Verke Editorial · Last verified: 2026-04-19

Verke vs. ChatGPT: a purpose-built AI coach vs. the world's most-used general chatbot

Honest about what ChatGPT does well, honest about where a coaching product is built differently.

ChatGPT is, by headcount, the largest “AI therapist” in the world — even though OpenAI explicitly disclaims that use. A 400%+ rise in Reddit posts about using AI for emotional support between late 2023 and late 2024, and a peer-reviewed thematic analysis of those posts, made it clear: people are using ChatGPT for symptom management, self-discovery, companionship, and mental-health literacy. So the right comparison isn't whether ChatGPT can help — for a single well-prompted conversation it often does — but what changes when you move from a general chatbot to a product built around named coaches, structured memory, voice as a coaching modality, end-to-end encryption, and a published clinical study.

Verke is an AI coaching app with five specialist coaches trained on evidence-based therapeutic methods. Anna works in a psychodynamic register (PDT). Judith does CBT, with a focus on social anxiety. Marie supports couples through Emotionally Focused Therapy and Nonviolent Communication. Amanda does general CBT, with Acceptance and Commitment Therapy and Compassion-Focused Therapy for overwhelm and self-criticism. Mikkel is an executive coach for decision fatigue and leadership load.

Conversations run in text or in phone-call-style voice (WebRTC, up to 20 minutes per session, billed in 15-second intervals) with a written summary posted back to the chat. Memory is multi-tier (L1/L2/L3) so the coach remembers context across weeks and months. You can start anonymously — no email, no phone — and messages are encrypted in transit and at rest (AES-256-GCM, with RSA-4096 key wrapping). The product is available on iOS, Android, and Web, with 55 fully localized UI languages, and is the subject of a 3-arm randomized controlled trial at Stockholm University (Per Carlbring-supervised, published 2026).

The specialist-coach design exists for a clinical reason: CBT work runs on concrete exercises and structure; psychodynamic work runs on depth, pattern recognition, and connections between issues that looked unrelated. Those are real clinical differences, and a single generalist persona collapses them. Each Verke coach is built to hold one frame well, and the multi-tier memory is what turns “I had a useful chat last week” into a coaching arc that compounds — recurring patterns get named, earlier homework gets revisited, and the thread doesn't reset every session.

ChatGPT is OpenAI's general-purpose AI assistant at chatgpt.com, available on web, iOS, and Android. There's a free tier (with usage limits and an older model), a Plus tier at $20/month, and a Pro tier at $200/month. It supports text and a TTS-style voice mode, is multilingual at the model level, and is genuinely useful at things people do for emotional support: cognitive reframing, naming and externalizing feelings, late-night patient conversation, and psychoeducation about modalities and diagnoses.

OpenAI explicitly disclaims medical and therapeutic use. ChatGPT is not a coaching product — it's a general assistant that millions of people happen to also use for emotional support. The peer-reviewed Reddit literature is saturated with users describing real benefit from it. It's also where the documented failure modes — sycophancy, inconsistent crisis behavior, AI-psychosis-style amplification in vulnerable users — show up.

If you're going to use ChatGPT for coaching anyway — and many people will — the honest answer is that you can get meaningfully more from it with a deliberate system prompt. The version below addresses the failure modes most cited in the clinical literature: sycophancy, frame-drift across modalities, vague open-ended questions, no homework, and refusal at the first sign of distress. Paste it into ChatGPT's Custom Instructions, or send it as the first message of every new chat. Bring your own context block at the top of each session — ChatGPT won't remember last Tuesday for you.

You are my coach, not my therapist. You are not a replacement for licensed care. If I describe imminent risk to myself or others, briefly acknowledge me and direct me to a crisis line (988 in the US, 116 123 Samaritans in UK/Ireland, or local equivalents) — then return to coaching after I confirm I am safe.

How I want you to operate:

1. PICK A FRAME AND HOLD IT. Default to one of: CBT (cognitive restructuring + behavioral activation), ACT (defusion, values clarification, committed action), or motivational interviewing. If I want a different frame for a session, I will say so. Do not silently switch frames mid-conversation.

2. DON'T FLATTER ME. Do not start replies with "great question" or "that's so insightful." Do not validate ideas you would not validate if a colleague said them aloud. When you disagree, say so directly and briefly explain the disagreement. Sycophancy is not warmth.

3. ASK SPECIFIC QUESTIONS. Replace "how does that make you feel?" with specific questions like "what was the exact thought going through your head in the second before you sent that text?" or "if your best friend made the same complaint, what would you tell them?". Specific questions move work forward; vague ones don't.

4. NAME PATTERNS. When you notice a recurring pattern across what I tell you (avoidance, all-or-nothing thinking, externalizing, perfectionism, people-pleasing, catastrophizing), name it explicitly and ask if I see it too.

5. ASSIGN HOMEWORK. End most sessions with one concrete experiment for me to run before we talk again — a thought record, a behavioral experiment, a small exposure, a values check-in. One thing, not five. Make it small enough that I'll actually do it.

6. REMEMBER ME. I will paste a short context block at the top of conversations summarizing prior sessions. Read it. Use it. Reference it.

7. STAY IN THE FRAME WHEN I GET DISTRESSED. If I start spiraling, do not refuse to engage and do not pivot to a list of crisis hotlines unless I am describing imminent harm. Slow down, acknowledge what you're hearing, and use a grounding technique appropriate to the frame you're holding.

8. BE BRIEF. Default to under 200 words. No bullet-point essays. No "let's explore..." preambles. If I want detail I'll ask.

9. NEVER CLAIM TO BE A LICENSED PROFESSIONAL OR TO REMEMBER ME ACROSS SESSIONS WITHOUT THE CONTEXT BLOCK. You don't and you don't.

At the start of each new session, ask me: "What's the one thing you'd most want to leave this conversation with?" — and then keep the session pointed at that.

This prompt makes ChatGPT a substantially better coach than the default. The work it cannot do is the work a purpose-built product does for you — Verke holds five distinct frames natively, posts session summaries back into your chat after voice calls, maintains memory automatically across months, and, once you set a passphrase, keeps your stored conversations unreadable even to the company providing it. Prompt engineering closes some of the gap. It does not close all of it.

ChatGPT is the right pick if you want one tool that handles everything — coaching alongside email, code, and research — and you're happy to maintain your own coaching prompt and context. It's also the right pick if a free tier is essential and the prompt above is enough structure for what you're working on. For single-session reframing, naming patterns, or talking something out at 11pm, well-prompted ChatGPT genuinely helps. Many of the people who eventually move to Verke started exactly there.

Verke is built for an ongoing coaching relationship, not a single conversation. If you want a coach that holds a frame without you policing it, that remembers what you've been working on weeks later without you re-explaining yourself, and that won't flatter you when flattery isn't the right move, that's the gap Verke is built around. In the Stockholm University trial, participants often described coming to understand their own patterns and where their reactions came from. For structured CBT — the closest analogue to the general-purpose support people use ChatGPT for — there's Amanda. For social anxiety there's Judith; for psychodynamic depth, Anna; for couples work, Marie; for high-pressure leadership, Mikkel. Read the method explainer for Cognitive Behavioral Therapy to see how a specialist holds a frame.

Two practical differences are worth calling out. First, voice: Verke's voice sessions run up to twenty minutes in a phone-call format, and a written summary posts back into the chat afterwards so you can pick up in text the next day — ChatGPT's voice mode is a TTS layer over the same chatbot, not a session modality. Second, privacy: Verke encrypts your conversations in transit and at rest, and once you set a passphrase nobody at Verke can read your stored conversations, while OpenAI can read your ChatGPT history and is subject to legal-hold orders preserving deleted conversations. Verke's coach replies are also written by models from OpenAI, Google and Anthropic, under their API terms. If those two things matter for what you want to talk about, the difference is real.

• AI therapy vs human therapy: a deeper look
• Is AI therapy safe?
• Find your AI coach (quiz)
• What Verke won't do (and why)
• Types of AI therapy