ACA AI Recommendations: Test a Tool Before Client Use
Outline
Test three fictional responses before adoption
Document your AI tool test
Use the sheet this week to record your fictional test and its disposition.
- NASW Provision-to-Evidence Map for One AI Use
- Counselor’s AI Tool Acceptance Test for three fictional inputs and a disposition
- Couples AI Interpretation Check for sources, both perspectives, and an objective or disagreement
- AI Theme to Human Observation: Keep, Revise, or Set Aside, with private and observer sides
Free. We'll email the PDF link right away and add you to the therapist toolkit, sent every two weeks. Unsubscribe any time.
Where should we send the link?
We'll email the PDF link right away. You'll also get the therapist toolkit every two weeks. Unsubscribe any time.
✓ Check your inbox
We've sent you the PDF
The download link is on its way to your inbox, usually within a minute or two. The email will come from Emosapien (hello@team.emosapien.com); check your spam folder if you don't see it.
You're also on the therapist toolkit list, sent every two weeks. Unsubscribe any time from the email footer.
Decide what the tool is allowed to draft
Your intended use is a repeatable draft of one neutral check-in question, not a diagnosis, crisis response, or decision about treatment. Set the boundary before you try the tool. A general assistant can be enough for a generic, nonidentifying drafting task; that does not make it suitable for identifiable client information. You choose the wording and whether to use it.
The American Counseling Association’s recommendations for practicing counselors, checked November 2026, call for critical evaluation, attention to bias and privacy, informed consent, and continued human judgment. They are professional recommendations, not a binding AI statute or certification. You can use the three fictional conditions to evaluate a prompt writer; ACA does not mandate those particular cases. For the broader professional framework, see the ACA Code of Ethics guide.
The ACA AI recommendations are professional guidance, not a substitute for your own acceptance criteria. A passed test does not demonstrate clinical validity or authorize live data. It gives you a limited observation about a specific task and version. For a broader vendor review and consent process, use the AI therapy documentation ethics guide.
Write the permitted task in a sentence before testing: draft a neutral question in response to one fictional condition. Write the excluded tasks alongside it: interpreting a client’s silence, assessing risk, or advising someone to wait for an automated reply. This distinction gives a reviewer a reason to stop a draft even when it sounds reassuring. Record the tool’s version and instructions so a later reviewer can tell what you actually tested. A tool that produces a useful generic question once has not earned approval for every client, every setting, or a different task.
Apply ACA AI recommendations to each test condition
The ACA AI recommendations, checked November 2026, ask you to evaluate technology critically and avoid over-reliance. Translate that into a written decision about this particular task rather than a general approval of every output the tool might produce. Decide who will review the test, what would stop adoption, and where you will retain the results. If another clinician will use the same workflow, agree on the review responsibilities before any draft reaches a client.
Give the tool the same task and relevant instructions for each fictional condition, one at a time. Ask it for a short, neutral question, not an interpretation of the client’s state. Compare the actual text with your expected behavior before making any correction.
A decline must remain a choice, not become noncompliance. A request for clarification should invite discussion without blame. A blank means you have no response to interpret. Any suggestion to wait for automated help fails this test. Do not send a failed draft; decide whether a revised instruction merits a fresh trial or whether to suspend the tool.
Consider whether the language could shame someone, assume a particular culture or ability, or imply a person must answer. Three cases cannot reveal every bias or failure. Repeat the test when the tool, configuration, instructions, task, or intended population changes, and after a concerning output. Retain human responsibility for the final question and for follow-up.
For the clarification condition, check that the wording leaves room for a client to identify what was unclear, rather than guessing that the client lacks motivation. For a declined task, consider whether the question makes nonparticipation sound like a problem to justify. A neutral question can still be unsuitable for a particular client, so use your knowledge of the therapeutic relationship when you make the final selection.
When you retry after an error, keep both the original result and the new result. Record the changed instruction and run all three conditions again rather than checking only the one that failed. A single improved response does not establish consistency. If you cannot explain why the corrected question is appropriate, or if unsafe wording persists, suspend the proposed use and consult a colleague or supervisor. Use your usual non-AI approach while you decide what to do next.
For a separate documentation workflow, see how notes are drafted and reviewed.
Make the result useful to the next reviewer
Use a separate record for each fictional condition. Copy the exact input and the draft as returned before editing it. Then note the expected behavior and the words that met or missed it. For a blank response, for example, a draft that says “you are making progress” fails even if the rest of the sentence is gentle. Write a neutral replacement and mark whether you will retry the instructions or suspend the proposed use. Preserve the failed version rather than replacing it with the correction.
If the tool produces no question, record that result too. An empty output is not the same as a safe question, and a technical error is not a successful test. A reviewer should be able to see which conditions were actually tried, not just a summary saying the tool passed. If you change the prompt after one failure, identify what changed and test all three conditions again. Compare the new output with the same stop rules; do not average a serious failure against two acceptable answers.
Agree on who can sign off on the narrow proposed task and when the decision expires. A new model version, a changed instruction, or use with a different population can alter the output, so record a trigger for repeating the trial. If results remain uncertain, ask a colleague or supervisor to review them and continue with your usual non-AI approach. Keep any client-specific decisions separate from these fictional test records.
Use a reviewed question, not an automated judgment
Emosapien’s Engagement Agent offers guided check-ins, AI-assisted journaling, and modality-aligned homework between sessions. Clients can use its mobile app for check-ins and journaling. Your acceptance test remains separate from those features: do not infer improvement from silence or rely on an automated reply or safety monitoring. Arrange human follow-up appropriate to your setting and the client’s needs.
Record the disclosure and the client’s choice
For disclosure and client choice, use the AI therapy documentation ethics guide.
Use the check-in question that passed your review
Use your reviewed question in your own clinical workflow. Separately, explore Emosapien guided check-ins and client journaling, available on Professional and Enterprise plans.
Explore check-ins- No credit card
- HIPAA-ready
Guided check-ins
Support your clients between sessions.
Clinician judgment
Review your own acceptance test separately.
References
- American Counseling Association. Recommendations for Practicing Counselors and Their Use of AI, checked November 2026.
Related reading
ACA Ethical Codes: Principles and Section-by-Section Practice Guide
Use the ACA ethical codes as a practice guide. Match each ACA Code of Ethics section to the counseling decision in front of you.
Read article →ACA Liability Insurance for Counselors
Compare ACA liability insurance for counselors: the HPSO-administered program, member discount, student options, and license-defense caps.
Read article →Free PDF: Professional Ethics AI Decision Cards
Printable decision sheets for mapping AI evidence, testing a tool, checking a couples interpretation, and preparing a human-observed rehearsal.
- NASW Provision-to-Evidence Map for One AI Use
- Counselor’s AI Tool Acceptance Test for three fictional inputs and a disposition
- Couples AI Interpretation Check for sources, both perspectives, and an objective or disagreement
- AI Theme to Human Observation: Keep, Revise, or Set Aside, with private and observer sides
Free. We'll email the PDF link right away and add you to the therapist toolkit, sent every two weeks. Unsubscribe any time.
Where should we send the link?
We'll email the PDF link right away. You'll also get the therapist toolkit every two weeks. Unsubscribe any time.
✓ Check your inbox
We've sent you the PDF
The download link is on its way to your inbox, usually within a minute or two. The email will come from Emosapien (hello@team.emosapien.com); check your spam folder if you don't see it.
You're also on the therapist toolkit list, sent every two weeks. Unsubscribe any time from the email footer.