UF’s journalism school unveils Authentically, a new AI-powered program that reduces bias in writing

After more than three years of research and development, the College of Journalism and Communications is unveiling Authentically, a writing program assisted by artificial intelligence to identify potential bias in news coverage. Authentically can comb through an article, flagging language that could reflect unconscious bias or editorializing, then provide suggestions for language that is more neutral.

Created by Janet Coats, the managing director of the Consortium on Trust in Media and Technology and Kendall Moe, the senior project manager for Authentically, the program is set for a wide launch later this year. 

“Our next goal is to license this technology and get it into a newsroom, not just to stress test it but also to prove its value,” said Coats.

The idea for Authentically was generated from research that the consortium pursued around language use in journalism. The goal was to apply data, not only an editor’s instinct, to ferret out problematic word choices.

Moe, who earned a degree in linguistics at UF, applied academic linguistic methods to identify patterns of problematic word choices. Her work showed that subtle word changes could improve the precision, accuracy and authenticity of language, reducing perceptions of bias and building audience trust.

Using that research, the consortium team set out to build a practical tool that could check for biased language in real time, assisting journalists in making more thoughtful word choices rather than defaulting to common story framing. Recognizing the need for a range of skill sets, Moe and Coats recruited a multi-disciplinary team from colleges across UF.

Computer science students from the College of Engineering developed the technology, and visual communication design students from the College of the Arts created the user interface and branding materials.

The team wanted journalists to be able to use the tool seamlessly as they were writing. In its current version, users copy and paste their story into Authentically, and the tool returns a marked-up version of the story, highlighting language that may be imprecise or biased. 

While the primary goal is to combat bias, it can also provide suggestions that make the story more concise or trim flowery language. The tool is descriptive, not prescriptive, so it offers suggestions that users can accept or dismiss, returning to their original language if they choose.

Users can customize the tool to meet the needs of their particular organization, importing their own style guides or prioritizing particular types of bias for the tool to search. Users can also choose from different large language models for analysis, including a custom LLM designed by the team or models such as Gemini, ChatGPT and Claude. The ability to customize Authentically could expand its utility to fields outside of journalism, including strategic communication, health communication and education.

“This tech is a foundation to build on,” said Moe. “Right now, it’s a trust tool for journalists, but we also envision it as an image tool for public relations or an analysis tool for speeches and video scripts.”