Common Pitfalls & How to Avoid Them
From the course AI for Therapists & Counselors
Therapists who encounter problems with AI in their practice typically run into one of five predictable failure modes. Each has a clear solution.
Pitfall 1: Using Non-HIPAA-Compliant Tools for Client Content
The most common and most serious mistake: pasting session notes, client details, or intake information into a standard ChatGPT or Claude account. Standard consumer tiers of these tools do not carry BAAs and may use submitted data for model training.
How to avoid it:
- Create a clear mental (and physical) firewall: standard AI tools are for administrative and educational content only — no client names, no session details, no identifying information
- For clinical documentation AI, use only platforms with signed BAAs
- Post a physical reminder at your workstation if needed: "No client data in ChatGPT"
If this boundary is ever breached, treat it as a potential HIPAA breach and follow your practice's breach notification protocol.
Pitfall 2: Signing AI Notes Without Full Review
AI documentation platforms are trained on clinical language and produce notes that look correct. The risk is that a plausible-sounding note may contain a clinical error — misattributing a symptom, inventing a coping strategy the client did not mention, or omitting a safety disclosure.
How to avoid it:
- Read every AI-generated note completely before signing — never approve a batch
- Compare the AI note to your mental recall of the session: does it accurately reflect what was discussed?
- Flag any instance where AI added content you do not remember — this is a training data contamination issue to report to the vendor
The legal standard is clear: your signature means you attest to the accuracy of the note. "The AI wrote it" is not a defense.
Pitfall 3: AI Replacing the Therapeutic Relationship
Some therapists begin using AI chatbots to answer client questions between sessions or to provide psychoeducation in an interactive format. This crosses the line from tool to intervention and creates liability, scope-of-practice, and relationship disruption risks.
How to avoid it:
- AI should never communicate directly with clients on clinical matters
- Between-session tools that clients interact with (apps, journals) should be pre-vetted, evidence-based platforms — not custom AI chatbots you configure yourself
- If a client asks whether they can use ChatGPT for mental health support, provide psychoeducation about the limitations and refer them to validated self-help apps like Woebot (with appropriate caveats)
Pitfall 4: Skipping the Opt-Out Conversation
Therapists sometimes update their consent forms to mention AI but never discuss it verbally with existing clients. For a new presenting issue or a vulnerable client, discovering that AI was used in their documentation can feel like a violation of trust — even if it was legally disclosed.
How to avoid it:
- Have an explicit verbal conversation with each existing client before implementing AI in their documentation
- Frame it positively: "I am using a new documentation tool that helps me write notes more quickly so I can spend more time focused on you during sessions"
- Document the consent conversation in the chart
- Honor opt-out requests promptly and without defensiveness
Pitfall 5: Neglecting Continuing Education on AI Ethics
AI capabilities and the ethical guidelines around them are evolving faster than standard CE cycles. APA, NASW, and ACA are all publishing updated guidance, but therapists who are not actively monitoring the field may be operating on outdated assumptions.
How to avoid it:
- Set a quarterly calendar reminder to check your primary professional association for updated AI guidance
- Attend at least one CE course per year specifically on technology and ethics in mental health practice
- Raise AI usage in clinical supervision — your supervisor's perspective is a built-in ethics check
- Join a peer consultation group that includes discussion of technology in practice
Key Takeaways
- •Never paste client data into consumer AI tools — only HIPAA BAA platforms for any clinical content
- •Read every AI note fully before signing; 'the AI wrote it' is not a defense in a licensing complaint
- •AI must never communicate directly with clients on clinical matters — that crosses into intervention
- •Have explicit verbal consent conversations with existing clients before implementing AI documentation
Progress is kept in this browser only.