A new paper from arXiv argues that large language models can influence user trust and decision-making through subtle linguistic cues, but notes that the behavioral effects of such outputs remain poorly quantified. The authors describe this as a missing piece in understanding how LLM-generated text affects real-world choices.
To address this, they introduce a behavioral computing framework, though the abstract provides only a brief sketch of its methods. The framework appears intended to track and measure the downstream effects of manipulative language on users over time.
Since the source is a single abstract, there are no conflicting findings to compare. The main takeaway is that the researchers see a gap in current measurement and propose a systematic way to fill it.