Column
From Fixed to Fluid: Why Conversational AI Should Switch Personalities with Context
Agent personality works best at moderate intensity — and the optimal personality shifts with context. A look at the fluid personality framework, which co-adapts persona metaphors (coach, tutor, librarian, tool) and expression intensity, and how it maps onto contact-center CX.
Hello. I am Miura from Affectosphere Group.
How do you decide your chatbot’s character?
Bright and friendly. Calm and polite. In most organizations, someone picks one personality and locks it in.
A paper published on arXiv in July 2026 (Rahman & Desai, arXiv:2607.01034) argues that this very act of fixing a personality can be a design flaw. Think of the last time an urgent support request was met by a relentlessly cheerful bot. That irritation is the phenomenon in question.
Three takeaways
- Synthesizing prior work, the paper shows that moderate personality expression — the middle of low, medium, high — maximizes user trust, enjoyment, and adoption intent. Too much or too little backfires.
- But the optimal personality depends on context. A statically fixed personality misaligns as task context, urgency, and user characteristics shift.
- The proposed fluid personality framework co-adapts a metaphorical persona (coach, tutor, librarian, tool) and personality expression intensity as a function of task context, user goals, and situational urgency. It is a design philosophy that maps directly onto contact-center CX differentiation.
① The surprising finding: moderate personality wins
The foundational insight is counterintuitive.
When conversational agents are compared at low, medium, and high personality expression, it is the moderate level that maximizes trust, enjoyment, and intention to adopt.
An agent with too little personality feels mechanical and cold. One with too much comes across as annoying and untrustworthy. The relationship is an inverted U, with the sweet spot in the middle.
This cuts against the marketing intuition that a stronger character always drives more engagement.
② Fixed personalities break when context shifts
The real subject of the paper, though, lies one step further.
Even if moderate personality is best, fixing it creates a new failure mode: the agent cannot follow contextual change. The paper calls this misalignment.
Consider a user browsing health information. During casual exploration, a friendly coach-like manner feels pleasant. The moment the user starts worrying that a symptom might be serious, that same manner feels tone-deaf. What is needed now is the precise, economical style of a librarian.
Fitness coaching. Reflective learning support. Every use case the paper highlights is a domain where contextual properties can flip mid-session. Fix a single personality, and a mismatch is guaranteed somewhere.
③ The fluid personality framework: two dials turned by context
The proposed framework is structurally simple. It has two dials.
The first is the metaphorical persona. The agent’s role is typed through metaphors: a coach (encouraging companion), a tutor (guiding teacher), a librarian (precise navigator), a tool (silent executor).
The second is personality expression intensity. Within any persona, how much emotional expression and personal address to layer on — low, medium, or high.
The framework then co-adapts both dials as a function of task context, user goals, and situational urgency. Urgency spikes? Shift toward tool, low intensity. Accompanying a user toward a goal? Shift toward coach, medium intensity. Personality is treated not as a setting but as a state.
A practical proposal: the context-sensing CX engine
The most obvious use case is contact-center and customer support AI.
Complaint handling, routine inquiries, and emergencies demand entirely different interaction personalities. Most current chatbots are monolithic here, so mismatches — “too casual for a complaint,” “too heavy for a simple question” — happen daily.
The implementation skeleton looks like this. Extend your existing intent classification with urgency and emotional-state estimation. Feed the output to a personality-switching module that dynamically sets persona and expression intensity in the system prompt. With an LLM-based chatbot, you can start with prompt control alone — no model changes.
The owning teams are CX improvement and support operations. KPIs: segment-level customer satisfaction (CSAT) and escalation rate. The first validation point is whether CSAT improves specifically in the urgent, high-emotion segment.
The same skeleton transfers to healthcare apps and HR tech such as mental health support desks. If anything, it should work better in domains where emotional swings are larger.
Is a shape-shifting AI dishonest?
Persona switching may trigger the reaction: “isn’t this manipulating users?”
But excellent human support staff do exactly this, naturally. They read the other person’s state and adjust how they speak. That is not manipulation; it is consideration.
The real question is not whether to switch, but what the switching serves. Adapting personality to help users reach their goals is one thing; cranking up emotion to farm session time is another. The framework is identical — the objective function decides its ethics.
For anyone working in CX, this paper offers a genuinely well-structured way to think about designing context-appropriate behavior while preserving brand consistency.
Cheers!
References
- Rahman, Hasibur, & Desai, Smit (2026). Behavior-Adaptive Conversational Agents: Toward a Fluid Personality Framework. arXiv preprint.
* This article was written in part with AI assistance and may contain inaccuracies.