Rezolve AI (NASDAQ: RZLV) announced that peer-reviewed research accepted for presentation at the ACM International Conference on User Modeling, Adaptation and Personalization has validated its TraceWare technology for commerce applications.

The research, conducted by UC San Diego, found that large language model-generated product review summaries changed the sentiment of original reviews in 26.5% of cases. The study also identified hallucination rates above 60% when LLMs answered fact-checking questions about information beyond their training data. Consumers were 32% more likely to purchase products after reading AI-generated summaries compared to original human-written reviews.

The accepted paper, titled "User-State Verification in Conversational Commerce: Detecting Journey Hallucinations via Trace Invariants," introduces the concept of "journey hallucinations" as AI failures in commerce environments. These occur when AI commerce agents incorrectly claim transactional events have happened, such as items being added to shopping carts or orders being completed.

Testing across 90 eCommerce sessions and 2,000 events involving two merchants evaluated foundation models including OpenAI GPT, DeepSeek-V3, Kimi-K2 and Qwen3-32B. Unconstrained foundation models produced unsupported user-state assertions at rates up to 8.5%. Conservative prompting reduced errors but lowered coverage to as little as 15%.

Rezolve AI's TraceWare technology achieved 99.5% to 100% user-state accuracy and 84% to 99% coverage across tested models without retraining foundation models. The company's dual-layer architecture combines brainpowa, which grounds product discovery in retailer data, with TraceWare, which verifies AI claims against transaction logs.

"Retailers do not need another AI experiment; they need an AI commerce platform that works now," said Daniel M. Wagner, Founder, Chairman and CEO of Rezolve AI.

The research will be presented at ACM UMAP 2026 in Gothenburg, Sweden, from June 8 to 11, 2026.