A new arXiv paper, “Advertising to Agents: Measuring Description-Framing Bias in LLM Tool Selection” (2605.23916v3), asks whether sales language in tool descriptions affects which tools an AI agent picks. In many agent systems, tools are chosen from registries where each provider writes its own description, so the wording is not neutral.
The researchers constructed pairs of listings that differ in one way—with praise as one example—to isolate the effect of framing. Their results indicate that such language can shift an agent's choice, meaning the description itself acts as a form of advertising.
Because this summary draws on a single source, there are no conflicting findings to compare. The paper's own abstract presents the bias as a measurable outcome, though the full details of the experimental design are not included in the excerpt.