Who authored the note when AI drafted it

The clinical problem

If a language model drafted language that entered the clinical record, authorship is not decorative. The record still needs a responsible clinician.

Default: the treating psychologist remains the author of the note. AI is a drafting aid, not a co-signer and not a hidden scribe. Disclose use in the record when the draft shaped clinical content, and edit until every claim is one you would defend under review.

Do not paste unreviewed AI text into the chart. Do not let the model invent history, quotes, or risk language. If you cannot verify a sentence, delete it.

The clinical question is accountability: who stands behind the note when a later reader — client, board, or colleague — treats every line as yours.

This note is a practice map, not legal advice and not CE credit. Research Assistant to attach dated sources before Ready for mock.

What to notice

Which sentences entered the chart unchanged from the model.
Invented history, invented quotes, or risk language the clinician did not observe.
Whether disclosure of AI use is present, absent, or vague.
The pull to finish notes faster than judgment allows.

Questions to ask yourself

Would I defend every line of this note under board review?
What did the model add that I cannot verify from the session?
If the patient read this note, whose voice would they hear?
Am I using AI to think, or to avoid thinking?

Possible psychodynamic formulations

Displacement of authorship

Handing draft work to a model can enact a wish to be free of the burden of standing behind clinical language — while still signing it.

Speed as defense

Late-day charting pressure recruits tools that flatten particularity. The defense is against fatigue and guilt, not against the patient.

False coherence

Models produce fluent narrative. Fluency can hide missing affect, missing risk, or a formulation the clinician never actually held.

Omnipotent scribe fantasy

The fantasy that an invisible helper “got it right” can replace the slower work of deciding what belongs in the record.

Common clinician mistakes

Pasting unreviewed AI text into the chart.
Letting the model invent history, quotes, or risk language.
Treating AI as a co-signer or hidden scribe instead of a drafting aid.
Omitting disclosure when the draft shaped clinical content.

What might change the interpretation

Institutional policy that requires or forbids AI drafting (follow the stricter rule).
A patient request for the record that exposes fluent but inaccurate language.
Consultation or audit that cannot map sentences back to observed clinical facts.
New board or payer guidance on AI-assisted documentation in your jurisdiction.