“I need a clear, precise translation that’s suitable for high-standard journalism platforms like The New York Times, The Washington Post or The Guardian. The tone should be natural and conversational, but professional, ensuring the meaning is adapted for an English-speaking audience. Also, please follow The Associated Press Stylebook.”
That’s the prompt used by the Center for Investigative Journalism, or CPI for its initials in Spanish, an award-winning independent news organization in Puerto Rico, to translate its stories from Spanish to English using artificial intelligence.
In December 2023, CPI received a grant from the American Journalism Project, a Washington-based organization that supports local and nonprofit news outlets, to expand the reach of its stories to English-speaking audiences. The result has been the integration of an AI assistant into CPI’s translation process—one that does not compromise the quality or cultural nuances of the stories.
Reaching that outcome took months of trial and error.
Noel Algarín, CPI’s English-language editor, led the project, which began in the summer of 2024. His first step was to take multiple previously published stories in Spanish and translate them using five different AI tools: ChatGPT, DeepL, Microsoft Word, Google Translate, and Claude.
The AI tool that demonstrated the best performance was ChatGPT. “The other tools showed significant inconsistencies, such as changing the gender of people mentioned in the stories, omitting entire passages of text, ignoring accents, or producing overly literal translations,” Algarín told LatAm Journalism Review (LJR).
Noel Algarín, CPI’s English-language editor. (Photo: Courtesy)
From there, Algarín developed a workflow in which ChatGPT’s translations were reviewed by editors. He also tested different prompts to provide the tool with more detailed translation instructions.
However, he noticed that ChatGPT’s translations were inconsistent. "Sometimes, it wouldn’t translate entire stories or would summarize them instead of translating," he said.
With the help of Liam Andrew, head of product and technology at AJP’s AI Studio, he pursued the idea of creating a customized AI assistant using API, OpenAI’s application interface.
"That’s when we started seeing more promising and consistent results," Algarín said. "This chatbot, specialized within OpenAI’s AI environment, was trained specifically for CPI’s translation needs, and we saw a clear improvement compared to previous efforts."
By working with the API, the team was able to reduce randomness or creativity in the AI’s responses, making it feel as if the article was originally written in English, Algarín said. The translations also better preserve the nuances of Puerto Rican contexts.
Over the past three years, CPI has sought ways to expand its reach beyond Puerto Rico. The newsroom has analyzed demographic data, conducted listening sessions, and visited different communities to better understand its audience’s needs.
"As part of that effort, we realized that making more of our content fully bilingual was essential," CPI’s executive director, Carla Minet, told LJR. "Thirty-five percent of our audience is in the United States."
When the American Journalism Project launched its Product & AI Studio—a new initiative to explore the smart application of emerging technologies in local journalism—CPI didn’t hesitate to apply.
"CPI's work on a translation GPT stood out as an opportunity to test how AI can support local newsrooms in serving multilingual audiences," Roshni Neslage, AJP’s director of communications, told LJR.
As part of the grant, CPI received $20,000 to carry out the project.
Before this funding, the process involved a human translator followed by an editor reviewing the translated text. "The translator is still part of the process, but instead of translating entire stories—many of which exceed 2,500 words—their role is now more focused on reviewing, correcting, and ensuring quality control," Algarín said.
Carla Minet, CPI’s executive director. (Photo: Courtesy)
In early 2025, CPI received a second grant from AJP—this time to explore AI tools for other organizational needs, including revenue generation, newsletter creation, and audience data analysis.
At the end of 2024, Algarín published a report detailing each experiment and lesson learned from the project. Neslage said this document could serve as a guide for other newsrooms looking to use AI in translations.
Among Algarín’s recommendations are the importance of direct interaction with the tools and refining them through trial and error. He also stresses the need for transparency about AI use to maintain audience trust.
The report also makes it clear that while AI has proven to be a valuable tool for improving translation productivity, human oversight remains essential—especially for refining storytelling and ensuring cultural sensitivity.
"The rigor in our translations has actually increased because there is now a review process with at least three layers of editing," Minet said. "We are not considering replacing anyone in the newsroom with AI. Our guiding principle has always been that this allows us to do more, speed up certain processes, but never replace our editorial methods. We remain firm in that position."
