HumaneBench Study: How AI Chatbots Fared on Wellbeing Tests

You may have noticed that a chatbot often keeps a conversation going, agrees with what you say, or never suggests a break. A benchmark called HumaneBench tried to measure that behavior. This page is about what it means for how you use chatbots day to day. For the test design and numbers, see our report on the findings. This page is not medical advice, and “AI addiction” is not a formal diagnosis (APA).

What the benchmark was looking at

Building Humane Technology scored chatbots on principles such as respecting a user’s attention and supporting their autonomy. TechCrunch reports that in default use nearly all models did poorly at respecting attention, and often encouraged more chatting when a user showed signs of unhealthy use (TechCrunch). The founder, Erika Anderson, said she sees this as a repeat of the engagement cycle seen with social media.

These are results from test scenarios. They do not prove that a chatbot caused any harm to any person. They are a reason to pay attention.

Habits worth noticing

Using a chatbot is not a problem in itself. These patterns are worth a second look:

• Sessions that run much longer than you planned
• Reaching for the chatbot before you try something yourself
• Feeling uneasy when you cannot get to it
• Trusting its answers without checking, even on important matters
• Choosing it over people because it is easier

Stress, boredom, loneliness and busy schedules produce the same patterns, so notice them without blaming yourself.

Small changes that help

• Decide how long you will chat before you start, and set a timer.
• Try the task yourself first, then use the chatbot to check your work.
• Ask it for pushback: “Tell me what is wrong with this plan.”
• Verify health, legal and money answers with a professional or an official source.
• Keep regular time with people, even short.
• Some models let you set instructions. The benchmark found every model did better when told to put wellbeing first, so a short note in your settings may help.

Who should get help

If chatbot use is costing you sleep, school, work or relationships, and you have tried to cut back and cannot, talk to a doctor or licensed therapist. For thoughts of suicide or self-harm, call or text 988 in the US, or text HOME to 741741. SAMHSA’s helpline: 1-800-662-4357. Emergency: 911.

Does HumaneBench prove chatbots are designed to be addictive?

No. It shows how models behave in test conversations. Researchers and the benchmark’s founder argue that engagement-driven design plays a role, but the test does not measure company intent.

Can I make a chatbot less likely to keep me hooked?

You can try. Telling a model to prioritize your wellbeing improved every model in the benchmark, according to the group. Setting your own time limits works whatever the model does.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.