A new artificial intelligence writing tool made online petitions longer and more alike, but did not improve two measures of early engagement, according to a study involving Cornell researchers. The researchers examined 1.5 million petitions across four English-speaking countries.
The study published Wednesday in Nature Human Behaviour examined Change.org as the platform gradually introduced its writing assistant in 2023. A Cornell Chronicle account described the rollout and findings. The study concerns access to an in-platform tool, not a verified count of individual petitioners who used it.
The tool became available first in the United States, Britain and Canada, then in Australia 11 weeks later. Researchers compared petition trends before and after rollout across those countries, using the staggered introduction to help separate the tool's effects from broader changes over time. They also analyzed repeat petition writers who posted before and after gaining access.
Petitions created after access to the tool grew longer and used more varied, complex words, while titles became more similar to one another. The team found more titles calling on readers to “implement,” “mandate” or “urge” an action, wording that had been less common before the rollout.
But those language changes did not translate into stronger outcomes on the study's measures. Researchers tracked the share of petitions receiving at least one comment within 30 days and the share reaching 10 signatures by the time data were collected. Neither measure improved; the comment measure fell about 5% relative to its pre-tool baseline, the Cornell account said. That is a relative change, not a five-percentage-point drop.
Those measures capture early engagement, not whether a petition changed a law, policy or institutional decision. The researchers did not establish that AI assistance caused any particular petition to fail, nor did they show that every petition created after rollout used AI.
Lead author Isabel Corpus, a Cornell information-science doctoral student, said the limited effect surprised the team. Cornell professor Mor Naaman suggested one possibility: people may be more likely to sign a petition shared by someone they trust than one they encounter while browsing the platform. That explanation remains a hypothesis, not a measured cause of the study's results.
The study included researchers at Cornell Tech, Cornell's Bowers College of Computing and Information Science, and the University of Michigan. Its findings apply to Change.org during the period studied, not necessarily to all AI-assisted writing or to petitions on other platforms.
FingerLakes1.com is the region’s leading all-digital news publication. The company was founded in 1998 and has been keeping residents informed for more than two decades. Have a lead? Send it to news@fingerlakes1.com.