Fact Checking Chatbot: A Misinformation Intervention for Instant Messaging Apps and an Analysis of Trust in the Fact Checkers

TL;DR

This paper investigates misinformation on Singapore’s mobile instant messaging services and assesses the effectiveness of a chatbot-based fact-checking intervention. Using a within-subjects design with 527 Singaporean participants, it tests three fact-checking sources—Government, News Outlets, and Artificial Intelligence—against a control condition, across 16 headlines with true/false veracity and truth-labels, while also measuring trust in each source. Results show mixed effects for accuracy, but AI-based fact checking yields the highest adherence to labels, revealing a notable attitude–behavior gap where high trust in government does not translate into greater adherence to government-issued labels. The findings highlight the potential and risks of AI-based fact checking on MIMS, suggesting that high-performing automated checkers can improve veracity judgments if they maintain integrity, while also underscoring the need to address blind trust and source biases in design and deployment. The work contributes to understanding how to deploy rapid, private fact-checking interventions on encrypted messaging platforms and informs policy and system design for scalable misinformation countermeasures.

Abstract

In Singapore, there has been a rise in misinformation on mobile instant messaging services (MIMS). MIMS support both small peer-to-peer networks and large groups. Misinformation in the former may spread due to recipients' trust in the sender while in the latter, misinformation can directly reach a wide audience. The encryption of MIMS makes it difficult to address misinformation directly. As such, chatbots have become an alternative solution where users can disclose their chat content directly to fact checking services. To understand how effective fact checking chatbots are as an intervention and how trust in three different fact checkers (i.e., Government, News Outlets, and Artificial Intelligence) may affect this trust, we conducted a within-subjects experiment with 527 Singapore residents. We found mixed results for the fact checkers but support for the chatbot intervention overall. We also found a striking contradiction between participants' trust in the fact checkers and their behaviour towards them. Specifically, those who reported a high level of trust in the government performed worse and tended to follow the fact checking tool less when it was endorsed by the government.

Figures (6)

• Figure 1: The introductory message describing the purpose and usage of the chatbot.
• Figure 2: The mock WhatsApp chatbot interface showing conversations of four news items in different Fact Checker × News Veracity × Fact Check Label conditions.
• Figure 3: Average AccuracyPV across different levels of (a) News Veracity, (b) Fact Check Label and (c) Label Precision. AccuracyPV scores are between 4 (high accuracy) and 1 (low accuracy). Error bars show 0.95 confidence intervals.
• Figure 4: Average AccuracyPV across different levels of (a) Fact Checker and News Veracity, (b) Fact Checker and Fact Check Label and (c) Fact Checker and Label Precision. AccuracyPV scores are between 4 (high accuracy) and 1 (low accuracy). Error bars show 0.95 confidence intervals.
• Figure 5: Average AdherenceFCL across different levels of (a) Fact Checker, (b) News Veracity and (c) Label Precision. AdherenceFCL scores are between 1 (high adherence) and 0 (low adherence). Error bars show 0.95 confidence intervals.