Should we be polite to ChatGPT?

Politeness, as defined by Brown and Levinson (1987), functions as a system of strategies for managing and protecting face, the social regard reflecting a person's identity. In this sense, face embodies the socially exposed part of that identity; the part susceptible to elevation, threat, or repair in interaction. As such, the use of politeness relies on the assumption that each participant possesses an identity that their treatment can validate or undermine.

So, when evaluating whether a conversational participant, such as ChatGPT, deserves politeness, we must first establish what an identity worthy of politeness constitutes and whether ChatGPT possesses these conditions.

Examining the accounts of identity formation in humans, we find identity arises from how we build a self through lived, physical, and social experience, a process that large language models do not undergo. So, because ChatGPT does not form an identity grounded in these conditions for courteous interaction, I contend we should not be polite to ChatGPT.

Foundations of Human Identity and the Role of Politeness

Politeness protects an identity built on two things: the experiences a person physically lives through, and the shaping of these experiences based on the feedback of others. Beginning with how we accumulate experiences, the process starts with gathering inputs, such as the sight of people, the sound of speech, the body's position, and concurrent emotions such as fear or pride. Then, the brain's hippocampus binds these factors into coherent events, linking people, actions, and locations (Eichenbaum, 2017). These events constitute episodic memories, which are recollections of particular experiences situated in time and place (Tulving, 1985). We revisit them through autonoetic consciousness, re-living the emotions of those events in the same first-person perspective in which they occurred. This concept constructs a ‘self’, proving crucial because it allows individuals to mentally time-travel, constructing an inherently subjective personal identity (Suddendorf & Corballis, 2007). Politeness protects this personal identity because the use of courtesy helps maintain relationships, and healthy relationships ensure survival. Humans evolved into cooperative groups, where collaboration dictates how we share resources and necessities like shelter and food. Without this collaboration, we are susceptible to physical and mental decline (Baumeister & Leary, 1995). Consequently, achieving our fundamental goals depends on how others perceive us, and we use politeness to maintain these relationships. To internalize politeness, we continuously monitor our reactions and gauge how others react in approval or disapproval, which results in innate emotions of pride or shame (Cooley, 1902; Leary, 2005), which we internalize into our identity, thereby teaching us to be polite to maintain our relationships through softening requests, offering thanks, or apologies, which all validate the other person's social value and autonomy, actively reassuring their brain's monitoring systems that their standing in the group remains secure. This maintenance of group inclusion safeguards a goal-driven “self” which we have built across years of remembered experience. Politeness, then, belongs to those whose identity forms and holds this way. The question remains whether ChatGPT fits this definition.

Why ChatGPT Appears to Qualify

ChatGPT appears to qualify because it reproduces the signals of politeness with a fluency that mimics genuine social participation. Chatbots can draw on a toolbox of politeness and fluent natural-language interaction, so we readily interpret their contingent responses as human-like, even though we know ChatGPT is not a living creature.

For example, in a classic study by Nass and Moon (2000), participants completed a task with the help of a computer and then evaluated how well it had performed. Some evaluated the system on the same machine they had just used; others did so on a different computer or on paper. Even though the computer's performance was identical, evaluations entered on the same machine were significantly more positive and restrained. When people “spoke” to the computer itself, they softened criticism while increasing praise, because they would do the same to a human interlocutor. Now, roughly two decades later, ChatGPT acts as a vastly more sophisticated version of that interface, partly because it effectively imitates the social cues through which we recognize and practice politeness with one another. Chatbots excel at politeness because the way they are trained on what is polite mirrors human socialization, where individuals act, receive a response, imagine how they appeared, and remember what to do on the next interaction based on whether they felt pride or shame. With ChatGPT, contractors interacted with the chatbot and outlined which responses felt appropriate and which did not. They did so by creating a reward model that assigned a scalar score to these responses, with higher scores indicating a greater margin of human approval. Based on these scores, the developers modified ChatGPT's parameters for textual output to reinforce what the moderators deemed socially appropriate while omitting the rest, until the chatbot's behavior reflected what humans preferred across thousands of social situations (Ouyang et al., 2022). With its polished politeness and anthropomorphic traits, ChatGPT triggers instinctive human reflexes, making the urge to reciprocate politeness natural to the point where it invokes a deep discussion on whether to do so.

Why ChatGPT Fails to Qualify as an Entity with a Self that Warrants Politeness

At its root, politeness serves the self, and a chatbot has no self to serve. We already defined identity and how humans build one through a physical and social reality, but ChatGPT organizes information through a drastically different process, so whatever it possesses does not constitute an identity worthy of politeness. To highlight this contrast, consider the nervousness of finishing an essay close to the submission deadline. That anxiety stems from years of past experiences, with goals, ambitions, standards, and fears about the judgment of past work, and what outcomes follow. The deliberation of success and failure matters because the situation of writing has emotional parallels tied to a narrative identity, where a continuing self that remembers these past experiences, shaped by the reactions of others. Architecturally, ChatGPT cannot undergo this existential process. During pre-training, the AI model's network processes a prodigious collection of human text divided into tokens, which are datapoints. The model then repeatedly predicts the next tokens, marks when it gets the predictions wrong, and then corrects itself until, after trillions of tokens are traversed, the model builds a sophisticated topology of how humans comprehend and articulate their experiences (Vaswani et al., 2017). As a result, ChatGPT can describe the nervousness of writing this essay close to the deadline with remarkable accuracy, because similar accounts of anxiety are apparent in its tokens. Yet, even when the model creates sophisticated descriptions of nervousness, the model never retrieved memories through autonoetic consciousness or felt the emotional deliberation of writing this essay, because it lacks the continuous self through which past experiences acquire significance and shape present concerns. One of the godfathers of AI, Yann LeCun, recognizes this limitation himself. He observes that “language is a very approximate representation of our percepts and our mental models” (LeCun, 2024). For that reason, ChatGPT's reality strictly interprets linguistic tokens, nowhere close to how humans perceive the world. Lacking an identity grounded in autonoetic and social experience, ChatGPT has no face, which is the part of identity that politeness protects. Therefore, if it doesn't have ‘face’, then it doesn't warrant politeness.

Objections and Rebuttals

However, even without a moral obligation to extend chatbots' politeness, some argue that impoliteness toward these systems can carry over into our own interactions or affect the helpfulness of their outputs. Evidence indeed grounds the first concern. Researchers found that when participants interact with AI, they shift towards candid, command-like language. This candid language, which users learn later, seems to affect the judgment of other people's work, subsequently carrying over habits of rudeness from human-AI interaction to human-human interaction (Tey et al., 2024; Gu et al., 2026). But, even when the transfer of rude habits exists, using this reason to justify conceding politeness dismisses the practicality of the chatbot. ChatGPT fundamentally serves as a tool for practical gain, so users do not have to waste time editing messages or worrying about courtesies like ‘please’ and ‘thank you’. Our blatant utility stems directly from our awareness that it lacks a face, as even OpenAI's CEO insists the technology ‘is a tool, not a creature’ (Altman, 2023). So, arguing ‘yes’ for politeness already dismisses the purpose of the chatbot, which exists to maximize practical gain. A less parasocial approach to address the issue of the habitual spillover is not to fake politeness to ChatGPT, but to create a distinct line between AI and humans, which is precisely what the researchers contend (Gu et al., 2026). Designers of these large language models should keep a boundary between machine and human, so we keep our social heuristics, rather than directing them towards a machine. Instead of extending warmth towards machines, the creators of the machine should reduce the sycophancy of AI, so we can continue to use it efficiently as a tool. Additionally, the concern that impolite interactions with AI models may degrade output quality lacks merit. Existing research finds that models like ChatGPT's responses depend far more on clear, specific instructions than polite phrasing, and that any effect of tone is inconsistent and task-dependent (Yin et al., 2024; Dobariya & Kumar, 2025). Because no strong link connects prompt etiquette to output quality, applying our uniquely human politeness heuristics to a weak claim undermines what LLMs should provide us.

Conclusion

Politeness mediates interactions for those who have accumulated an identity, and a self within that, a calibrated history of experience, feeling, and participation in the physical and social world. ChatGPT has learned from this world in a sophisticated manner, producing technology that, on some levels, surpasses humans in reasoning, summarizing, and translating. Yet, never experiencing the world as a living being remains a bottleneck for ChatGPT not having an identity, because ChatGPT has no social standing to protect. There is no “face” for politeness to safeguard, and therefore, we do not owe it consideration, regardless of whether it reflects on our morals.