Accommodation Made Cheap: AI Impacts on Replies to Online Reviews and the Signal Value of Tailoring

Accommodating a message to its recipient takes effort, and that effort is part of what accommodation signals. AI language models now reduce this effort. We study a masspersonal setting, restaurant replies to diners’ reviews on OpenTable, in six U.S. metropolitan areas from 2019 to mid-2026, with an interrupted time-series design around the introduction of ChatGPT. The detected AI likelihood of replies rose to 44% by mid-2026, compared with 1.2% for customer reviews. Replies became more tailored to the reviews they answered, most in quarters when AI likely wrote at least half of a restaurant’s replies. They also became more similar across restaurants but less similar within each restaurant. Before 2023, more tailored replies were followed by more reviews over the next year; from 2023, they no longer were, although this decline cannot yet be attributed to AI authorship. The form of accommodation may spread while the effort behind it falls.