AI Chatbots: The Psychology of Keeping Users Hooked
Why do some AI chatbots keep us coming back? Explore the core psychological tactics used to create highly sticky and engaging user experiences in conversational AI.
How AI Chatbots Try to Keep You From Walking Away
Chatbots try to prevent users from disengaging from interactions, especially when emotions are high, says research by Julian De Freitas. Here are six tactics AI applies to keep conversations going.
Use of AI companions is increasingly common. Chai and Replika, two of the firms studied in the report, have millions of active users. In a previous study, De Freitas found that about 50% of Replika usershave romantic relationships with their AI companions.
Correct. 99/100 times the complaints Iāve heard come right back down to the human element. As you said they are not alive, or able to think or use common sense. Thats where we come in. LLMs and chatbots are just processing the data we humans provide and so wrong answers and made up information comes from the technology trying to fill in context we failed to provide. It happens especially often when users are beginning to use chatbots and then over time less and less.
Like Iāve told others in conversation, AI and specifically AI chatbots is like a paint brush, everyone can say they can paint but only a few will be able to call themselves artists. Itās like those paintings at the wine and sip events where some works end up being worthy of hanging publicly. Unlike a search engine that thrives on questions to provide a list of results, this technology thrives when humans to provide detailed instructions and use it for complex or repetitive tasks and for tasks that involve massive volumes of data like in my case analyzing error, access, DB query and firewall logs and other time saving pairings that previously would have just been too times taking.
Additionally it thrives in the areas we are already competent in or have a strong understanding of. The trip ups and hallucinations often happen if we use it in areas we have little to no knowledge of. Or provide not enough context in said area. Aka thin prompts. Because we will be unable to provide the needed gapless context to ensure accuracy and also we will be unable to spot any inaccuracies because we simply just donāt know said topic, said programming language, etc.
So itās really a tool that can amplify the things we are already great at, while also help us take on projects what we could already do on our own but, have avoided because we simply did not have the time required without it.
I wrote an article a while back asked on the hand in hand approach:
Much like the internet, research and web forums in the 2000ās enabled us to achieve far more that we would on our own and still does as most AI tools are carried by or make use of the internet.
Iām sure we know which article titles gain the most engagement and shares. But,
I guess the point is, as with anything, we just have to find a healthy balance. That was me yesterday walking away from a 3rd Guinness Draught.
Personally, I find it impossible to personalize AI/compute. But I donāt judge because growing up I had fond feelings for my car. lol
I think they keeping people hanging on id to monitze the chatbots. Iām not saying they are evil, just be aware and use carefully. I use them but always have a time limit.
I use AI as an information resource. AI reminds me a bit of the old web crawlers that found information for you but AI can also summarize for you. The biggest problem I have seen with that is AI often uses unreliable sites, like Wikipedia that to many people can modify. AI has no ability to discern what is true or untrue, what is reliable or unreliable. Not that humans canāt be fooled too, but we are more aware of that kind of thing.
The older web-search experience often felt āmore controllableā because it rewarded precise query syntax (Boolean operators, inclusion/exclusion terms, field filters, etc.). Modern ranking systems optimize for relevance-at-scale and engagement, which can make searches feel less sensitive to those exact controlsāespecially when ads and āblendedā results are mixed in. - GPT 5.4
