How Hearst's DevHub is Building AI Tools That Work for Local News
HNP’s DevHub has been building AI tools that solve real problems for local newsrooms while driving value for subscribers and subscription revenue
Left: The Houston Chronicle’s Meeting Monitor is the latest tool fresh out of the DevHub. Right: Tim O’Rourke (Screenshot and photo courtesy of O’Rourke)
Tim O'Rourke is Vice President, Editorial Innovation and AI Strategy, at the Hearst media company’s newspaper division (HNP) and leads the DevHub, a central team of technically skilled journalists and programmers serving all Hearst newspapers from the San Francisco Chronicle to small hyperlocal newsrooms. (Hearst owns 24 daily and 52 weekly newspapers.)
What started as a data visualization team has evolved into an ambitious laboratory for AI experiments that respect journalistic principles while pushing innovation boundaries. Last week, I had the chance to talk to Tim and he gave me the run-down on what the lab has been working on.
Human-in-the-loop Principle
When ChatGPT launched in 2022, the DevHub became the natural home for Hearst's AI experiments. But O'Rourke's team took a deliberately cautious approach that puts accuracy first. "We're in the accuracy business," O'Rourke told me. "We have to ensure that what we're publishing is accurate and that we aren't putting an unfair burden on our readers."
This philosophy shapes everything the DevHub builds. Take their quiz generation engine EmCee (multiple choice), which transforms reporting into interactive content. The AI automates the grunt work, but human editors still review and refine every output. What used to take a full day of work now requires just 30-60 minutes of editing review.
The same human-in-the-loop principle applies to their reader-facing tools. Chowbot, their restaurant recommendation chatbot, launched in San Francisco because the Chronicle has "the best local regional food team in the country bar none”, according to O’Rourke, and years of structured restaurant data. The AI leverages this expertise rather than replacing it.
Building Tools That Make Money
Hearst's AI tools consistently drive business results. O'Rourke tracks traditional audience metrics, but more importantly, conversion rates for HNP’s subscription-focused sites.
"All the kind of stuff the DevHub works on tends to over perform for conversion and at a really high rate," he explains. "And that's why we continue to get support." The tools also perform well with their most committed readers, serving as value-adds that differentiate Hearst's local offerings from national competitors.
This success stems from the DevHub’s strategy of building unique local reader tools rather than generic content generation. Chowbot works because it's powered by local food critics' expertise, though O'Rourke acknowledges the conservative trade-offs this requires. As he told Generative AI in the Newsroom, "We don't want our readers to have to go and verify accuracy." They'd rather leave questions unanswered than risk providing incorrect information.
Beyond reader-facing tools, internal systems like Producer-P help journalists create optimized headlines and SEO content through Slack integration. The tool handles over 1,000 requests monthly across Hearst's network while maintaining a zero-error track record for factual accuracy.
How Internal Tools Become Public Products
The DevHub's most ambitious project so far is Meeting Monitor which recently launched for Houston Chronicle readers. The tool started as an internal journalist aid, helping reporters track local government meetings that they couldn't attend in person.
"We're already over 250 meetings covered across the group and all of our major markets," O'Rourke says, from vector control boards to city councils to state assemblies. "The reporters love it. And it's not because they don't go to these meetings. They go to the important ones, but what if there's two meetings on the same night?"
The internal success convinced the DevHub to build a public version. But this required rebuilding the models and user experience for general readers rather than trained journalists. "We had to redo the models and tweak the way that we were presenting information because it's different for a journalist than for a normal reader."
Meeting Monitor uses AI to increase transparency and put more civic information in citizens' hands without replacing core journalistic functions like analysis, context, and watchdog reporting.
Innovation Doesn’t Have to Cost Much
When I asked about costs, O'Rourke emphasized efficiency over raw spending. "We are extremely disciplined in how we use compute power," he notes. For a tool like Meeting Monitor covering hundreds of meetings, "something of that scale is not going to cost us much."
The real investment is human capital: the technical journalists who build these systems.
The DevHub operates on quarterly cycles, publishing "one big thing, whether it's AI driven or a data driven thing, per quarter with a lot of smaller scale stuff happening in collaboration with our newsrooms." This sustainable pace allows for the careful testing and refinement that makes their tools actually useful.
Scaling Without Losing Your Local Soul
The challenge for any innovation hub serving multiple markets is maintaining local authenticity while achieving economies of scale. Hearst's approach centers local expertise while leveraging shared technical infrastructure.
Chowbot's expansion to San Antonio illustrates this balance. Rather than simply copying the San Francisco version, they rebuilt it around San Antonio Express News' food critic and the local dining scene. "We really wanted to build on what we had done in San Francisco, but center the human, center the reporter even more," O'Rourke explains.
The San Antonio version performed better, with increased return sessions thanks to improved communication about database updates and promotion strategies. Based on these learnings, Hearst plans to update the original San Francisco version.
Data-driven projects might analyze national datasets, but local reporters provide the context and expertise that make stories relevant to specific communities. "That's what separates us," O'Rourke notes. "Central technical expertise mixed with local market expertise."
What Newsrooms, Product Teams and Publishers Can Learn from Hearst’s DevHub
Local knowledge trumps technology: For news organizations with limited resources, O'Rourke's advice is clear: don't let technology replace local knowledge that took decades to build. "Using this technology as something to lift up instead of replace, I think is the absolute key, because in five years, everybody's going to have the same tools doing the same thing."
Define your local niche: The competitive advantage lies in local expertise, not the AI itself. "You're going to compete against a big tech company if you're playing the same game. That doesn't seem too winnable. But I think you can win the local knowledge, information, unique content game if you play it right."
Blueprint for sustainable newsroom innovation: start with real problems, maintain human oversight, measure business impact, and never forget that local expertise is your competitive moat. In a landscape littered with AI experiments that often promise everything and deliver little, Hearst has built tools that actually work in local news markets.
Tools used for this newsletter:
Notta (interview transcription)
Claude 4.0 Sonnet Pro and Deep Research: Transcription query, first draft newsletter