When artificial intelligence apologizes: Dynamic trust updating and the role of personality.

What is it about?

This study examines how different AI apologies affect users’ trust after the AI makes a mistake. In an experiment with 245 participants, a chatbot made errors and then either did not apologise or used one of six apology types. We tracked how trust changed across repeated interactions and also examined whether personality influenced these changes. We found that only a simple expression of regret (“I’m sorry”) reduced the immediate drop in trust, while no apology strategy improved longer-term recovery once the AI began performing correctly again. People who viewed the AI in more relational terms were also more responsive to apologies. Personality also mattered. Agreeableness and Openness were associated with higher baseline trust, while people higher in Openness adjusted their trust more strongly in response to both AI errors and subsequent correct performance. However, personality did not reliably predict which type of apology worked best.

Featured Image

Photo by Steve DiMatteo on Unsplash

Photo by Steve DiMatteo on Unsplash

Why is it important?

The study shows that human trust-repair principles cannot simply be copied wholesale into AI design. AI apologies appear to work selectively. A short expression of regret can soften the immediate damage from an error, but actual subsequent performance is more important for rebuilding trust over time. It is also important methodologically because it treats trust as something that changes continuously, rather than relying on a single “How much do you trust the AI?” rating at the end. This gives a more realistic picture of how people calibrate trust when interacting repeatedly with AI systems.