If AI drafted this note, who actually signed it?

The OT finished the session. There's already an AI SOAP note on the screen. She clicked Sign on the way to the next room.

On Thursday the parent asks about a goal the note says they practiced. She didn't practice it.

The draft looked plausible. The signature made it a record.

You may recognize this. Clinics adopt AI therapy notes because the day's already full. The next child's already waiting, so Sign becomes a click between rooms.

The cost, when it goes wrong, is a signed lie, not a slow note.

Here's the check I'd run this week. The Sign Screen Check: before Sign is live, the screen shows this week's actual session, and the signer can point to where each line came from.

If they can, you pass. Stop reading. You don't have this problem.

If they can't, don't blame the therapist, and don't ban AI. Fix the sign screen.

Look at the screen before Sign is live

Pick one note from this week. Occupational therapy, speech, PT, or ABA. Open the screen the signer actually uses. Don't use a demo.

You should see this child, this date, this discipline, the named author, and who was in the room. You should see the goals they actually touched, and what happened in the hour.

Then they should be able to point. This sentence is the seating they used. This one is the two-word phrase.

This line is the plan they'd run on Thursday. If they can do that, Sign can go live.

That's a pass. That's this week's session, sitting in front of the person who'd defend it.

Don't run this from the owner's desk. Sit with the therapist who treated the child. Ask them to point, out loud, before Sign is clickable.

If they hesitate on a line, that line isn't ready. If they can walk the whole note in a minute, you're done.

If the screen could be any child, any discipline, any Tuesday, it failed. If they have to open last month's note to know who they're signing for, it failed.

If one person could sign OT, speech, and PT because the button was there, it failed. That would make it look like one clinician delivered every discipline.

If they can't say where a sentence came from, that sentence shouldn't enter the record. "Tolerated well" is the giveaway.

So is a plan that doesn't follow the data. So is assessment language sitting in a treatment note.

Can you use AI for therapy notes? Yes, as a draft. An AI scribe for occupational therapy or for speech therapy can structure what the therapist already observed.

It doesn't observe the child. The named human who reviews the draft and signs it is the author.

Not the software. Not, usually, the clinic owner.

The owner still owns the system. They own whether recording or AI was consented. They own who can see a child's record.

They own whether Sign can go live on a draft nobody would defend. Owning the clinic doesn't make them the person who signs the clinical note.

Each clinician signs their own discipline's note. If two people treated together, each contribution should still be identifiable. The person who turns an AI draft into a record is the person who must stand behind it.

If Sign went live before they could point

Don't tell therapists to write more. They already write.

You're not asking for a longer SOAP. You're stopping Sign until that picture is true.

Do this on the same note this week.

1. Keep Sign locked until the treating clinician is on that session. The person who ran the hour, or directly supervised it, is the only one who can make the draft a record.
2. Put this week's session on that screen before the button is live. Child, discipline, goals, what happened. If they have to hunt last month's note to know who they're signing for, Sign stays locked.
3. Remove the lines they can't source. If they can't say where a sentence came from, it doesn't enter the record. Later, an addendum is fine. After lock, don't overwrite in silence.

The screen should also make the state obvious. Draft, reviewed, signed, and amended shouldn't look like the same document.

Once it's signed, it should lock. If something was wrong, the next line is an addendum.

Then check the same therapist next week. Don't ask if they "reviewed." Ask if they could point.

If the software still lets Sign go live on a draft nobody would defend, you now know what you need from the screen. You don't need a longer policy.

The draft stays a draft until a person has looked

You can run that check without changing software. It'll tell you whether AI SOAP notes for therapists are saving the afternoon, or creating a record too early.

In Metanoa, AI is a starting point. A few rough notes become a SOAP draft. Nothing's saved without someone confirming it.

Signing locks the note. After that it is a record, not a draft you can quietly edit.

If you want to see that with your own disciplines, start at metanoa.ai.

FAQ

Can you use AI for therapy notes?

Yes, as a draft the treating clinician reviews and signs. The signature means this is an accurate record of the service they provided or directly supervised, and of their clinical judgment.

Who is responsible if an AI draft gets a detail wrong?

The clinician who signed it. The clinic's responsible for a governed system: consent, access, and a Sign button that can't go live on a draft nobody has looked at.

Does the clinic owner sign AI notes?

Not automatically. Each clinician signs their own discipline's note. The owner owns whether the system's safe to use.

Can AI write speech therapy or occupational therapy SOAP notes?

It can draft them from what the therapist already observed. It doesn't observe the child, and it doesn't become the record until a person has looked.

Do we have to review every AI-generated note before signing?

Yes. If the signer can't point to where a line came from, that line shouldn't enter the record.