Our software drafts prose from material a clinician supplies and reviews. It does not decide what a pattern of findings means, and it isn't built to. That's a line we drew on purpose, and it's the single design decision that has shaped the product more than any other.
I'm a founder, not a clinician. I'd be the last person who should be making calls inside a report, and I don't think software should either. What follows is why we hold that line, what it actually means in the product rather than in a slogan, and what it costs us — because a principle a vendor states without a cost attached isn't worth much.
The typing is ours to take on. The judgment isn't.
Why does authorship matter this much?
Because it's the thing the profession is actually selling, and every psychologist I've asked said so in almost identical words.
The interpretation is what they trained for. It's what their name on the document certifies. It's what a school, a physician, or a hearing is relying on. A report is not a summary of scores — it's a professional opinion, attributed to a named person who can be asked to account for it.
Which means a tool that produces conclusions isn't saving that person work. It's producing something they now have to verify rather than something they wrote, and verifying a conclusion you didn't reach is often slower and always riskier than reaching it yourself. Several clinicians described exactly this with general-purpose tools: fluent output that read plausibly and took longer to check than to have written.
What does the line mean in the product?
Three concrete things, not a philosophy.
It drafts from supplied material only. The draft is built from what the clinician provides. It isn't asked to infer findings that weren't given to it, and where information is absent the correct behaviour is to leave a gap rather than produce a confident sentence filling it.
It doesn't write the analytic conclusion. The integration section — the part psychologists consistently name as the hardest and the most theirs — is scaffolded, not authored. The tool can lay out what needs addressing. It does not decide what the findings indicate.
Everything is traceable back to the input. This is the part that makes review real rather than ceremonial. If a clinician can check a sentence against the material they supplied, review is a genuine check. If they'd have to re-derive it from scratch, "review and sign" is a formality dressed as a safeguard.
That third point is the operational version of the whole principle, and it's the one I'd press any vendor on, including us.
What does "review and sign" have to mean to count?
Every product in this category uses that phrase, mine included, which is exactly why it's worth pushing on.
A review is meaningful when the reviewer can tell where each claim came from and could reconstruct it if challenged. A review is theatre when the document is fluent, plausible, and unverifiable without redoing the work. Both look identical on a demo call and completely different in month three.
So the test I'd apply: take a paragraph of the draft and try to trace it back to the source material. If you can, the review is real. If you're relying on the prose sounding right, you've been handed liability with a time saving attached. There's a longer set of criteria in how to choose documentation AI for a clinical practice.
A review is real when you can trace a sentence back to what you supplied. Otherwise it's a formality.
What does holding this line cost us?
Genuinely quite a lot, and I'd rather say so than pretend the principle is free.
It costs us the impressive demo. The version where you press a button and a complete report appears is a much better thirty seconds than ours. We can't do that, because the interesting part of the document is the part we won't generate.
It costs us on comparison. Anyone willing to have their tool write conclusions will show a larger apparent time saving than we will, because they're removing work we deliberately leave in place.
It means the hardest hour is still the clinician's. We don't make the integration section faster. We make sure the time available for it isn't already spent on the blank page and the boilerplate. That's a narrower promise and it's the one we can actually keep.
I'm at peace with all three. The alternative is selling a shortcut through the exact part of the work that a person's name is attached to, and I don't want to be the company that did that.
Doesn't the clinician stay the author with any tool?
Formally, yes — the signature is always theirs. Practically, it depends entirely on whether the workflow makes that signature meaningful.
Authorship isn't a checkbox at the end. It's whether the person signing shaped the analytic content, and whether they can defend it because they reached it. A tool that generates conclusions and asks for approval has technically preserved authorship and materially eroded it.
That's the distinction we build against. It's also why the first question almost every psychologist asks me is whether a draft will still sound like them — a question I think is exactly right, and one I've written about in will an AI-drafted report still sound like me.
If you want to see where the line falls in practice, that's what the drafting does: a first draft from your own material, which you correct and sign. The judgment stays where it belongs.
— Ian
