Researchers at a business school studied how generative AI writes online headlines. They found that if the AI learns the reasons people click, it can create more engaging and more trustworthy headlines. The team examined A/B testing, where two headline versions are shown to different readers to see which one gets more clicks.
They used headlines from the site Upworthy and let the model propose simple explanations for why one headline did better. The researchers then tested those explanations across more data and kept only the validated ones. After that, they fine-tuned the AI so it writes headlines that work for the right reasons. In people tests, the new model’s headlines were chosen more often than standard AI headlines.
Difficult words
- researcher — person who studies a topic and collects informationresearchers
- generative — able to produce new text or content
- headline — short title used for a news articleheadlines
- test — action to check which option works bettertested, tests
- validate — check and confirm that something is correctvalidated
- fine-tune — make small changes to improve performancefine-tuned
- model — computer program that makes predictions or examplesmodel’s
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Have you ever clicked a headline because it looked interesting? Why?
- Do you prefer headlines written by people or by computers? Why?
- What makes a headline feel trustworthy to you?
Related articles
Antibody and EGFR–STAT1 pathway point to new fibrosis treatments
Researchers at Yale found a human antibody that blocks epiregulin and lowers fibrosis markers. They also show EGFR activates STAT1 in fibroblasts, suggesting two treatment paths: block epiregulin or target the EGFR–STAT1 pathway.
