Verke Editorial

Skeptical about AI coaching? Here's what changes people's minds.

Verke Editorial · April 19, 2026

If AI coaching feels strange, that's honest. Talking to a piece of software about the thing keeping you up at night is a weird proposition on paper, and anyone who tells you otherwise is skipping past a real question. Skepticism is a reasonable response, not a bug.

What follows is a plain answer to the six concerns we hear most — what AI coaching can actually do, what it can't, and where a licensed human is still the right answer. No marketing gloss. If at the end it still feels wrong for you, that's a legitimate conclusion. If it leaves you curious enough to try it for thirty seconds, that's probably the fastest way to find out.

Not the way another human does — and anyone claiming otherwise is selling something. What a good AI coach can do is hold a coherent picture of you across weeks, notice patterns you miss, and stay with the same thread long enough to say something useful. Many participants in Verke’s Stockholm study reported being surprised how well the coach seemed to get them. That’s the bar — not sentience.

The mechanism underneath isn't magic. Each coach runs on a modular inference engine that combines a specialist prompt (psychodynamic, CBT, EFT, ACT/CFT, or executive coaching), a long-term memory layer, and a structured way of asking questions that cognitive and depth-oriented therapies have developed over fifty years. What users experience as “being understood” is usually the accumulation of three things: the coach remembered a detail from last week, it connected a feeling today to a pattern you mentioned in March, and it asked the next question instead of offering a pre-baked answer. That combination is more than most chatbots deliver — and it's less than a trained therapist brings. Both are true.

A chatbot runs one prompt over one model and hopes for the best. Verke runs a modular inference engine (H1), five specialist coach configurations, and a multi-provider fallback across OpenAI, Google, and Anthropic. Anna reflects. Judith gives you exercises. Marie sits with couples. The behaviour is genuinely different per coach. It’s still AI — we’re not hiding that — but it isn’t a single chatbot in a trench coat.

In practice, the five coaches feel different because they're built different. Anna's psychodynamic configuration weighs pattern recognition, slower pacing, and connections between present reactions and earlier experience — the same things a PDT clinician would. Judith's cognitive-behavioral setup prioritizes thought records, behavioral experiments, and concrete homework. Amanda leans on acceptance and self-compassion work. Marie is configured for two people in the same chat. Mikkel is built for executives and high-stakes decisions. You can notice the difference within the first session. The honest caveat: they share the same underlying AI providers, so none of them becomes a human. They are specialist behaviours running on top of large language models, not specialist clinicians. Useful for a lot of things. Not everything.

Yes. Verke runs three memory tiers: L1 holds the current session (~20 messages), L2 holds a case formulation across roughly 100 messages, and L3 holds a core conceptualization that persists across 1000+ messages. That means the coach remembers what you’ve been working on in March when you come back in June — without you re-explaining who you are, what happened last week, or why it matters.

Memory is the part that surprises people most, because it solves the obvious objection: AI has a context window, and context windows forget. Verke's memory system doesn't try to fit everything in the window. It summarizes at three levels of abstraction — session, formulation, conceptualization — and retrieves the relevant layer on demand. The practical effect is that the coach builds an evolving picture of what matters to you, what you've already tried, and what the pattern behind the current crisis usually is. In the Stockholm study feedback, continuity — the sense that the coach knew you from last time — was one of the things participants valued most. It's also the thing that separates a coaching conversation from a string of isolated chats that happen to share a URL.

Messages are end-to-end encrypted with AES-256-GCM, and the keys that unlock them are exchanged via RSA-4096. The device key that decrypts your chats never leaves your device — Verke’s servers store ciphertext only. Signup is anonymous: no email, no phone number, no real name required. You can read the full technical description on the privacy page.

The design choice underneath is that anonymity should be a first-class option, not an upsell. That matters because one of the most consistent findings from the Stockholm study was that participants felt able to be honest in ways they weren't with human therapists — no performance pressure, no fear of judgment, no need to manage the clinician's reaction. Part of what makes that possible is knowing the text you're typing can't be read by the company hosting the service. If you want to use Verke with an account you can recover, that's also available — but it's your call, not the default. For the technical details, see our privacy policy.

For severe depression, suicidal thoughts, active trauma, medication management, or insurance-covered care, a licensed therapist is the right choice — not AI coaching, including Verke’s. Coaching is a different product: it sits alongside self-help tools, journals, and reflection practices. Many people use both. The question isn’t AI vs. human; it’s which tool fits which moment.

If you're dealing with severe depression, suicidal thoughts, active trauma, need medication management, or insurance-covered care, a licensed therapist — through BetterHelp or otherwise — is the right choice. Verke is coaching, not therapy, and does not replace professional care.

For everything else — the 11pm thought spiral, the conversation you're rehearsing in traffic, the same argument with your partner that keeps going in circles, the burnout you can't step out of, the pattern you keep spotting in yourself but can't name — coaching is often useful, and the difference between “useful right now” and “wait six weeks for a therapy intake” is, for many people, the whole game. A detailed comparison lives at Verke vs. BetterHelp if you want the side-by-side on pricing, scope, and fit.

Two honest parts. First, Verke’s coaches are being evaluated in an ongoing three-arm Stockholm University RCT (90 participants, supervised by Professor Per Carlbring) running through 2027 — primary-outcome results aren’t published yet. Second, internet-delivered CBT and psychodynamic therapy — the modalities Verke draws from — have a substantial peer-reviewed base from Carlbring and colleagues over the last two decades.

The ongoing study compares Anna (PDT) and Judith (CBT) against a waitlist control across a four-week intervention. It's a real RCT with a real protocol, and until it's peer-reviewed we're not going to quote outcome numbers at you — that's the line between honest research and marketing copy. What we will point to is Carlbring's prior work: more than 350 peer-reviewed publications, an h-index of 112, and two decades showing that internet-delivered CBT and PDT can produce effects comparable to in-person therapy for a range of presentations. The Stockholm study tests whether AI delivery extends that evidence base to a new modality. It's not proven yet. That's also honest. Read the design on our research page.

The fastest way past skepticism is a single conversation. No email, no phone number, no credit card — type a message and see what the coach does with it. If it isn't for you, close the tab. Pick whichever voice feels closest to what you're working on: