How far can personality push users toward misplaced AI trust?

Tanner Kohler's Nielsen Norman Group (NN/g) study guide proposes that users anthropomorphize AI along an escalating four-level scale: "courtesy, reinforcement, roleplay, and companionship." It grounds the scale in a basic relational drive — "People inherently seek interactions with people (or systems) that seem to listen to them and understand their needs. The better the AI is at seeming as though it understands a user, the more they trust it and project human-like attributes on it" — and in a specific trust split: "People trust AI more when it seems smart rather than sentient. AI emotions can reduce trust in factual, task-oriented work." The guide adds that "Users quickly attribute human-like characteristics to artificial systems, which reflect their personality back to them," and closes with a design warning: "Large-language models (LLMs) humanize by design, and adding personality or emotion amplifies the risk of misplaced user trust."

The mechanism the guide gives is twofold. First, anthropomorphism is partly a property of the systems themselves — LLMs "humanize by design" — and partly a property of users, who bring a standing preference for being understood into any interaction, human or not. Second, the guide separates two distinct triggers of trust that pull in opposite directions: perceived competence ("smart") builds trust in task-oriented work, while perceived sentience or emotional affect can undermine it in exactly those contexts. The four-level scale is presented as a path design choices can push users along — from minimal courtesy responses, through reinforcement and roleplay, to companionship-level attachment — with personality and emotion named as the specific levers that amplify risk as a user moves up that scale.

This connects to Does chatbot language style actually shape how much we trust it?: both describe trust forming from how an AI presents itself rather than from its underlying competence, though that note isolates expertise-sounding language while this one isolates personality, emotion, and perceived sentience as the risk vector. It extends Why do people trust AI outputs they shouldn't?: anthropomorphic projection is a plausible on-ramp to the intuition-reason conflation that framework names, since treating an AI as person-like licenses intuitive trust over reasoned evaluation. It also follows from Why do users fail with AI interfaces designed like conversations?, which argues users bring communicative-use competencies to a system that isn't communicating; the four-level scale here describes how far that misapplied competence can run, from simple courtesy to full companionship.

The excerpt does not define how the four levels are distinguished operationally, does not cite the study or method behind "People trust AI more when it seems smart rather than sentient," and does not quantify how much personality or emotional design actually amplifies misplaced trust. It is presented as summary, not as original measurement. What it supports, at the strength the excerpt allows, is a design caution rather than a proven effect: adding personality or emotional affect to an AI product is not a neutral engagement choice, since each step up the anthropomorphism scale the guide names raises the chance a user's trust outpaces the system's actual competence.