Summary: In technology, lower latency is usually considered an unambiguous improvement. Yet a recent study suggests that making AI respond faster can sometimes reduce users’ perception of its thoughtfulness and usefulness.
Researchers tested 240 people with chatbot responses deliberately delayed by different amounts (2, 9, or 20 seconds). Even when the content was identical, participants rated slower answers as more thoughtful and more useful. The results indicate people naturally attribute human-like “thinking” to pauses in AI responses, which affects subjective judgement even if objective behavior stays the same.
Key Facts
- The perception gap: Participants who waited 9 or 20 seconds for identical replies rated the AI higher than those who received 2-second responses.
- Anthropomorphizing latency: Users read pauses as the system “thinking,” applying human conversational norms—where a quick reply can seem impulsive and a pause suggests deliberation—to AI interactions.
- Behavior versus perception: Response speed did not significantly alter how people used the chatbot—frequency of prompts and interactions stayed largely the same. Only subjective evaluations of intelligence and usefulness changed.
- Task matters more than speed: The kind of work shaped interaction patterns: creation tasks (brainstorming, drafting) generated more iterative back-and-forth, while advice tasks (evaluation, recommendations) led to fewer, focused exchanges.
Source: NYU
Background: As developers optimize AI for faster responses, latency is often framed as a purely technical constraint to minimize. But how these speed gains influence real users’ impressions and decision-making is less well understood.
Previous human-computer interaction research has linked faster response times to better usability for deterministic systems like web pages or file transfers, where outcomes are predictable. AI models differ: they are probabilistic and generate responses that can vary. Because AI interfaces often mimic conversation, people instinctively apply social cues—such as timing and pauses—from human dialogue to machines.
A study presented at CHI’26 examined how response timing shapes both use and evaluation of AI assistants. Led by Felicia Fang-Yi Tan and Professor Oded Nov, the research recruited 240 participants who completed typical knowledge-work tasks via a chatbot. Tasks were split between creative activities—like brainstorming ideas and drafting text—and advisory activities—such as evaluating decisions and offering recommendations.
Participants experienced one of three controlled response delays. Some saw answers after two seconds, others after nine seconds, and a final group after twenty seconds. Content was kept consistent across conditions to isolate timing as the variable.
The findings run counter to the common presumption that faster is always better. While speed did not change interaction patterns—users prompted and edited at similar rates across conditions—it did affect how intelligent and useful the chatbot appeared.
Participants who received the fastest replies consistently rated the AI as less thoughtful and less helpful. Longer delays led users to attribute more deliberation and care to identical replies. Many interpreted the pause as a sign that the system was “thinking,” mirroring the role of pauses in human conversation where measured silence often signals reflection.
The distinction between observable behavior and subjective perception is important. The study shows that timing primarily changes appraisal rather than usage. For creative tasks, users engaged in more iterative exchanges regardless of delay; for advice tasks, exchanges were fewer and more targeted. But subjective trust and perceived intelligence rose with longer pauses.
These results carry practical and ethical implications for AI design. Instead of treating every millisecond of latency as wasted time, designers might consider “positive friction”—intentional, modest delays that encourage reflection and give outputs a perception of care. Such design choices could enhance user experience in contexts where perceived deliberation matters.
At the same time, designers should be cautious. If slower responses automatically evoke greater trust, users might over-rely on systems whose outputs are not actually superior. That raises ethical questions about whether and how systems should manipulate timing to influence perception—and whether users should be informed about such intentional delays.
Key Questions Answered:
A: The study introduces the idea of “positive friction.” Intentional slowdowns can encourage reflection and increase perceived trust, but they can also be used deceptively—making an inferior system seem smarter simply by adding a pause. Decisions to slow responses should weigh benefits against ethical concerns and transparency.
A: Because AI chat interfaces resemble human conversation and operate probabilistically, people apply the same social models they use with other humans. In human dialogue, a very quick reply to a complex question can feel dismissive, while a pause suggests genuine consideration.
A: Not necessarily. The effect is psychological: slower responses can feel more thoughtful but don’t change the underlying output. For repetitive, speed-sensitive tasks, a fast model is still preferable. For collaborative, advisory work where perceived deliberation matters, users may prefer the cadence of a slower model.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional context added by staff.
About this AI research news
Author: Leah Schmerl
Source: NYU
Contact: Leah Schmerl – NYU
Image: The image is credited to Neuroscience News
Original Research: Findings were presented at CHI’26