aceModerate evidence
More
Study of 245 people finds only a chatbot's expression of regret softened trust loss
Study title: When artificial intelligence apologizes: Dynamic trust updating and the role of personality.
- Evidence type
- Randomized controlled trial
- Sample
- n = 245
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-2518cite or check it
Study title: When artificial intelligence apologizes: Dynamic trust updating and the role of personality.
Researchers randomly assigned 245 participants to work through five business problems with a chatbot that was preprogrammed to make mistakes on two tasks. After the errors, participants saw no apology or one of six apology strategies, and the team tracked how trust changed after each task.
Read the study abstract
Artificial intelligence (AI) systems are susceptible to error, and many are designed to apologize.Yet, it remains unclear which apologies restore user trust, a question requiring repeated interactions with an AI agent and models capturing how trust evolves across encounters.Drawing on models of apology from human-human interaction research, we examined whether apology components that repair interpersonal trust function similarly in AI.Participants (N = 245) solved five business problems with a chatbot preprogram…
The gain
In a randomized experiment with 245 participants, a chatbot expressing regret after an error was the only one of six apology types that softened the immediate drop in trust compared with no apology.
In brief
Trust trajectories differed significantly by apology condition. Compared with no apology, only Expression of Regret buffered the usual immediate fall in trust after an error. The other five strategies did not do so, and none of the apologies boosted trust further once the chatbot went back to answering correctly.
The effects varied by person. Apologies reduced trust loss more among participants whose post-interaction evaluations were more relational than functional. Personality also mattered: Agreeableness and Openness were linked to baseline trust, and Openness was also tied to how trust shifted over time.
Main points
- The experiment randomized 245 participants into seven conditions: error with no apology, or error followed by one of six apology strategies.
- The six strategies were Expression of Regret, Explanation, Acknowledgment of Responsibility, Declaration of Repentance, Offer of Repair and Request for Forgiveness.
- Only Expression of Regret cushioned the immediate trust decline following the error-apology-correction sequence relative to no apology.
- No apology type further improved trust recovery once the chatbot resumed correct performance.
- Participants with more relational than functional evaluation profiles showed larger apology-related reductions in trust loss; Openness was linked to dynamic trust updating.
The full story
Participants solved five business problems with a chatbot that erred in two tasks at fixed intervals. A dynamic trust-updating model estimated how trust shifted after each trial, separating errors from correct performance across repeated encounters.
Results showed that apology effects were selective. Expression of Regret alone softened the immediate loss of trust, while explanation, taking responsibility, repentance, an offer of repair and a request for forgiveness did not outperform silence on that measure.
In separate analyses, all five Big Five traits were associated with baseline trust. When the traits were considered together, only Agreeableness and Openness stayed associated with baseline trust, and Openness was also linked to dynamic trust updating.
Sources
1 on file- Technology, Mind, and Behaviordoi:10.1037/tmb0000230
Talk it over
Who gets this benefit, and who was left out of the study or story?
Study of 245 people finds only a chatbot's expression of regret softened trust losshttps://truvace.com/item/when-artificial-intelligence-apologizes-dynamic-trust-updating-and-the-role-of-p
Who gets this benefit, and who was left out of the study or story?
The gain In a randomized experiment with 245 participants, a chatbot expressing regret after an error was the only one of six apology types that softened the immediate drop in trust compared with no apology.
Answer in the threadHow do we know? What can this kind of evidence show, and what can’t it?
The evidence This rests on a randomized controlled trial with a sample of 245, from Technology, Mind, and Behavior.
Answer in the threadWhat does it leave open? What else would you want to know before acting on it?
The record’s own caveat None of the apology conditions improved trust recovery once the chatbot resumed correct performance, and the effects were selective and context-dependent, depending on participants' evaluation profiles and personality.
Answer in the threadHave you seen AI used this way where you study, work or live? Did it go the same way?
Answer in the thread
The questions are the same on every record. The details under them are taken from this record as written.
What readers say