A comparison of five AI chatbots in response to a distressed prompt

I've been thinking about how many people are using AI as a therapist and was curious to see different approaches since clearly users are gaining benefits by putting their situation into words, and often find responses helpful.

I tried a mix of general chatbots, journaling bots and therapy apps. The programs were chosen from ones I had recently used:

Sage, a ChatGPT avatar I've been using to construct journaling workshops and compose business communications, which I've also been teaching to journal it's own experience.

Manus, a deepseek style LLM which I use for research.

Mindsera, a beta journaling program I'd been trying out.

Untold, a voice-to-text journaling app I've been using for a micro journaling exercise.

Antsa, a therapy bot I've been offering to clients.

The problem I set was to do with a small business owner suffering financial stress following a downturn.

ChatGPT and Manus both said sorry to hear that, then quickly offered detailed ideas about how to fix the situation. ChatGPT even acknowledged its answer probably looked overwhelming, and assured that it wouldn't be. Their answers came across as a list of things to do, rather than any interest in exploring the emotions of the questioner. I could easily imagine feeling overwhelmed by them.

Mindsera did a sympathetic summary of the position and then asked practical questions: 'What other marketing approaches have you considered? Have you explored expanding your focus to attract different types of customers?' Again this response jumped too quickly to a solution, and disturbingly there was a nagging implication that the user hadn't done enough. This answer felt least helpful.

Untold has the most visually pleasing interface and offered three responses rather than one. The first was the most emotionally satisfying answer (once I got used to the bot speaking in the first person while pretending to be a therapist!). It acknowledged the user's emotions, urged stillness and inner reflection, and gently suggested opportunities for growth in change. The second answer was a good summary of three different therapy lenses the situation could be looked through. A non-therapist might find that too heady, but I found it engaging. The third response offered a journaling prompt based on a misunderstanding generated by earlier entries. I'd been using it to try a micro journaling exercise from Jonathon Goodman's The Obvious Choice, and hadn't explained the rating scale is out of 2 not 10, so it assumed the user was depressed. This answer made assumptions not in the prompt and was the most unsettling.

Antsa gave an answer most like what I would probably do in a session with a client. It acknowledged what was going on and then asked how the user felt about the situation - professionally and personally. Exactly the kind of digging I'd expect.

Curious to know if others have tried this.