To path indicator Subpages of “Department“ To navigation To quick access To footer with other services
To content
Logo WIWI with text "Department of Business and Economics"
Students walk across the campus north. A large red sculpture in the shape of large gears can be seen. © Roland Baege​/​TU Dortmund

Publication of a Study on AI Explainability and Trust Repair in Conversational AI Systems

The image is an infographic illustrating four different trust repair strategies used by AI systems, divided into two categories: human-like and system-like approaches. In the left section, under the heading "Human-like trust repair strategies:", there are two illustrations of a small orange robot standing next to a smartphone. In the first scene, titled "Apology", the robot looks sad, places a hand on its chest in a remorseful gesture, and the chat interface on the phone screen reads: "I apologize for my mistake. You're right.". The second scene next to it is labeled "Asking questions". Here, the sad-looking robot is shrugging its shoulders, has a blue question mark hovering above its head, and the phone screen displays the message: "Could you please point out where I made an error?". In the right section, under the heading "System-like trust repair strategies:", the robot's demeanor changes, and it is depicted with a strictly neutral facial expression. The third scene, labeled "Local explanation", shows the robot pointing at the phone screen, which now displays dark panels with statistical data values such as "Jeans: +0.78", "Missing: +0.65", and "Cap: -0.13". The fourth and final scene is titled "Counterfactual options". Here, the neutral-looking robot points at the phone, which shows a user interface with three clickable buttons labeled "Option A", "Option B", and "Option C". The four distinct scenarios are visually separated from one another by vertical, dashed green lines. © Björn Konopka​/​TU Dortmund, AI-generated with Gemini Pro 3.0
Illustration of trust repair strategies
What strategies can AI-based conversational agents use to effectively rebuild user trust after making mistakes?

The article “Explainability in AI: Comparing Human-Like and System-Like Trust Repair Strategies” was published in the journal Information Systems Frontiers.

Authors: Björn Konopka and Manuel Wiesche

The study examines which strategies AI-based conversational agents should employ to restore user trust following a system error. The focus is on comparing human-like strategies (apologies, follow-up questions) in line with the Computers are Social Actors (CASA) paradigm with system-oriented eXplainable AI (XAI)-based strategies (local explanations, counterfactual options). Based on a controlled online experiment with 261 participants, in which conversational AI agents employed different repair strategies following a simulated system failure. The results show that both human-like and XAI-based strategies can restore subjective trust to a comparable degree. The key finding is that XAI-based explanations lead to significantly higher actual continued usage rates. This result challenges common design decisions that prioritize “human-like strategies as the default” and establishes XAI as an effective mechanism for restoring trust after the fact.

Read the full article: https://link.springer.com/article/10.1007/s10796-026-10751-1 

Back