
Consent should come first, in plain language, before any tool touches the work. The clinicians I respect most don't hide the fact that they use software — they explain it.
Several have told me they sit down with wary clients and walk them through it. What the tool does, what it doesn't, and that the clinician is still the one writing the clinical record. That conversation builds trust rather than spending it.
Some clinicians work under employers who restrict AI tools entirely, and that's a real constraint I take seriously. If your workplace has a policy, that comes first — full stop.
My honest take — consent isn't a checkbox or a line of fine print. It's a conversation, and it's part of the relationship. A good tool should make that conversation easier to have, not something you have to apologize for.
Where client data lives is the question clinicians are right to press hardest on.
Is client data safe with AI tools?
It depends entirely on the vendor — which is exactly why this question matters so much. Clinicians are right to ask where data lives and whether it's used to train models. Those are not paranoid questions. They're the questions.
The hard truth is that "de-identified" is doing a lot of work in this industry, and not all of it is honest. Removing a name isn't the same as making something anonymous. Research on automated de-identification of clinical notes shows it's a genuinely difficult technical problem, not a magic switch you flip (Nature, 2024).
It's worth going further than that. Privacy experts have argued that anonymization itself can be something of a myth in an AI world — that protecting people takes more care than simply stripping obvious identifiers (IAPP).
I share that not to scare anyone, but because I'd rather clinicians hold every vendor — including us — to a high bar. The clinicians I've talked to care deeply about whether their clients' information is treated with respect. That care is the whole point, and a tool should earn it, not assume it.
That's also why, when a clinician sets things up with us, what they send is de-identified by design. Privacy-first isn't a feature you bolt on later. It's the starting posture.
The software drafts; the clinician keeps every clinical decision.
Does AI replace clinical judgment?
No — and any tool that claims to should worry you. This is the line I hear most often, and I agree with it completely. AI can assist with the drafting. It cannot, and should not, decide.
The clinicians I talk to are clear about this. They want help with the parts that drain them — the blank page, the repetitive structure, the hour of documentation after a long day. That's exactly what we built a clinician-first alternative to ChatGPT for notes to handle. They do not want a tool that pretends to know their client better than they do.
I think of it like this — the clinical judgment stays entirely with the clinician. The software is there to take a first pass at the writing, so the expert can do what only the expert can do: read it, correct it, and sign off on it knowing it's right.
That's the only version of this I'm comfortable building. A draft you control beats a decision you didn't make, every time. (I've also weighed in on using ChatGPT for clinical notes and on whether an AI-drafted report still sounds like you.) If you want to see how we keep the clinician in that seat, that's the heart of what we've built — software that drafts, while you decide.
Good vendors welcome the scrutiny — the right questions keep everyone honest.
What should a clinician ask an AI vendor?
Ask the questions that protect your judgment, your clients, and your data — and don't accept vague answers. Over these conversations, a short list has emerged that I'd genuinely want any clinician to use, including on me.
A few I'd put near the top:
- Where does my data live, and is it used to train models? If the answer is fuzzy, that's an answer.
- What does "de-identified" actually mean here? A serious vendor can explain it plainly.
- Can I edit and override every draft? If the tool doesn't put you in control, walk away.
- Will it write in my structure and voice, or a generic template? The draft should serve your work, not reshape it.
- What happens if I want to leave? Your records are yours.
If a vendor gets defensive about any of these, that tells you something. The good ones welcome the scrutiny — I've come to see these questions as a gift, because they keep us honest.
The thread through all of it
If there's one thing these conversations have taught me, it's that the caution clinicians bring to AI is a strength, not a barrier. They're protecting the things that make their work matter — trust, judgment, and care for the people in front of them.
My job, as I see it, is to build software that respects all three. Clinician in control. Consent first. Privacy by default. A draft that reads like yours. That's it.
Thank you to everyone who has taken the time to share their worries and their hopes with me so candidly. It means a great deal, and it's made the work better. I'm still learning, and I'd love to keep the conversation going.
— Ian
