Personalizing a chatbot makes you trust it more and worry about your privacy more — at the same time, for the same reason.

This explores whether personalization makes people trust a chatbot more and worry more about their privacy at the same time. The short answer from the collection is yes, with an important complication. Longitudinal research finds personalization raising trust and anthropomorphism (treating the bot as human-like) while also raising privacy concerns Does chatbot personalization build trust or expose privacy risks?. It also finds a third, less obvious effect: every good interaction raises the user's baseline expectations, so later failures disappoint more. The broader personalization literature describes the same split. Memory, personas, and preference modeling are the features that build trust, and they are also the ones that make an AI more persuasive and open the door to manipulation Does personalization in AI increase trust or manipulation risk?.

The complication is that the trust half of this story may be weaker than it looks. A 5-day randomized trial with 992 participants found that people did feel closer to chatbots and rated them as more competent over time. But that happened whether or not the bot was personalized, and personalization itself had no measurable effect on closeness, trust, or how much people disclosed Does personalization actually build closeness with chatbots?. Much of what looks like personalization building trust may simply be familiarity. Trust in chatbots also seems to come from surface features rather than from feeling understood: conversational back-and-forth Does conversational style actually make AI more trustworthy? and language that sounds expert Does chatbot language style actually shape how much we trust it? both build trust regardless of accuracy. Timing matters too. Novelty effects in chatbot relationships fade predictably Do chatbot relationships lose their appeal as novelty wears off?, so a single-session study can mistake early excitement for lasting trust.

The privacy half is more solid, but it is often hidden from users. In one study, most of the 40 participants never noticed that their LLM's memory had stored or inferred private details about them. Once they were shown, every participant wanted the ability to edit and delete those memories Do users notice when LLMs infer and retain private information?. Those inferences are also often wrong. Every one of 12 models tested made claims about users that went beyond the evidence in 35–49% of cases Do large language models fabricate user attributes beyond available evidence?. So personalization can mean a bot quietly building a profile of you that is partly made up. Users' own worries focus on the moment data is collected rather than on how it is later used or kept What privacy concerns matter most to ChatGPT users?. That focus makes sense, because chatbots are very good at drawing out disclosure. People share more when no human is there to judge them Do chatbots help people disclose more intimate secrets?, and they match a bot's emotional openness the way they would with another person Do chatbots trigger human reciprocity norms around self-disclosure?.

The more useful finding is that the trade-off can be reduced. In a 450-person experiment, an agent that handed decisions back to the user whenever it detected a privacy risk caused less privacy concern and less loss of trust than a fully personalized agent. This worked even though it asked for approval less often overall, because what mattered was users feeling in control, not the number of times they were asked Can handing control back to users reduce personalization's privacy costs?. Read together, these studies suggest the privacy costs of personalization are real and partly unseen, its trust benefits may be smaller than they appear, and giving users control when it matters, not constant permission prompts, may be the way to keep the benefits without the unease.