Mynd
Using AI, Ethically and Effectively: A Therapist's Practical Guide to Prompting
May 12, 2025
Anas U.
The world of Artificial Intelligence (AI) tools, like ChatGPT, is buzzing. As mental health professionals, we're hearing about their potential to streamline administrative tasks, help with drafting notes, and even brainstorm ideas. The promise of freeing up valuable time is incredibly appealing in our often demanding work.
But with this potential comes essential responsibility. Just as you wouldn't leave sensitive client files exposed, using AI requires careful consideration, especially when it comes to client privacy and leveraging these tools effectively without compromising ethical standards or clinical judgment.
You might be wondering: How can I use these tools without risking confidentiality? And how do I get them to actually provide something useful for my specific needs?
The answer lies in prompting – essentially, learning how to ask the AI for what you need in a clear and specific way. You don't need to be a tech expert or a data scientist to do this. Think of it more like learning how to give excellent instructions to a new administrative assistant.
This guide is designed to help you understand the basics of prompting, see practical examples tailored for mental health practice, and most importantly, learn how to use AI ethically and safely.
Harness the power of AI to simplify your practice. All the capabilities discussed in this guide – including secure AI note-taking, AI assistants, session follow-ups, and homework suggestions – are available directly built into Mynd, our AI-first practice management platform, designed specifically for mental health professionals.
The Absolute First Rule: Client Privacy is Non-Negotiable
Before we dive into how to write a good prompt, we must address the most critical point: Protecting Client Confidentiality.
Pasting raw, identifiable client notes or information into any public AI tool is a serious breach of privacy and ethical guidelines (and likely legal requirements like HIPAA). AI models learn from the data they process, and you cannot assume information shared with a public tool remains private.
Therefore, the single most important step when using AI with anything related to client work is ruthless, complete de-identification.
What MUST you remove or change?
Names & Initials: Replace client names, family names, or any other specific names with placeholders (e.g., "Client A," "the individual," "Partner," "Child 1").
Specific Dates: Avoid using actual dates (birthdays, session dates, dates of events). Use relative terms (e.g., "last week," "in childhood," "several months ago") or generic placeholders.
Locations: Remove specific addresses, workplaces, schools, unique community names, or locations that could easily pinpoint someone. Use general descriptions (e.g., "their home," "work," "school," "local park").
Unique Identifying Details: Be cautious with highly specific details that could make someone identifiable when combined (e.g., a very rare medical condition plus a specific unusual hobby, or names of famous family members). Generalize or omit these if not essential to the task.
Think of it this way: If someone read the de-identified text, could they possibly figure out who you are talking about? If yes, de-identify further.
Only ever paste the fully de-identified information into the AI tool. Always review the privacy policy of any AI service you use, especially paid, professional-grade services, to understand how they handle data. Be extremely wary of free tools for any sensitive information, even de-identified.
What is a "Prompt," Anyway?
In the simplest terms, a prompt is the instruction or question you provide to the AI. It's how you tell the AI what you want it to do.
Example: You type, "Explain the concept of cognitive restructuring." That's a prompt.
Example: You type, "Summarize the following text." [then you paste text]. That's also a prompt.
Why Learn to Write Good Prompts?
AI models are incredibly powerful prediction engines – they predict the next most likely word based on the vast amount of text they were trained on and the input you give them.
Writing a clear prompt helps the AI make the right predictions to meet your specific needs. A good prompt means:
More Accurate Results: The AI is less likely to misunderstand your request.
More Relevant Output: The response will be tailored to the task you want to accomplish.
Reduced "Hallucinations": While AI can still make things up (more on this later!), clear instructions anchored to the context you provide can help minimize this.
Greater Efficiency: You get closer to the desired output on the first try, reducing back-and-forth or frustrating edits.
The Building Blocks of an Effective Prompt
Think of your prompt as having a few key components, like giving clear directions:
Your Goal/Instruction: What do you want the AI to do? Use strong action verbs.
Examples: Summarize, Draft, Brainstorm, Explain, Outline, Create, Write, Rephrase.
The Task: What specific task are you asking it to perform?
Example: Summarize de-identified session notes.
Example: Draft a section of a treatment plan.
Context (Your De-identified Information): This is the information the AI needs to work with (your de-identified notes, observations, concepts). Providing this context helps the AI focus.
Output Format (Optional but Helpful): How do you want the answer to look?
Examples: In bullet points, as a paragraph, in a table, 3-5 sentences, using simple language.
A simple formula:
[Action Verb] [Specific Task] based on [Your De-identified Context]. [Optional: Specify Output Format].
Practical Prompt Examples for Your Practice
Here are some examples of how you can structure prompts for common tasks. Remember: ALWAYS insert your de-identified information where indicated.
Task: Summarizing De-identified Session Notes
Prompt 1 (General Summary):
"Summarize the following de-identified session notes into 5 main points highlighting Client A's reported mood, key topics discussed, and any homework assigned.
[Paste your fully de-identified session notes here]"Prompt 2 (Focus on Specific Theme):
"From the following de-identified session notes for Client B, extract all information related to their reported experiences with anxiety symptoms. Present the information as a short paragraph.
[Paste your fully de-identified session notes here]"
Task: Drafting Sections of Notes or Treatment Plans (Using De-identified Info)
Prompt 3 (Objective Section Draft):
"Draft the 'Objective' section of a progress note based on these de-identified behavioral observations from Client C's session:
[Paste your fully de-identified observations, e.g., 'Client arrived on time, maintained consistent eye contact, spoke at a moderate pace, reported stable sleep hygiene, denied suicidal ideation.']"Prompt 4 (Assessment Section Draft):
"Draft a concise 'Assessment' section based on the following de-identified information for Client D, linking reported symptoms to diagnostic criteria:
[Paste de-identified info summarizing symptoms and relevant history, e.g., 'Client D reports persistent low mood, loss of interest in hobbies previously enjoyed, and difficulty concentrating over the past two weeks. De-identified GAD-7 score was 18. History includes a previous episode of depression two years ago.']"Prompt 5 (Goal Draft):
"Draft a measurable, time-limited treatment goal for Client E focused on managing stress, based on this de-identified description of their current challenge:
[Paste de-identified description, e.g., 'Client E wants to feel less overwhelmed by work deadlines and stop bringing work stress home every night.']"
Task: Brainstorming or Generating Ideas
Prompt 6 (Intervention Ideas):
"Client F is struggling with [De-identified broad issue, e.g., social isolation after a move]. Brainstorm 4-5 practical, concrete strategies for increasing social connection, drawing from principles of [Therapeutic modality, e.g., CBT]."
Prompt 7 (Psychoeducation Concept):
"Explain the concept of [Concept, e.g., the Window of Tolerance] in simple, metaphor-rich language suitable for an adult with no prior therapy experience. Keep it under 200 words."
Prompt 8 (Homework Ideas):
"Generate 3-4 potential homework assignments for a client working on [Topic, e.g., increasing self-compassion]. The assignments should be simple and take no more than 15 minutes per day."
Task: Drafting Practice Communications
Prompt 9 (Email Wording):
"Draft a polite and professional email template to a client about [Topic, e.g., an upcoming fee increase], explaining the reason concisely."
Prompt 10 (Policy Wording):
"Write a paragraph for a therapy practice's intake paperwork explaining [Topic, e.g., the cancellation policy] in clear, client-friendly terms."
Here are simple prompt templates for generating different therapy note formats using AI:
Prompt Template: SOAP Note (Simplified)
Goal: Generate a SOAP note from de-identified therapy session information.
Instructions: Read the de-identified session information provided below. Create a therapy note organized into the four standard SOAP sections: Subjective, Objective, Assessment, and Plan. Use ONLY the information explicitly mentioned in the provided text to populate these sections. Do not add or invent any details.
Format Guidelines:
Start with clear headers for each section: Subjective:, Objective:, Assessment:, Plan:
Follow each header with a summary of the relevant de-identified information from the provided text.
Present the information concisely.
Content Source: Use ONLY the de-identified text you paste below (can be de-identified session transcript, clinical notes, or contextual notes).
Writing Style:
Use clear, professional language.
Maintain an objective tone, especially in the Objective section.
Be concise and factual.
(Critical Constraint: Never fabricate or include information not explicitly mentioned in the provided text. Ensure the final output adheres strictly to the information you provided. Always review and edit the generated note.)
[Paste Your Fully De-identified Session Info Here]
Prompt Template: DAP Note (Simplified)
Goal: Generate a DAP note from de-identified therapy session information.
Instructions: Read the de-identified session information provided below. Create a therapy note organized into the three standard DAP sections: Data, Assessment, and Plan. Use ONLY the information explicitly mentioned in the provided text to populate these sections. Do not add or invent any details.
Format Guidelines:
Start with clear headers for each section: Data:, Assessment:, Plan:
Follow each header with a summary of the relevant de-identified information from the provided text.
Present the information concisely.
Content Source: Use ONLY the de-identified text you paste below (can be de-identified session transcript, clinical notes, or contextual notes).
Writing Style:
Use clear, professional language.
Maintain an objective tone in the Data section.
Be concise and factual.
(Critical Constraint: Never fabricate or include information not explicitly mentioned in the provided text. Ensure the final output adheres strictly to the information you provided. Always review and edit the generated note.)
[Paste Your Fully De-identified Session Info Here]
Prompt Template: GIRP Note (Simplified)
Goal: Generate a GIRP note from de-identified therapy session information.
Instructions: Read the de-identified session information provided below. Create a therapy note organized into the four standard GIRP sections: Goal, Intervention, Response, and Plan. Use ONLY the information explicitly mentioned in the provided text to populate these sections. Do not add or invent any details.
Format Guidelines:
Start with clear headers for each section: Goal:, Intervention:, Response:, Plan:
Follow each header with a summary of the relevant de-identified information from the provided text.
Present the information concisely.
Content Source: Use ONLY the de-identified text you paste below (can be de-identified session transcript, clinical notes, or contextual notes).
Writing Style:
Use clear, professional language.
Maintain an objective tone in the Response section.
Be concise and factual.
(Critical Constraint: Never fabricate or include information not explicitly mentioned in the provided text. Ensure the final output adheres strictly to the information you provided. Always review and edit the generated note.)
[Paste Your Fully De-identified Session Info Here]
Prompt Template: BIRP Note (Simplified)
Goal: Generate a BIRP note from de-identified therapy session information.
Instructions: Read the de-identified session information provided below. Create a therapy note organized into the four standard BIRP sections: Behavior, Intervention, Response, and Plan. Use ONLY the information explicitly mentioned in the provided text to populate these sections. Do not add or invent any details.
Format Guidelines:
Start with clear headers for each section: Behavior:, Intervention:, Response:, Plan:
Follow each header with a summary of the relevant de-identified information from the provided text.
Present the information concisely.
Content Source: Use ONLY the de-identified text you paste below (can be de-identified session transcript, clinical notes, or contextual notes).
Writing Style:
Use clear, professional language.
Maintain an objective tone in the Behavior and Response sections.
Be concise and factual.
(Critical Constraint: Never fabricate or include information not explicitly mentioned in the provided text. Ensure the final output adheres strictly to the information you provided. Always review and edit the generated note.)
[Paste Your Fully De-identified Session Info Here]
Refining Your Prompts (The "Tinkering")
Getting the perfect output from AI often requires a little back-and-forth.
Experiment: Try slightly different wording in your prompt. Does asking it to "Summarize" yield a different result than asking it to "Extract key points"?
Iterate: Use the AI's initial response as a starting point. You can then give follow-up instructions like:
"Make that summary shorter."
"Can you rephrase that using less clinical language?"
"Add a sentence about the importance of practice."
"Organize those points by date mentioned in the notes."
Treat the interaction like a conversation where you guide the AI towards the desired outcome.
Important Warnings & Limitations of AI
While helpful, AI is not a magic bullet and has significant limitations that are particularly important in the context of mental health:
Hallucination: AI can generate information that sounds plausible but is completely false or not supported by the de-identified context you provided. Always cross-reference the AI's output with the original de-identified information and use your clinical judgment.
No Clinical Judgment or Empathy: AI does not understand the nuances of the therapeutic relationship, non-verbal cues, or the complex emotional landscape of a client. It cannot replace your training, experience, or intuition. AI output is a starting point or a draft, never a final clinical decision or intervention.
Based Only on Input: The AI only knows what you give it in the current prompt. It doesn't "remember" past sessions (unless you include summaries in your prompt) or have access to the full, identifiable client file you hold.
Potential Bias: AI models are trained on vast amounts of text from the internet, which can contain biases. Be mindful of this and critically evaluate any AI-generated content for potentially biased or inappropriate language or suggestions.
Implementing AI ethically and efficiently is key to modern practice. Instead of piecing together tools, find everything you need for AI-powered notes, assistance, session follow-ups, and homework suggestions securely built into one integrated platform. Learn more about how Mynd, the AI-first practice management solution, streamlines your workflow.
Conclusion: AI as a Tool in Your Toolkit
AI tools, used responsibly and ethically, can indeed be valuable assistants in managing the administrative load and sparking ideas in your mental health practice.
The key is to master the art of prompting – giving clear, specific instructions – and to prioritize client privacy above all else through rigorous de-identification.
By understanding what AI is good at (processing text, summarizing, drafting based on patterns) and what it is not (clinical judgment, empathy, confidentiality), you can leverage its capabilities to potentially free up time, allowing you to focus on what truly matters: the human connection and skilled therapeutic work you provide.
Start simple, practice crafting clear prompts, de-identify absolutely everything client-related, and always, always apply your professional judgment to the output.
What are your biggest questions or hesitations about using AI in your practice? Share them below!