6 Requirements from APA’s Ethical Guidance for Automated Report Drafting
The checklist already exists. Most practices are following half of it without knowing that.
A psychologist downloaded APA’s Ethical Guidance for AI in the Professional Practice of Health Service Psychology the week it was updated, in December 2025, looking for direction on how to use automated report drafting in her intake process without crossing an ethical line. She read it twice at her kitchen table, checking each requirement against what she already had in place. It read like an inspection sheet, and her documentation practice matched most of the items on it already.
The default read is backwards
Ask a room of psychologists what the APA’s AI guidance means for their practice and most will say some version of “be careful” or “maybe hold off.” That reading treats the document as a warning label. Read the actual guidance, developed by APA’s Mental Health Technology Advisory Committee and adopted by the APA Ethics Committee, and it reads differently: a set of conditions under which AI-assisted documentation is defensible. A psychologist who already keeps human oversight, discloses the tool to the family, and reviews every line before signing is closer to full compliance than the “be careful” framing suggests.
Six requirements repeat throughout, most careful practices are already hitting them.
What the APA ethical guidance actually asks for
APA’s guidance is organized around ethical principles rather than a numbered checklist, but six requirements recur across it.
Disclosure and informed consent. Families need to know when a tool assists with a report, in language they understand, before the report is used. For documentation: state in the report or the intake paperwork that a drafting tool was used, plainly and up front.
Human oversight of every output. The clinician stays responsible for evaluating the content before it goes anywhere, at the start of use and on an ongoing basis. For documentation: no report leaves the building without a clinician who reads every section.
Bias and fairness. Tools should be evaluated for whether they perform consistently across different populations, with results checked group by group rather than only in aggregate. For documentation: ask a vendor how their tool was tested across race, language, and diagnostic presentation.
Data privacy and security. Tools handling behavioral health data need to meet HIPAA and comparable standards, with the psychologist responsible for confirming it. For documentation: a signed BAA and a written data-use policy.
Validation before use. Tools should be tested and shown to work, with evidence a practice can review, before they touch a live caseload. For documentation: ask what the tool was validated against, and whether that validation is available to review.
Legal and liability awareness. The psychologist carries the liability for what the tool produces, and training staff on its limitations is part of managing that risk. For documentation: whoever signs the report needs to understand what the tool did and didn’t do.
None of these are a reason to write report drafts manually
Run this against whatever solution you use
Does a clinician read it before it goes out? If a workflow lets a report leave without a human review step, that’s the exact risk the guidance names.
Has it been tested across the caseload you actually serve? Ask the vendor directly. If the answer is a features list, ask again.
Is the privacy and security answer a signed document or a sales pitch? A BAA and a written no-training policy are documents you can read. A compliance badge on a landing page is not.
Does your team understand what it doesn’t do? The riskiest gap in a practice is a clinician who never learned where the tool stops.
One or two “no” answers should be your first priorities to address.
Where Diagnosis Builder sits against this
A documentation platform built for autism diagnostic evaluations satisfies most of the six by design. Every draft is generated for review, marked as a draft pending the clinician’s signature. Every draft requires a clinician’s edit and approval before anything is finalized. Data stays in an isolated tenant under a signed BAA, with no training on patient records. What the platform does not do is evaluate its own accuracy across a practice’s caseload. That validation step belongs to the practice, and the APA’s guidance is right to ask for it. The psychologist at the kitchen table has her checklist now.
