Liveliness is something that feels so elusive in designing NPCs for almost any type of game. We want our worlds to feel responsive, but responsiveness feels different across genres and settings. I am currently developing Restle and Floppy/Write. Restle is turn-based at the end of the day, and that makes it hard to find the sweet spot for our NPC reactions. The opponent AI in Restle has three modes of responding: one is in its actions within the game, the decisions it makes based on what the player does during their turn. This can impact how much the player feels the NPC is paying attention to them. This becomes extremely evident when dealing with offering different difficulty levels for the player. On easier difficulties, I am working to make sure that the AI is actively choosing a less optimal play rather than just choosing the “wrong” move. I worry about the AI making choices that feel either detached from the player’s moves or making them feel as if the AI is actively throwing the match, which also feels bad and somewhat dismissive. This means that we need the AI to be capable of making a range of quality moves after any given play the player might make. Another way I am trying to create liveliness in the NPC is by having it engage in various forms of narration, some of which are responsive to the player in nature, while others are geared towards telegraphing the NPC’s decisions and encouraging the idea that the NPC is actually thinking about their response. And finally, each difficulty has its own personality, trying to engage with the player in different meters and through different literary voices so that it feels like a unique opponent with individuality every match.
The relationship to the player is one of the most crucial tools we have in making the AI feel intelligent. Liveliness in NPCs comes down not just to what the NPC does but to how the NPC feels like it is doing it. One thing I think is crucial is responsiveness that isn’t just mechanical but tuned to player expectations. For example, the decisions the AI makes after a player’s move must cover a spectrum: nearly optimal, somewhat suboptimal, maybe intentionally flawed but plausible. If the easier modes always feel like the AI is randomly bad or like it is throwing the match, the player will lose trust and feel disconnected. Research shows that players’ enjoyment increases when NPCs have behavior plus dialogue that reflect decision-making. In stealth games, simple dialogue with more complex behavior yields more enjoyment, and even simple behavior works better when its intentions are telegraphed via dialogue (Al Enezi & Verbrugge, 2023).
Another pillar is personality and identity. If each difficulty level or each NPC voice has a distinct character and the NPC’s narration reveals what it is thinking or considering, that reinforces the sense the NPC is alive in that moment. Research on NPC dialogue generation in RPGs finds that dynamic memory, character-consistent dialogue, and experience-response histories (so the NPC “remembers” player behavior) markedly improve immersion and the sense of an NPC as its own agent rather than as a predictable script (Weir et al., 2022). Also, identity matters: in one study, players behaved differently toward NPC companions depending on whether they thought the NPC was human or nonhuman, even though the AI underneath was identical. That kind of perception shift shows how big a role character identity plays (Williams et al., 2016).
When I think about the economy of attention, the way players are trained now by contemporary trends in gaming, it becomes clear that liveliness in NPCs has to compete with a constant demand for novelty. Players are conditioned by fast cycles of reward, by live service loops, by games that flood them with stimuli to keep them invested. That means our designs need to provide a different kind of anchor. A responsive NPC, one that telegraphs thought or makes mistakes that still feel deliberate, can hold attention not by overwhelming but by making the player believe there is someone else in the room with them. It shifts the focus from raw stimulation to sustained engagement. In that sense, giving NPCs personality, memory, and variability can be a counterweight to the attention economy because instead of just feeding the player another notification or quick reward, the game is cultivating a relationship that the player is curious to return to (Bogost, 2007).
When you read the attention economy as a designer, the first thing to notice is that players aren’t neutral recipients of whatever you put in front of them; they are trained to look for spikes, surprises, and repeatable payoffs. Modern live-service loops and many mobile and multiplayer designs teach players to scan the screen for instant signals that promise progress or novelty. That pattern of scanning rewires what counts as “interesting” to the player. That conditioning matters because it biases attention away from slow, social entanglements toward fast, predictable reward hooks. The mechanics that work there—variable rewards, intermittent reinforcement—are exactly the ones psychology flags as producing the highest persistence (King et al., 2010).
So, if you want an NPC that keeps someone’s attention without competing with every light and chime in the UI, you have to treat the NPC as a social anchor rather than a reward dispenser. People apply social rules to technology, so the moment an NPC gives social cues—a remembered slight, a voice that hesitates, a quip that references something you did last session—players treat it like another actor in the room. That isn’t just poetic; it’s CASA and parasociality in action. The brain’s social scripts get engaged, and those scripts are sticky in a way that raw points are not. Designing for parasocial continuity—consistent voice, believable memory, and signs that the NPC has internal reasons for its moves—reorients attention from “what do I get now” to “what will they do next,” and that curiosity sustains return play in a deeper way than a one-off loot drop (Zhu et al., 2021; Williams et al., 2016).
Practically, that means combining three levers instead of one. First, give the NPC a graded competence profile—not simply “good” or “bad,” but a range of plausible responses that change with context—so lower difficulties feel deliberate rather than random. Second, bake in episodic memory and decay so the NPC references prior encounters without becoming a logging dump. Curated forgetting keeps callbacks meaningful and avoids fatigue. Third, pace the informational rewards; telegraph intent sometimes, withhold some information other times, so the player experiences pattern and violation rather than flat predictability. These levers can be automated with dynamic difficulty and memory frameworks, and recent work shows both that dynamic difficulty adjustment can be used to maximize engagement and that memory-aware, generative NPCs materially increase the sense of presence when implemented carefully (Zhu et al., 2021).
Finally, watch the ethical and practical tradeoffs. Leaning too hard on intermittent, gambling-like rewards will win attention but erode goodwill and can cause harm, while leaning too hard into polished social presence without robust affordances can create the uncanny, brittle NPC that breaks immersion when it fails. The work you do to make an NPC feel like a conversational, remembering agent is the same work that converts attention into attachment rather than addiction—small, characterful moments, rhythm and pacing, and constraints on when and how the NPC interrupts the player. Measured correctly, NPCs become an attention economy counterweight. They don’t chase every eyeball with noise; they hold the player by being interesting as persons, not as machines that hand out points (Bogost, 2007; King et al., 2010).
References
- Al Enezi, W., & Verbrugge, C. (2023). Investigating the influence of behaviors and dialogs on player enjoyment in stealth games. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 19(1), 161–169. https://ojs.aaai.org/index.php/AIIDE/article/view/27512
- Bogost, I. (2007). Persuasive games: The expressive power of videogames. MIT Press. https://archive.org/details/persuasivegamese00bogo
- Weir, N., Thomas, R., D’Amore, R., Hill, K., Van Durme, B., & Jhamtani, H. (2022). Ontologically faithful generation of non-player character dialogues. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP 2022), 520–530. https://doi.org/10.18653/v1/2022.emnlp-main.520
- Zhu, J., Villareale, J., Javvaji, N., & Risi, S. (2021). MemoryRepository for AI NPC. CHI Conference on Human Factors in Computing Systems (CHI ’21). https://doi.org/10.1145/3411764.3445307
- King, D. L., Delfabbro, P. H., & Griffiths, M. D. (2010). The convergence of gambling and digital media: Implications for gambling in young people. Journal of Gambling Studies, 26(2), 175–187. https://doi.org/10.1007/s10899-009-9153-9
- Williams, A., Blackwell, A., & Griffiths, N. (2016). How the perceived identity of a NPC companion influences player behavior. International Journal of Human-Computer Studies, 94, 1–11. https://doi.org/10.1016/j.ijhcs.2016.06.002
Boukie Games 