Conference Presentation, Q&A
Adrian Blair, CEO & Co-Founder, TrustPilot: Winning Customers. Trust in the Age of AI
- Trustpilot anticipates that the evolution of agentic AI toward active decision-making, such as switching bank accounts or purchasing items, will necessitate embedding trust signals within these systems to prevent dangerous outcomes caused by a current crisis where 54% of Americans refuse to trust AI with decision-making.
- The company predicts that widespread AI adoption depends on systems learning from real human experiences rather than inferring conclusions from data, requiring the processing of hundreds of metadata points per review to determine trustworthiness.
- Trustpilot plans to launch the "Trust Layer" API this year as its primary AI product, designed to provide access to over 350 million historical human experiences and current feedback, utilizing raw review data to enable AI systems to listen before acting.
- Expected efficacy of the Trust Layer is projected to increase as daily feedback volume grows beyond 200,000 bits, with the API offering value beyond simple web scraping by containing non-public data that allows for the correction of cultural biases across different geographies.
- To ensure data integrity, Trustpilot expects to proactively remove untrustworthy content, following the precedent of removing 4.5 million reviews in 2024, and will deploy fraud investigation teams to action against bad actors like review sellers.
- Strategic plans prioritize maximizing review visibility across external platforms rather than developing an internal advertising business, with the expectation that large language models, due diligence investors, and consulting firms will be among the primary beneficiaries of the Trust Layer system.
- The organization maintains a stance that users will likely only adopt AI agents if they are trusted to operate safely based on real human data, necessitating the continued growth of positive word-of-mouth recommendations to improve the quality of trust embedded in future AI.