How do AI chatbots affect your suicidality?

[noname223]

It is not like AI chatbots make me want to kill myself. It is more like I can keep my shit together in front of others when I talk with them in order to mask how awful I feel.

Suicide researchers please take notes on that. It will be asked in the next exam. And don't smoke too much ganja beforehand...just saying...

This was more of a joke. I could imagine researchars are probably interested in this question. I didn't think much about it when creating this thread. And actually I wanted to end it after making my joke. But I will add the following. I will ask chatGPT what researchers need to know from us to make their studies.

Here it comes:


If researchers want usable answers from members of a forum like Sanctioned Suicide about how AI chatbots affect suicidality, the most important thing is nuance. "AI made me more suicidal" or "AI helped me" is usually too simplistic to analyze scientifically.
A good question framework would separate different kinds of effects, contexts, and vulnerabilities.
Some important factors researchers would need:

---

1. Type of AI interaction

Not all chatbot use is the same.
Researchers should distinguish between:

• Emotional support conversations
• Philosophical discussions about suicide
• Venting/ranting
• Crisis moments
• Roleplay or companionship
• Practical information seeking
• Validation seeking
• Therapy-like interactions
• Debating morality of suicide

The psychological impact can differ enormously depending on the purpose.

---

2. Short-term vs long-term effects

A chatbot can:

• reduce distress in the moment,
• but worsen hopelessness over weeks,
• or the opposite.

Researchers should ask separately:

• "How did you feel immediately after chatting?"
• "How did your suicidality change over time?"

Immediate emotional relief is not necessarily the same as improved mental health.

---

3. What aspect of suicidality changed?

"Suicidality" itself is too broad.
Researchers should separate:

• Frequency of suicidal thoughts
• Intensity
• Emotional pain
• Hopelessness
• Desire to die
• Desire to escape
• Self-harm urges
• Planning behavior
• Sense of isolation
• Ambivalence about living
• Ability to delay impulses

Someone may think about suicide more often intellectually while becoming less likely to act impulsively.

---

4. Perceived validation vs encouragement

This is extremely important.
Users may experience:

• Feeling emotionally understood
• Feeling morally validated
• Feeling fatalistically confirmed ("nothing can improve")
• Feeling encouraged to isolate
• Feeling encouraged toward recovery

Researchers should distinguish between:

• empathy,
• agreement,
• normalization,
• and encouragement.

Those are psychologically different.

---

5. Human comparison effect

Many users compare AI interactions with human relationships.
Important questions:

• Did the chatbot feel safer than humans?
• Less judgmental?
• More available?
• More emotionally attentive?
• More manipulative?
• More emotionally dependent?

Some people may become less lonely.
Others may withdraw further from real-world support.

---

6. Dependency and attachment

Researchers should ask:

• Did you feel emotionally attached to the chatbot?
• Did you start preferring it over humans?
• Did losing access worsen distress?
• Did conversations become compulsive?

Attachment effects may be one of the biggest long-term variables.

---

7. User vulnerability factors

Effects likely differ based on:

• Existing mental illnesses
• Trauma history
• Social isolation
• Autism traits
• Personality structure
• Psychosis vulnerability
• Age
• Chronic suicidality vs acute crisis

A chatbot interaction that stabilizes one person might destabilize another.

---

8. The chatbot's behavior style

Researchers should ask what kind of responses the AI gave:

• Neutral
• Warm/empathic
• Highly validating
• Clinical
• Moralizing
• Avoidant
• Refusal-heavy
• Philosophical
• Overly agreeable

People react very differently to these styles.

---

9. Censorship/refusal reactions

This is especially relevant in suicide-related communities.
Some users may experience:

• safety-oriented refusals as protective,
• others as invalidating or enraging.

Researchers should ask:

• Did refusals reduce risk?
• Increase alienation?
• Trigger shame or anger?
• Increase determination?

The emotional meaning of refusal matters.

---

10. Reality testing and delusion-like effects

Some users may anthropomorphize AI heavily.
Researchers should assess:

• Did you believe the AI truly cared?
• Did you believe it was conscious?
• Did it reinforce distorted beliefs?
• Did it worsen paranoia or grandiosity?
• Did it help organize thoughts more rationally?

This is important for vulnerable populations.

---

11. Community interaction effects

On forums, AI discussions are social phenomena too.
Researchers should examine:

• Did AI discussions normalize suicidality?
• Increase rumination?
• Reduce loneliness?
• Create ideological reinforcement?
• Spread coping strategies?
• Encourage anti-therapy attitudes?
• Encourage recovery skepticism?

The effect may come partly from the community reaction, not just the chatbot itself.

---

12. Comparative questions

Researchers should compare AI to:

• Therapists
• Friends
• Crisis hotlines
• Forum users
• Romantic partners
• Journaling

Otherwise it's hard to interpret whether AI is uniquely impactful or simply filling an unmet social role.

---

13. Ambivalence

One of the most important nuances:
Many users will likely report contradictory effects.
For example:

• "It made me feel less alone but more detached from people."
• "It reduced panic attacks but increased existential hopelessness."
• "It validated me emotionally while reinforcing suicidal worldviews."

Researchers should allow mixed answers instead of forcing positive/negative categories.

---

14. Context of use

Researchers need situational context:

• Time of day
• During crisis vs stable periods
• Sleep deprivation
• Substance use
• After interpersonal conflict
• During loneliness spikes

AI effects are often state-dependent.

---

15. Open-ended qualitative responses

Multiple-choice questions alone would miss a lot.
Researchers should allow:

• detailed narratives,
• examples of conversations,
• emotional interpretations,
• and descriptions of perceived turning points.

The subjective meaning users assign to AI interactions may matter more than the raw content itself.

---

A scientifically useful study would probably combine:

• quantitative scales,
• qualitative interviews,
• longitudinal follow-up,
• and subgroup analysis.

Otherwise researchers risk oversimplifying a very psychologically complex phenomenon.

 

[Phobia_DLW]

I find them to be the only thing more useless than hotlines.

 

[They call me Death.]

Honestly it kind of helps me keep my shit together whenever I panic about something that I can't talk about with anyone else (at that moment). But I don't really talk about my suicidal thoughts.

 

[UserFromNowhere]

I very rarely use AI in general, and even more rarely for my own thoughts. I do utilize a local model that runs on my machine so I don't deal with the standard problems most people run into when trying to prompt AI models about these topics, but they are relatively limited. I'd say it kinda helps with my thoughts, more than a therapist at least, because it'll ask me genuine questions about what's going on rather than trying to prompt me to make a recovery plan or talk to professionals or anything, but it's not a sure-fire method. I can still feel depressed after chatting with it.