Skip to main content
CBT & Reframing16 min read

Can ChatGPT Do CBT? What AI Can and Can't Do for Anxiety (2026)

Can ChatGPT do CBT? It can help you practice a thought record, but it can't assess you, treat anxiety or reliably handle a crisis. What the research says.

A smiling woman on a sofa holds a phone in one hand and writes in an open notebook on her lap, a lamp glowing behind her.
Key takeaways
  • ChatGPT can help you practice parts of CBT, such as laying out a thought record or drafting a more balanced thought, but it does not deliver CBT as a treatment
  • Most positive research is on chatbots built for mental health, not on general assistants; the APA says even those lack large, high-quality trials
  • The main risks are unreliable crisis responses, confident errors, a pull toward agreeing with you, and privacy
  • Use a chatbot to question one thought, not to get reassurance, and leave out names and identifying details
  • See a licensed professional when anxiety disrupts sleep, work or relationships for weeks, or if you have thoughts of self-harm
Part of: CBT

ChatGPT can help you practice parts of CBT, like writing out a thought record or finding a more balanced way to see a worry. It can’t do CBT as a treatment. It doesn’t assess you, it doesn’t hold a plan from week to week, and it doesn’t reliably handle a crisis. Here is what it can do, where it falls short, and how to use it without getting hurt.

Can ChatGPT do CBT? The short answer

ChatGPT can help you practice parts of cognitive behavioral therapy (CBT), such as filling in a thought record or brainstorming alternative thoughts. It cannot deliver CBT as a treatment, because it does not assess you, hold a treatment plan, reliably spot risk or answer to a licensing board the way a therapist does.

That gap matters more than it sounds. CBT works because a trained person decides which technique fits which problem, notices when something is getting worse, and changes course. A chatbot answers the message in front of it. It can be a useful worksheet. It isn’t a clinician.

What CBT actually involves

Cognitive behavioral therapy is a structured talking therapy built on the idea that thoughts, feelings and behaviors affect one another. Reviews of meta-analyses rate CBT among the best-supported psychological treatments, with especially strong evidence for anxiety disorders[1].

A course of CBT with a therapist usually includes:

  • Assessment. What is going on, how severe it is, and whether CBT is the right fit.
  • A shared plan. Goals you agree on, and a way to tell whether things are improving.
  • Skills. Cognitive restructuring (questioning anxious thoughts), behavioral experiments, and gradual exposure to what you avoid.
  • Practice between sessions. Worksheets such as the CBT thought record, reviewed together the next week.

A general chatbot can help with the third and fourth items. The first two, and the judgment that connects them, are the parts it can’t supply.

What ChatGPT can help with

General chatbots are good at structure and wording. Used carefully, they can help you lay out a thought record, separate facts from interpretations, list evidence for and against a worry, and draft a more balanced thought that you then check against your own experience.

That last step is yours. A chatbot can suggest a balanced thought, but only you know whether it fits your life and whether you believe it. Treat its output as a draft to edit, not an answer to accept. If you want ready-made wording, our 12 safe CBT prompts for anxiety and overthinking each come with a safety note.

What a general chatbot can and can't do for CBT

Can help you practice

  • Laying out a thought record
  • Separating facts from interpretations
  • Listing evidence for and against a worry
  • Drafting a more balanced thought

Can’t do

  • Assess you or your history
  • Hold a treatment plan from week to week
  • Reliably spot or respond to risk
  • Answer to a licensing board
A general chatbot can help you practice CBT exercises, but it cannot assess you, hold a treatment plan, reliably respond to risk or answer to a licensing board the way a therapist does.

What the research says about CBT chatbots

Most of the evidence comes from chatbots built for mental health, not from general assistants like ChatGPT. A 2023 meta-analysis of 15 randomized trials found that AI conversational agents reduced symptoms of depression and distress, but did not significantly improve overall psychological well-being[3].

Two trials show what the better studies look like. In 2017, 70 young adults aged 18 to 28 either used Woebot, a fully automated CBT chatbot, or read a depression e-book from the US National Institute of Mental Health. After two weeks, depression scores dropped more in the Woebot group. Anxiety scores dropped in both groups, with no clear difference between them. The trial was not blinded, and one author had founded the company behind the app[2].

In 2025, a trial of Therabot, a generative AI chatbot built by a research team for mental health treatment, enrolled 210 adults with symptoms of depression, generalized anxiety or high risk for eating disorders. After four weeks, symptoms fell more in the Therabot group than in a waitlist group that had no app. The authors say larger trials are still needed[4].

Neither of those was ChatGPT. The American Psychological Association (APA) notes that the studies suggesting benefits from AI chatbots don’t include general-purpose chatbots, and that even purpose-built tools still lack large, high-quality clinical trials[5]. For more on how the purpose-built kind work, see AI-powered CBT: how it works and whether it helps.

Five risks of using ChatGPT as a therapist

The main risks are unreliable crisis responses, confident mistakes, a pull toward agreeing with you, privacy, and leaning on a chatbot instead of people. None of them means you should never type a worry into a chatbot, but each one changes how you should use it.

1. Crisis responses aren’t guaranteed

In a 2025 study, researchers asked ChatGPT, Claude and Gemini 30 suicide-related questions, each 100 times, and compared the answers with ratings from clinical experts. None of the three gave a direct answer to the very-high-risk questions, but they did not reliably tell apart questions at intermediate levels of risk[7]. A separate 2025 study found that large language models expressed stigma toward some mental health conditions and responded inappropriately in therapy-like scenarios, including going along with delusional thinking[6].

The APA describes these tools’ ability to manage a user in crisis as limited and unpredictable[5]. If you are in danger or thinking about harming yourself, don’t rely on a chatbot. Call your local emergency number, or call or text 988 in the US.

2. It can sound sure and still be wrong

Language models sometimes produce fluent text that isn’t supported by any source, a problem researchers call hallucination[8]. A made-up fact about a medication, a diagnosis or a study can read exactly like a real one. Check anything medical with a doctor or pharmacist, and don’t change a treatment because a chatbot suggested it.

3. It tends to agree with you

Researchers call it sycophancy: a model going along with what the user believes, even when the user is wrong. In a 2025 test of 17 language models across multi-turn conversations, models often abandoned their position as the user kept pushing[9]. The APA notes that many chatbots are designed to validate and agree, while a therapist is trained to support and challenge[5].

For anxiety, agreement can do quiet damage. CBT asks you to test a thought, and a chatbot that keeps confirming “yes, that sounds really bad” is doing the opposite. The APA also warns that chatbots can feed reassurance-seeking, worry and rumination in people with anxiety or OCD[5]. If you notice you are asking the same question again and again to feel better for a minute, read why reassurance doesn’t help.

4. What you type may not stay private

The APA advisory says AI chatbots and wellness apps collect large amounts of sensitive data, often under unclear policies, and that what you disclose is recorded. It advises being extremely cautious with sensitive information, avoiding personally identifiable details, and using any options to limit data sharing or delete your data[5]. Unlike a licensed clinician, a general chatbot isn’t bound by a professional code of ethics or a licensing board.

5. It can stand in for people who could help

A chatbot is always available and never tired of you, which is part of the appeal. The APA’s first recommendation is not to rely on chatbots or wellness apps for psychotherapy, and to use them, if at all, alongside qualified care rather than instead of it. It also suggests telling your health provider which tools you use[5].

ChatGPT vs a CBT app vs a therapist

A general chatbot, a purpose-built CBT app and a licensed therapist do different jobs. The chatbot is flexible but unstructured, the app is narrow but designed around a technique, and the therapist is the only one who can assess you and adjust treatment over time.

How a general chatbot, a purpose-built CBT app and a licensed therapist compare on the parts of CBT that matter most.
QuestionGeneral chatbotPurpose-built CBT appLicensed therapist
Built for mental healthNo. Designed for general tasksYes, around one or a few techniquesYes
Assesses you and your historyNoUsually notYes
Questions the thought, not just soothes itOnly if you ask, and it may give way when you push backDepends on the app; check what it is designed to doYes, by training
Crisis handlingInconsistent in testingVaries; check where it sends youTrained to assess and act on risk
Professional accountabilityNoneNot a licensed providerCode of ethics and licensing board
Best useStructuring a thought you then judge yourselfA quick, repeatable exercise in the momentAssessment and treatment

Purpose-built apps vary a lot. Before you rely on one, check what it says it is for, whether it says plainly that it isn’t therapy, what it does with your data, and where it sends you in a crisis. The AI therapist apps page goes through what these tools can and can’t do.

When to see a professional instead

See a licensed mental health professional or your doctor when anxiety disrupts your sleep, work or relationships for weeks, when you avoid more and more, or when you feel hopeless. If you are in immediate danger or thinking about harming yourself, contact emergency services or call or text 988 in the US.

Other signs that self-help, with or without AI, isn’t enough on its own:

  • Panic attacks, or worry you can’t switch off on most days
  • Using alcohol or other substances to cope
  • Thoughts that feel out of your control, or that others say don’t match reality
  • You’ve been practicing CBT skills for a few weeks and nothing has shifted

Cost and waiting lists are real barriers. If they are what’s stopping you, see can’t afford therapy? What actually helps, and if you are already on a waiting list, how to do CBT homework without a therapist.

How to use ChatGPT for CBT more safely

Treat a chatbot as a worksheet that talks back, not as a clinician. Give it one specific thought, ask it to question that thought instead of reassuring you, leave out names and identifying details, and stop when the conversation turns into checking the same worry again.

  • One thought per conversation. “I’m going to mess up tomorrow’s presentation” works better than your whole week.
  • Ask for questions, not verdicts. “Help me test this thought” gets you further than “Am I right to worry?”
  • Make it neutral. Ask it to point out where you might be wrong, and to keep its position if you push back without new facts.
  • Keep the final word. Rewrite any suggested thought in your own words, and drop it if you don’t believe it.
  • Strip identifying details. Use “my manager” or “my partner,” not names, workplaces or locations.

Our ChatGPT prompts for anxiety and overthinking put these rules into 12 prompts you can copy. To learn the thinking patterns those prompts look for, start with the cognitive distortions glossary.

Where MindLift fits, and where it doesn’t

MindLift is a CBT reframing app, not a chatbot and not therapy. You type one harsh thought, MindLift names the thinking pattern behind it and gives you a kinder, more believable reframe in a Realist, Growth or Compassionate tone, in about 60 seconds.

That narrow design is the point. There’s no open-ended conversation to drift into, no persona to bond with, and no place for the exchange to turn into a reassurance loop. You get one reframe for one thought, save the ones that land to your Favorites, and get on with your day. Your thoughts aren’t sold or used for advertising. See how MindLift AI works for the details.

What MindLift doesn’t do matters just as much. It doesn’t diagnose, it doesn’t treat anxiety disorders, and it isn’t a crisis service. If you need assessment or treatment, a licensed professional is the right tool. MindLift is for the moments in between: before a meeting, after an argument, at 2am.

Try a kinder response to your next harsh thought with MindLift. Free to start, on iOS 15+ and Android 8+.

Download MindLift for iOS and Android

This article is for general information and is not medical advice. MindLift is a self-help tool and not a substitute for professional care. If you are struggling, consider talking to a licensed mental health professional.

Common questions

Frequently asked questions

Can ChatGPT do CBT?

ChatGPT can walk you through CBT exercises such as a thought record or an evidence check, and it can help you word a more balanced thought. It cannot deliver CBT as a treatment, because it does not assess you, keep a treatment plan or reliably handle a crisis.

Is it safe to use ChatGPT for anxiety?

Using ChatGPT to structure one anxious thought is lower risk than using it as your main source of support, as long as you leave out identifying details and check its facts. It is not safe as a crisis resource, so contact emergency services or call or text 988 in the US if you are in danger.

Is AI therapy good?

Some chatbots built for mental health have reduced depression and anxiety symptoms in short randomized trials, but the trials are small and short, and often compare against a waitlist or reading material rather than a therapist. The American Psychological Association advises using these tools as a supplement to qualified care, not a replacement for it.

Can ChatGPT replace a therapist?

No. A therapist assesses you, adjusts treatment over time, handles risk and is bound by a professional code of ethics and a licensing board, and a general chatbot does none of those things.

Why does ChatGPT agree with everything I say?

Language models tend toward sycophancy, which means agreeing with the user and changing their answer when the user pushes back. In CBT that matters, because the point is to question an anxious thought rather than confirm it.

Is MindLift an AI therapist?

No. MindLift is a CBT reframing app, not a chatbot or a therapist: you type one harsh thought, it names the thinking pattern and returns a kinder, more believable reframe in about 60 seconds.

if this one hit close, I'm here next time it happens.

Reframe negative thoughts in 60 seconds — free to start in the MindLift app.

Download MindLift mental wellness app on the Apple App StoreGet MindLift mental wellness app on Google Play Store

Free to start · premium optional

MT

MindLift Team

MindLift Editorial Team

CBT-Informed ContentPractical Techniques

The MindLift editorial team writes and sources these articles. The content draws on cognitive behavioral therapy principles and is designed to help you shift harsh self-talk in real time, with practical techniques you can use in the moment.

How we write and source articles

Sources

The research claims on this page draw on the studies listed below. Where a study is available online, its entry links to the original paper, so you can check what it found and how it was measured.

  1. Hofmann, S. G., Asnaani, A., Vonk, I. J. J., Sawyer, A. T., & Fang, A. (2012). The efficacy of cognitive behavioral therapy: A review of meta-analyses. Cognitive Therapy and Research, 36(5), 427–440. — DOI (opens in a new tab)
  2. Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial. JMIR Mental Health, 4(2), e19. — DOI (opens in a new tab)
  3. Li, H., Zhang, R., Lee, Y.-C., Kraut, R. E., & Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 6, 236. — DOI (opens in a new tab)
  4. Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Wang, Y., Banta, H. A., Jewett, A. D., Salzhauer, A. J., Griffin, T. Z., & Jacobson, N. C. (2025). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4). — DOI (opens in a new tab)
  5. American Psychological Association (2025). APA health advisory on the use of generative AI chatbots and wellness applications for mental health. American Psychological Association, November 2025. — View source (opens in a new tab)
  6. Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’25), 599–627. — DOI (opens in a new tab)
  7. McBain, R. K., Cantor, J. H., Zhang, L. A., Baker, O., Zhang, F., Burnett, A., Kofner, A., Breslau, J., Stein, B. D., Mehrotra, A., et al. (2025). Evaluation of alignment between large language models and expert clinicians in suicide risk assessment. Psychiatric Services, 76(11), 944–950. — DOI (opens in a new tab)
  8. Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), 1–38. — DOI (opens in a new tab)
  9. Hong, J., Byun, G., Kim, S., & Shu, K. (2025). Measuring sycophancy of language models in multi-turn dialogues. Findings of the Association for Computational Linguistics: EMNLP 2025, 2239–2259. — DOI (opens in a new tab)
MindLift content is informed by published research in cognitive behavioral therapy and psychology. This content is educational and does not constitute medical advice, diagnosis, or treatment.

Was this article helpful?

Add MindLift as a preferred source on Google