Do people judge an AI agent's personality separately from whether it actually gets the job done well?

This explores whether people judge 'who the agent seems to be' separately from 'how well the conversation goes.' The corpus says they partly do. But the two show up as overlapping strands of one impression, not as two clean boxes. The clearest evidence comes from the Partner Modelling Questionnaire. People's mental picture of a dialogue agent splits into three factors: perceived competence, human-likeness, and communicative flexibility How do users mentally model dialogue agent partners?. Human-likeness is roughly the 'identity' question, while competence and flexibility are about interaction quality. They are statistically distinct, so users do pull them apart. The surprise is the weighting: competence explains almost half of how people size up an agent, and human-likeness about a third. Getting the job done matters more than the persona.

The disclosure research adds a twist. When people learn their partner is an AI, they first steer away from it. That bias reverses after repeated interactions where they can see the results Does revealing AI identity help or hurt user trust?. So identity isn't a fixed perception. It gets revised by how good the interactions turn out to be. Disclosure without visible outcomes changes nothing. On first contact the two dimensions feel separate, and over time interaction quality absorbs identity.

On the practitioner side, the two problems tangle. Fin's team found that one warm opening message lifted customer satisfaction by several points Does conversation design ownership measurably improve AI agent performance?. Tone and warmth are partly persona choices, yet they paid off as interaction quality. Work on proactive agents makes a similar point from another angle. An agent can be smart and adaptive but still feel intrusive if it lacks 'civility': good timing, respect for boundaries, not overriding the user How can proactive agents avoid feeling intrusive to users?. Civility is the place where character and conversation quality become the same thing.

Separating the two matters most where the stakes are emotional. A heuristic review of nine deployed agents found that companion-style and advisory products like Character.ai scored worst on interaction principles such as preventing dependency and making intent visible. General-purpose productivity tools scored best Do productivity agents actually protect users better than emotional ones?. Products that put the most effort into a vivid identity may be neglecting the basics of interaction. Meanwhile, engineering work on keeping personas consistent treats identity as its own technical problem. Examples include layered persona scripts Can layered persona architecture sustain coherent character behavior? and RL training that cuts 'persona drift' Can training user simulators reduce persona drift in dialogue?. Those papers measure consistency, not whether users experience it as separate from quality.

A useful parallel comes from phone agents. Task success, privacy compliance, and reusing saved preferences turn out to be independent capabilities, and success rankings don't predict the other two Do phone agents succeed at all three critical tasks equally?. Agent quality may look like that in general: several dimensions that move separately even when users experience them as one overall impression. The corpus doesn't have a study that asks users directly whether they see identity and interaction as separate design problems. The closest answer is that perception keeps them apart, while design and trust over time keep merging them.