At Hearst Newspapers, the emphasis is on creating innovative, high-impact, high-value content that can’t be found elsewhere. During the recent INMA Webinar Future-proofing news in the GenAI era, members heard how the company uses AI innovation to build loyalty.

Tim O’Rourke, vice president/content strategy for Hearst, is founder of the Hearst DevHub, a small team of journalists who learned different programming languages and approaches to code-based journalism to create unique content. The DevHub plays a key role in helping provide interactive local content at the calibre of much larger publications such as The Wall Street Journal and The New York Times.

“We really think about quality first, uniqueness first, and how can we separate ourselves in a very crowded news environment that’s only going to get more crowded with AI-generated content flooding the scene,” O’Rourke said.

For years, the team has relied on three strategic lanes:

• Editorial engineering. This is a team of backend coders that build tools to create new experiences for the newsrooms.
• Visual storytelling. This is the source for high-level data analysis and visualisations.
• Self-service templates. This offers “Click 2 Publish” templates that newsrooms can use to create interactive storytelling without coders.

The strategic approach has had remarkable results on the bottom line:

“This kind of interactive work, this project-level data-driven work, outperforms our core content in every metric that we trace,” O’Rourke said. “It doesn't mean that we want to stop doing day-to-day news coverage, but it does mean that when we make these investments, when we take the time to do a little bit extra, when we think about a different way to tell stories to our audiences and we wow them, that we get a benefit from that.”

In fact, these data-driven projects enjoy four times the reach of regular articles and boast an 18 times conversion rate.

“When it comes to taking passive readers and turning them into paying subscribers, that this kind of high-level interactive data-driven work really does outperform,” he said.

Now, as it moves into the future, Hearst has added a fourth lane: AI and automation.

Driving in a new lane

As Hearst moves into automated territory, the company is taking a conservative approach to how it works with AI and are not rushing in to use AI-generated content, O’Rourke said. The DevHub is in “constant conversations” with the leadership team, legal team, and newsrooms, he added.

“We’re involving our journalists. Of course, we’re building scalable tools, but we’re doing so with caution in mind and, every step of the way, we are keeping humans involved.”

The company is aware of the risk of using AI-generated content, particularly for such a local-minded company, he said: “Our reputations [and] our brands are what separate us in a lot of ways, and we want to make sure that that’s front of mind at all times.”

At the same time, this is no time to stand still; the company is embracing this transformative technology but is clear that it won’t mass-publish AI-generated content or allowing any content to be published without human oversight.

“We are being as transparent as we can possibly be. Overkill would probably be a good way to put it, but at this point, we feel that that’s necessary.”

One of the approaches Hearst is preaching to its journalists is what it calls the sandwich approach, with humans on either side of the AI-generated content. That means humans come up with the ideas, think about them critically, and will do the final check to edit at the end before something is published.

In the middle, he said, is where technology can accelerate the work: “This is where we can take some of these rote tasks we’ve implemented over the last decade and reduce the amount of time that we spend on them so that our journalists can be thinking on higher-minded things.”

Creating with AI

A recurring internal question is how AI can help Hearst create, build, and produce more high-quality (and better) local journalism.

“This isn’t about how we can make a buck,” he said. “It’s about how we can use this transformative technology to do better work locally, which is really, really hard — and ever harder as belts tighten.”

One answer has been to bring an AI engineer onto the DevHub team and create a series of products O’Rourke shared with INMA members.  

• Radar: This automated programme is built on the back of traditional scrapers and helps the newsroom get stories ahead of its competitors. The scrapers comb Web sites and then plug the fresh information into the company’s Slack interface, allowing reporters and editors to get information faster. “We’ve been able to get in front of the story and get in front of the competition and create scoops that have really put us ahead and had a tangible positive effect on our coverage and our bottom line,” he said

• Producer-P: This is the first internal AI tool the company developed about 10 months ago. The audience optimisation bot is a “co-pilot for producers” that saves time on story production tasks such as SEO URL suggestions and SEO-friendly headlines. O’Rourke said it has been particularly helpful in smaller markets and has saved “a tonne of time” since newsrooms began using it.

• Chowbot: This audience-facing local restaurant recommendation engine is built from more than 60 guides and 1,000+ restaurant reviews from its food and wine team at the San Francisco Chronicle. Instead of giving users info from Yelp, it provides recommendations from a trusted source. The app has been live for more than a month and has “gotten a little bit of notice.” More importantly, O’Rourke said, it showed the team what it was capable of doing and provided a springboard to bigger initiatives.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .