AI Writing Tools Lengthen Change.org Petitions, Yet Fail to Improve Outcomes

Cornell University researchers have determined that artificial intelligence writing assistance on Change.org alters petition language but fails to improve campaign success. The study, led by doctoral student Isabel Corpus and Professor Mor Naaman of Cornell Tech, analyzed 1.5 million petitions to evaluate how the platform’s AI drafting tool affected user engagement and outcomes. Published September 30 in Nature Human Behaviour, the research leveraged a staggered rollout of the tool across English-speaking markets, including the United States, United Kingdom, Canada, and Australia. By comparing petition trends before and during the October to December 2023 deployment window, the team isolated the AI’s influence from broader platform usage patterns. The data revealed that AI-enhanced petitions consistently increased in length and lexical complexity while displaying greater homogeneity in phrasing. Titles and body text shifted toward formal imperatives such as mandate, implement, and urge, reflecting training data drawn from historically successful campaigns. Despite these stylistic changes, campaign effectiveness remained stagnant or declined. Metrics tracking early engagement, specifically the proportion of petitions receiving their first comment within thirty days and those reaching ten signatures, showed no measurable improvement. In fact, comment acquisition dropped approximately five percent relative to pre-implementation baselines. Researchers attribute the lack of impact to behavioral and psychological factors inherent in grassroots advocacy. AI-generated drafts tend to be less specific and lack the granular details that typically galvanize supporters. Furthermore, the study suggests that petition sharing relies heavily on trust networks, with users primarily signing causes recommended by personal contacts or established organizations rather than discovering them through platform browsing. When AI drafting reduces the author’s personal investment, the resulting decrease in organic sharing likely suppresses broader engagement. The findings challenge prevailing assumptions that algorithmically optimized text automatically drives persuasion or mobilization, highlighting instead the continued centrality of human authenticity and specificity in digital activism.