Thinking / Feature
Talk to Me Like a Human
The conversational interface trap
Natural language makes AI systems easier to use.
It also invites expectations they cannot meet.
Part of
Metaphor Infrastructure
Forthcoming
A Common Design Instruction
“Just talk to it like you’d talkto a person.”
It also invites the user to expect it to act like a person, even if they know it isn’t.1
01
Why It Works
The interface succeeds by borrowing from human communication.
Natural language is among the most practiced forms of human interaction. It lets users express intent, revise a request, supply context, and recover from misunderstanding without learning a command language or reading a manual.2
The more sensitively an AI system responds to context and purpose, the more useful it becomes as a tool for thought. That is one of the central achievements of conversational interface design.
Conversation is the most practiced interface a person can bring to a human-machine interaction
Why conversation is such a powerful interface
01
No manual required
People can begin with the language and social practices they already know.
02
Expressive task specification
Users can describe goals, examples, constraints, and uncertainty in their own terms.
03
Rapid repair and refinement
A user can correct, redirect, or elaborate without restarting the interaction.
Familiar language
Lower friction
Wider access
More powerful use
02
The Person Metaphor
The interface is wearing a person.
A conversational AI uses language within an interface that assembles a familiar social situation from many small cues: a first-person voice, alternating turns, typing indicators, memory claims, apologies, and labels such as “thinking” or “reasoning.”2
01
Linguistic Costume
The system speaks as “I.”
It apologizes, hedges, reassures, and phrases outputs as beliefs or intentions.3
02
Visual Costume
The interface stages a dialogue.
Turns, avatars, bubbles, and typing animations borrow the grammar of human conversation.4
03
The Chat Interface Trap
Both are true.
The same design strategy that makes AI unusually accessible also encourages users to import assumptions the system may not be able to satisfy.1–4
Like A Person
Conversation is irreplaceable.
Natural language is a general-purpose interface that users already know. It covers tasks no designer could list in advance, accommodates partial expressions of intent, and lets users repair misunderstandings within the conversation.2
- Familiar language
- Partial intent is okay
- Easy repair and refinement
But the more human the interaction feels, the easier it is to assume there is a person behind it.
Like A Tool
Clear boundaries matter.
A more system-like interface can make it clearer that the response comes from a computational system, not a stable person with intentions, commitments, or feelings. But pushing too far in this direction can make interaction rigid, unfamiliar, and harder to use.
- Clearer system boundaries
- Fewer human assumptions
- Clearer expectations
But the more mechanical the interface becomes, the more we lose what makes conversation useful.
How can designers keep the power of conversation without making the system seem more human than it is?
Neither extreme works. The challenge is to keep what makes conversation useful without making the system seem like a person.
04
Designing Around the Trap
Natural, without implying a person.
Designers can preserve capability and conversational fluency while making the system’s nonhuman status clear.2
Principle 1
Preserve conversational flexibility
Support ordinary language without implying a stable identity behind every response.
Principle 2
Make boundaries legible
Where appropriate, show significant changes in the structure of the interaction, e.g., when session context changes through compaction or when turns in a session are routed to distinct model types.
Principle 3
Name mechanics carefully
Intermediate tokens are not ‘thinking tokens’; a recommendation does not need to be framed as ‘What I’d do’.
Principle 4
Surface multiplicity
Where appropriate, show that one polished answer is one generated possibility rather than a settled conviction3, and show the contributions (and even revisions) of distinct sub-agents.
A Small Interface Experiment
Three ways to present the same result
Not a solved pattern. A question made visible.
Highly Personified
“I found three places I think you’ll love.”
Warm and familiar, but it implies preference, judgment, and a stable speaker.
Mechanical
QUERY COMPLETE: 3 RESULTS
Conceptually cautious, but it throws away the expressive benefits of conversation.
System-Legible
Three restaurants identified based on your criteria.
Natural enough to read easily, while naming an operation without the first person pronoun.
05
Open Questions
A research program.
Questions that can be investigated through design, prototyping, and research with people using deployed systems.01
Which interface cues most strongly trigger assumptions of personhood?
02
Can conversational fluency coexist with more accurate mental models?
03
When does anthropomorphism help, and when does it become deceptive?
04
Which forms of transparency improve calibrated trust?
05
Can different interaction modes support different relationships to the same system?
Further Reading
03 Sources
Further reading.
Three sources that extend the article’s account of conversation, language models, and social cues.Source 1
ConversationA concise account of the practices and cognitive capacities that make ordinary conversation possible.
Read Source ↗Source 2
Talking about Large Language ModelsAn argument for describing language-model behavior without treating fluency as evidence of thought, understanding, or inner life.
Read Source ↗Source 3
A Taxonomy of Social Cues for Conversational AgentsA systematic account of the verbal, visual, auditory, and less visible cues through which conversational interfaces invite social responses.
Read Source ↗References
05 Citation Groups
References.
Numbered groups correspond to the superscript citations used throughout the article and research note.
Weizenbaum, J. (1966). ELIZA—A computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36–45. https://doi.org/10.1145/365153.365168
Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. W. H. Freeman.
Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
Ciston, S., Berry, D. M., Hay, A. C., Marino, M. C., Millican, P., Schwarz, A. I., Shrager, J., & Weil, P. (2026). Inventing ELIZA: How the first chatbot shaped the future of AI. The MIT Press. https://doi.org/10.7551/mitpress/15790.001.0001
Kendrick, K. H., & Holler, J. (2024). Conversation. In Open encyclopedia of cognitive science. MIT Press. https://doi.org/10.21428/e2759450.3c00b537
Clark, H. H. (2025). Common ground. In Open encyclopedia of cognitive science. MIT Press. https://doi.org/10.21428/e2759450.671362c9
Levinson, S. C. (2025). The interaction engine. In Open encyclopedia of cognitive science. MIT Press. https://doi.org/10.21428/e2759450.e3df24b2
Hutchins, E. L. (1989). Metaphors for interface design. In M. M. Taylor, F. Néel, & D. G. Bouwhuis (Eds.), The structure of multimodal dialogue (pp. 11–28). North-Holland.
So, J., Cheng, C., & Krishna Murthy, S. (2026). Beyond anthropomorphism: A spectrum of interface metaphors for LLMs. In Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems (pp. 1–7). Association for Computing Machinery. https://doi.org/10.1145/3772363.3798591
Shanahan, M. (2024). Talking about large language models. Communications of the ACM, 67(2), 68–79. https://doi.org/10.1145/3624724
Shanahan, M., McDonell, K., & Reynolds, L. (2023). Role play with large language models. Nature, 623(7987), 493–498. https://doi.org/10.1038/s41586-023-06647-8
Cohn, M., Pushkarna, M., Olanubi, G. O., Moran, J. M., Padgett, D., Mengesha, Z., & Heldreth, C. (2024). Believing anthropomorphism: Examining the role of anthropomorphic cues on trust in large language models. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (Article 54, pp. 1–15). Association for Computing Machinery. https://doi.org/10.1145/3613905.3650818
Feine, J., Gnewuch, U., Morana, S., & Maedche, A. (2019). A taxonomy of social cues for conversational agents. International Journal of Human-Computer Studies, 132, 138–161. https://doi.org/10.1016/j.ijhcs.2019.07.009
Gnewuch, U., Morana, S., Adam, M. T. P., & Maedche, A. (2018). Faster is not always better: Understanding the effect of dynamic response delays in human-chatbot interaction. In Proceedings of the 26th European Conference on Information Systems (Research Paper 113). Association for Information Systems. https://aisel.aisnet.org/ecis2018_rp/113/
Go, E., & Sundar, S. S. (2019). Humanizing chatbots: The effects of visual, identity and conversational cues on humanness perceptions. Computers in Human Behavior, 97, 304–316. https://doi.org/10.1016/j.chb.2019.01.020
Cox, S. R., Martin-Lise, J., Hosio, S., & van Berkel, N. (2026). Watching AI think: User perceptions of visible thinking in chatbots [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2601.16720