Don't Detect, Just Correct: Can LLMs Defuse Deceptive Patterns Directly?

Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA), pp. 1–11

Abstract

Figure 1: We provided the LLM GPT-4o with the HTML source of UIs like the one on the left, a fictional ticket shop using several deceptive patterns to manipulate users into purchasing VIP tickets.Several iterations of simply prompting the LLM to "Make that less manipulative" (our minimal prompt) resulted in the fair design on the right.

Authors 5

1. René Schäfer Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Aachen, Germany

2. Paul Miles Preuschoff Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Aachen, Germany

3. Rene Niewianda
   Affiliation as printed

   Independent Researcher, Aachen, Germany

4. Kevin Fiedler Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Aachen, Germany

5. Jan Borchers Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Aachen, Germany