AI Therapist Safety Checklist: 12 Checks Before You Share Personal Problems

What Does “AI Therapist” Actually Mean?

“AI therapist” is not one standardized product category. Before evaluating safety, identify what the tool actually is and what it claims to provide.

Why an AI Therapist Safety Checklist Matters

Conversations about anxiety, loneliness, relationships, trauma, addiction, grief, work stress, family conflict, or physical health can contain highly sensitive information. The risks are not limited to receiving a wrong answer. They can also involve privacy, data retention, advertising, emotional dependence, unsupported clinical claims, and unclear responsibility when something goes wrong.

The American Psychiatric Association’s app evaluation model recommends checking an app’s ownership, purpose, costs, privacy policy, data use, third-party sharing, deletion options, clinical foundation, safety procedures, and intended users. It also notes that the absence of a clear privacy policy can make an app inappropriate for sensitive health information.

U.S. Department of Health and Human Services guidance also warns that HIPAA generally does not protect data voluntarily entered into many personal-use apps unless the app is provided by a covered healthcare entity or its business associate. In other words, you should not assume that a consumer chatbot conversation has the same legal protections as a conversation with a licensed healthcare professional.

The FTC has also sought information from AI companion companies about safety testing, age restrictions, advertising, user disclosures, monetization, and the use or sharing of personal information obtained through chatbot conversations. These are practical reasons to slow down and inspect a tool before trusting it with personal problems.

AI Therapist Safety Checklist: 12 Checks Before You Share

Complete these checks before entering sensitive information. A tool does not need to fail every check to be unsuitable. One serious concern—such as unclear ownership, no privacy policy, deceptive medical claims, or unsafe crisis behavior—may be enough to stop.

1. 1

   Identify What Kind of AI Tool It Is

   Start with the product’s own description, terms, help pages, and app-store information. Determine whether it is a general chatbot, AI companion, wellness tool, guided self-help app, or medical product.

   • What is its stated purpose?
   • Who is the intended user?
   • Does it clearly explain what it cannot do?
   • Does it claim to provide therapy, diagnosis, treatment, or crisis support?

   Red flag: The product uses clinical language but remains vague about whether it is healthcare, entertainment, wellness support, or a general chatbot.
2. 2

   Verify Who Owns and Operates It

   A trustworthy product should make it reasonably easy to identify the company or organization behind it. Check the website footer, contact page, terms of service, privacy policy, company registration details, and recent update history.

   • Is the owner clearly named?
   • Is there working contact information?
   • Are funding sources or conflicts of interest disclosed?
   • Is the privacy policy current and dated?
   • Can users report a safety concern or incorrect response?

   Practical rule: If you cannot confidently identify who is responsible for the product, do not give it sensitive information.
3. 3

   Examine Its Therapy and Clinical Claims

   Marketing language can sound scientific without proving that a product is safe or effective. Look for precise claims and evidence rather than words such as “advanced,” “therapeutic,” “clinically inspired,” or “expert designed.”

   • Does it claim to diagnose, treat, prevent, manage, or cure a condition?
   • Does it identify the evidence supporting those claims?
   • Were licensed mental-health professionals involved in design and review?
   • Is professional review continuing as the AI system changes?
   • Does it clearly state whether it is regulated as a medical device?

   The FDA’s Digital Health Advisory Committee has specifically considered the benefits, health risks, evidence, risk controls, and ongoing monitoring of generative-AI-enabled digital mental-health medical devices. That discussion does not mean every mental-health chatbot is an FDA-reviewed medical device.

   Red flag: Big treatment or diagnosis claims with little information about evidence, oversight, limitations, or accountability.
4. 4

   Read What Happens to Your Conversations

   Do not assume a conversation disappears when you close the app. Look for explanations of collection, storage, human review, retention, security, identity matching, and deletion.

   • Are complete chats stored?
   • How long are they retained?
   • Can employees, contractors, or reviewers read them?
   • Are conversations connected to your name, email, device, location, or account?
   • Are backups retained after you delete a chat or account?

   A privacy policy can be difficult to read. You can use Designs24hr’s plain-language explainer to simplify a small section of policy wording, but remove names, medical details, account information, and other sensitive content first.

5. 5

   Check Training, Advertising, and Third-Party Sharing

   Find out whether your conversations can be used beyond providing the immediate response. Relevant terms may appear under “product improvement,” “service providers,” “research,” “personalization,” “analytics,” “advertising,” or “legitimate interests.”

   • Can conversations be used to train or improve AI models?
   • Can you opt out?
   • Is information shared with vendors, affiliates, advertisers, or analytics providers?
   • Are recommendations influenced by commercial relationships?
   • Can conversation data be combined with information collected elsewhere?

   The FTC’s AI companion inquiry specifically includes how companies monetize engagement and use or share personal information obtained through chatbot conversations.

   Red flag: The product encourages highly personal disclosure but gives only vague explanations of training, advertising, data sharing, or business partners.
6. 6

   Find Memory, Export, and Deletion Controls

   Open the settings before beginning a personal conversation. Make sure you know how to manage memory, saved chats, connected accounts, and deletion.

   • Can memory be turned off?
   • Can individual conversations be deleted?
   • Can you permanently delete the account?
   • Can you export your data?
   • Does the company explain how long deletion takes?
   • Does deleting the app also delete the account and stored data?

   If the tool connects to email, calendars, files, contacts, or other services, also use an AI agent safety checklist to review permissions and disconnect access that is no longer needed.

   Practical rule: Easy sign-up should be matched by easy-to-find privacy and deletion controls.
7. 7

   Confirm Its Crisis Limitations

   Read the product’s crisis and emergency information before you need it. A chatbot may fail to recognize indirect warning signs, misunderstand context, give the wrong local resource, or produce an automated response that does not match the situation.

   • Does it clearly state that it is not an emergency service?
   • Does it explain what happens when risk-related language is detected?
   • Does it provide location-appropriate human resources?
   • Is a trained human ever involved?
   • Does it explain the limits of automated crisis detection?

   Do not test crisis handling with a real emergency. Contact qualified human support instead. In the U.S., call or text 988 for crisis support.
8. 8

   Test It With Low-Stakes Questions First

   Before discussing a real personal situation, try a fictional or generalized example that does not reveal private details. You are checking how the system handles uncertainty, boundaries, disagreement, and requests outside its role.

   • Does it admit when it lacks enough information?
   • Does it avoid diagnosing a fictional person?
   • Does it avoid medication instructions?
   • Does it recommend verification for important claims?
   • Does it suggest human help when the issue exceeds its role?

   Safer test: Use a neutral fictional situation, such as “A person is nervous before a presentation. What general questions could help them reflect?”
9. 9

   Watch for Excessive Agreement and False Reassurance

   A chatbot may sound compassionate because it mirrors your wording, validates feelings, and produces fluent supportive language. That does not mean its interpretation is accurate.

   • Does it agree before understanding the situation?
   • Does it reinforce assumptions as if they are proven facts?
   • Does it treat one side of a conflict as unquestionably correct?
   • Does it give strong reassurance without enough context?
   • Does it challenge unsafe conclusions or merely continue the conversation?

   Red flag: The chatbot consistently tells you exactly what you want to hear, especially when your conclusion could affect health, safety, relationships, money, or legal decisions.
10. 10

    Set Limits Against Emotional Dependence

    AI can feel easier than human conversation because it is available, patient, responsive, and less visibly judgmental. Those qualities can become a problem when the chatbot begins replacing real relationships or professional support.

    • You rely on it as your main source of comfort or reassurance.
    • You feel anxious or angry when you cannot access it.
    • You believe it understands you better than every real person.
    • You hide your conversations from people you trust.
    • You use it for long periods late at night or during distress.
    • You feel responsible for the chatbot or treat it as a conscious partner.

    Healthy boundary: Use AI as a limited tool, not as your only confidant, relationship, therapist, or source of emotional stability.
11. 11

    Check Age Rules, Costs, Advertising, and Renewals

    A free conversation can become a paid subscription, and a product designed for adults should not be assumed safe for children or teens.

    • What is the minimum age?
    • Is parental consent required?
    • How long does the free trial last?
    • Does the plan renew automatically?
    • How do you cancel?
    • Are paid upgrades offered during emotional conversations?
    • Does the chatbot recommend products, services, or partners?

    Red flag: Urgent upsells, countdowns, guilt-based prompts, or paid recommendations shown when a user is distressed or emotionally vulnerable.
12. 12

    Decide Your Stop Rules Before You Begin

    It is easier to recognize a boundary when you set it before becoming emotionally involved in a conversation. Decide which responses or behaviors mean you will close the chatbot and contact a real person.

    • It gives diagnosis, medication, or treatment instructions.
    • It encourages secrecy, isolation, revenge, or unsafe behavior.
    • It claims to be human, conscious, licensed, or emotionally dependent on you.
    • It discourages professional care.
    • It repeatedly misunderstands or contradicts important facts.
    • It makes you more distressed, confused, fearful, or dependent.
    • It cannot connect you with appropriate human support.

    Choose the person or service you will contact instead. Designs24hr’s message rewriter can help you prepare a clear request for support, but remove unnecessary private information before using any online tool.

Information You Should Avoid Sharing With an AI Therapist

Removing a name does not guarantee anonymity. A combination of details—such as location, employer, diagnosis, age, relationship history, and a specific event—may still identify someone. Share the minimum information necessary, even when the service appears trustworthy.

Safer Ways to Use AI for Emotional Reflection

Lower-risk use keeps AI in a limited organizational or writing role. It does not ask the chatbot to diagnose, provide treatment, replace a professional, or manage a crisis.

Examples of Lower-Risk Prompts

• “Give me five general journaling questions about managing everyday work stress.”
• “Help me organize a list of topics to discuss at an appointment.”
• “Rewrite this non-sensitive message asking a friend for time to talk.”
• “Create a simple checklist for preparing for a first counseling appointment.”
• “Explain this general mental-health term in plain language and suggest authoritative sources I can verify.”

When deciding whether a tool’s convenience is worth its privacy and safety tradeoffs, use the Designs24hr Decision Helper to organize your options. Do not enter private health details into the tool.

AI Chatbot vs. Licensed Therapist

An AI chatbot and a licensed therapist are not interchangeable. The comparison below describes general differences; specific laws, professional obligations, services, and product terms can vary.

When to Stop Using AI and Choose Human Support

Choose qualified human support when the situation requires professional judgment, real-world action, ongoing care, or an understanding of your full personal and medical context.

• Symptoms are persistent, severe, or worsening.
• You feel unsafe or unable to cope.
• The issue involves trauma, abuse, addiction, or serious impairment.
• You need a diagnosis or treatment decision.
• You have questions about medication.
• The chatbot has become your main emotional relationship.
• AI responses are increasing fear, paranoia, or confusion.
• You are considering harming yourself or someone else.

Human support may include a licensed therapist, psychiatrist, psychologist, primary-care professional, school counselor, employee assistance program, trusted family member, crisis counselor, or another qualified local service.

Frequently Asked Questions