Ai can organize your thoughts. only a human can believe in you.

The honest part: the chatbots do something

let’s start where most people won’t, because it’s true. a 2025 meta-analysis of ai-driven conversational agents for young people’s mental health (Feng et al., journal of medical internet research) found a measurable short-term reduction in depression symptoms (g=0.61) — and, worth stating plainly, a null result for generalized anxiety (g=0.06, confidence interval crossing zero). a separate 2024 meta-analysis of chatbots in short-course treatments (Zhong et al., journal of affective disorders) reached a similar conclusion.

so the anti-ai reflex — “it’s all fake, it does nothing” — isn’t honest. the effects are modest and short-course. but they’re real. pretending otherwise just makes people stop trusting you.

Where it stops: the mechanism a model can’t run

here’s the turn. the thing that holds a person up over time isn’t information delivery — it’s being witnessed by someone whose belief in you is costly, and therefore real. in a 2006 neuroimaging study (Coan, Schaefer & Davidson), people facing the threat of a shock showed a measurably calmer brain response when a supportive human simply held their hand — and the effect got stronger the closer the relationship.

a model can generate the words “i believe in you.” but belief you can’t disappoint, can’t burden, can’t actually be seen by, doesn’t regulate a nervous system. that’s not a bug you patch in the next version. it’s the whole thing.

What people actually say they want

and people already know this. in surveys, roughly half of adults cite concerns about ai’s lack of empathy and about privacy, and say they want a human in the loop. journals are now warning clinicians against attributing “empathy” to language models at all, because of the dependency risk.

read that generously, not smugly: people aren’t being fooled. they’re being underserved. the chatbot is what’s available at 2am when no human is. the fix isn’t to shame the tool — it’s to make humans available.

The both/and

use ai for what it’s good at: organizing a racing mind, drafting the hard text, naming what you’re feeling. then take that to a person. peer support — other people who’ve been where you are — keeps showing up in the research as a complement that improves hope and engagement. in a five-country randomized trial across six sites (UPSIDES, british journal of psychiatry, 2025), peer support improved measures of social inclusion, empowerment and hope against a wait-list — a real but mixed result: two of three subscales moved in the main analysis, and the overall social-inclusion score reached significance among those who actually attended. it’s not a rival to professional care. it’s the part professional care can’t be.

that’s the whole reason veisund exists — a free, anonymous place where the thing answering you at 2am is a person, not a prediction. i built it, so i’m not neutral, and i’ll name that. but the point holds no matter which app you open: the thing that helps isn’t the software. it’s a human who’s been where you are.