The Haitian Times was founded in 1999 to bridge the generational and geographical gaps among Haitians. They’re a small, nimble team led by Publisher and Editor-in-Chief Vania André.
When André first experimented with ChatGPT, she wasn’t trying to transform her newsroom. She was trying to get her editors to fill out a form.
A submission form that the editors kept skipping became a custom GPT that now sits at the heart of the Haitian Times’ editorial workflow, cutting story production time from nearly two hours to 40 minutes.
The foundation came first
Before AI entered the picture, the Haitian Times had already done the hard work of operationalizing its editorial process. Because they have a small, multilingual team, they realized quality was at risk because small errors kept slipping through the cracks. So, they documented the submission process and built a form capturing every field needed to publish a post: Headline, subheading, meta description, article summary, and more.
They also developed a detailed internal style guide that went beyond AP style to include psychographic information about their readers, guidance on transcreation versus translation for Haitian Creole and French content, and specific cultural context (for example: don’t describe Haitians as “resilient” — a word the community finds condescending).
“We prefer transcreation versus translation. So, translating the meaning and the spirit of a phrase versus the literal translation,” explained André. These details on preferences specific to their newsroom are all ones they supplied the custom GPT as training materials.
Custom GPTs are a user-created version of ChatGPT tailored to niche use cases. It’s similar to a Claude skill where you can create a structured, reusable set of instructions that can help you perform repeatable tasks more efficiently. Alongside the style guide and submission form, André fed it Google Search Console reports, Google Analytics data, and their site’s content taxonomy — essentially everything an experienced editor would need to make smart decisions about framing, copy, and SEO.
Getting buy-in: show, don’t tell
The team was skeptical at first. Even after André walked them through how she trained the tool, the questions kept coming: Will it understand our voice? Will it get our style right?
Her answer was to stop explaining and start demonstrating. During daily editorial calls, she’d share her screen, paste in a draft, and let the GPT generate the submission form in real time with the correct format, SEO-optimized, and taxonomy categories populated. Seeing it work live, repeatedly, is what moved editors from apprehensive to bought-in. Today, all editors use it as a standard part of their workflow, without any prompting from André.
Two edits, two use cases
Every story gets two editorial passes, and the custom GPT plays a different role in each.
First edit focuses on voice, structure, and framing. Editors also use the GPT to flag data visualization opportunities. If a paragraph is dense with numbers, it goes into the tool to assess whether there’s enough for a chart or graph, and if not, what additional data to pursue.
Second edit is a copy polish: grammar, style, and active voice. Editors paste in specific sentences or paragraphs that feel clunky and ask the GPT to clean them up using explicit instructions not to change tone or overhaul copy, just tighten it. The tool also generates the complete submission form at this stage, including headlines, subheadings, and categories.
The result: from two hours to 40 minutes
The time from filed copy to a live story has dropped from roughly an hour and a half to two hours, down to about 40 minutes — a reduction of more than half. That’s a significant gain for a lean, under-resourced newsroom covering fast-moving news.
What’s next?
André is experimenting with building a Claude agent that would automate social media content creation by pulling published stories, applying brand templates in Canva, and generating carousel content without manual intervention. It’s a work in progress, but it reflects the Haitian Times’ operating philosophy: identify the bottleneck, experiment without fear of breaking things, and build incrementally.
The Beacon’s approach to AI was cautious but experimental. Its team adopted a “testing mindset” to explore how AI could help their newsroom.
