As AI began entering newsrooms, Hearst Newspapers created a distinct plan for integrating into its network of local publications.

During the Webinar Using AI to help reporters, presented by the INMA Generative AI Initiative, Hearst’s Tim O’Rourke shared how that approach has been implemented. O’Rourke, who works on the company’s San Francisco-based DevHub team, explained that Hearst’s approach to AI is not about replacing journalists but empowering them with tools that streamline workflows, enhance reporting, and expand audience engagement.

Since the debut of ChatGPT in late 2022, Hearst’s DevHub — a 12-person innovation team — has worked across its network of local news sites to establish clear guardrails and practical applications for generative AI. From major metros like the San Francisco Chronicle and Houston Chronicle to smaller hyperlocal outlets, the team is committed to developing tools tailored to their unique needs.

“We always say, from an editorial perspective, we’re going to embrace GenAI, but we’re going to do it responsibly and conservatively and cautiously,” O’Rourke explained. “We keep humans involved in everything we do.”

That doesn’t mean backing away from innovation, but O’Rourke describes Hearst’s approach as one of “cautious innovation,” where trust with audiences and brand safety is key, and is anchored in transparency:

“We’re not going to mass-publish [AI-generated] content without review. We err on the side of explaining how we use this,” he explained. At the same time, “We really do try to push the envelope within the guardrails that we have in place.”

Technology with a human touch 

A key part of Hearst’s strategy is fostering a culture of thoughtful experimentation.

The DevHub has launched an AI Editorial Council composed of reporters, editors, and tech-focused staff from various newsrooms. The council meets biweekly to tackle ethical questions and operational challenges, and distributes a monthly newsletter to keep everyone updated.

“We’ve seen a lot of great ideas come through there, and we think it’s going to grow in size and scope in the coming months,” O’Rourke said.

Training also plays a crucial role in Hearst’s implementation process. The DevHub conducts high-level strategy sessions and hands-on workshops to demystify AI tools and share how journalists can use them. Sessions cover custom GPT development, approved vendor tools like NotebookLM, and practical use cases such as interview prep and source organisation.

This is important because the tools can be intimidating — which means they won’t be used.  

“If somebody’s out there and doesn’t understand how their people in other newsrooms or how their leadership wants them to use the tools, it can be kind of a Wild West,” O’Rourke noted. “We help lead people and supplement that with one-on-one talks and trainings and newsroom-wide conversations.”

The three buckets of AI integration

When it comes to AI, Hearst organises its efforts into three categories: digital production, news gathering, and reader-facing experiences.

1. Digital production. “These are things that you all probably have in your newsrooms in some capacity,” O’Rourke said, explaining this bucket includes tools for audience optimisation, headline generation, and SEO enhancement.

Hearst’s custom GPTs are designed to support fast-paced environments, especially in smaller newsrooms, but he noted they are intended to “speed up the process but not replace our editors.”

One standout tool is Producer P, a Slack-integrated bot that streamlines production tasks like crafting summaries for social media, alerts, and newsletters. Reporters and editors can customise outputs, ensuring editorial quality while saving time.

2. News gathering. AI is helping support core reporting functions with new tools. O’Rourke said Hearst’s Assembly tool, its most popular innovation, monitors public meetings by scraping video content and converting audio into transcripts using OpenAI’s Whisper.

These transcripts are timestamped and integrated into Hearst’s CMS, automatically alerting relevant reporters and editors, which allows for coverage of meetings that might otherwise go unnoticed.

This is particularly helpful on nights when two big meetings are happening at the same time and ensures reporters and editors will have information on both.

“It has really put us ahead, especially in city hall coverage,” he said.

Document analysis is another area where AI shines.

In Oakland, Hearst reporters used AI to sift through thousands of e-mails dropped during an election cycle, identifying patterns and constructing timelines that led to timely and impactful coverage.

Similarly, the Albany Times Union used AI to investigate “never events” in hospitals — cases where surgical tools were left inside patients — by analysing years of complaint data.

Audio file analysis has also proven valuable. During severe flooding in Texas, Hearst combined AI with data reporting and on-the-ground journalism to build a timeline from hundreds of EMS calls. This provided readers with a clear, factual account of the disaster’s progression and supported deeper investigative work.

Speech analysis is another emerging application. Hearst has explored changes in political rhetoric over time, examining speeches by prominent figures.

“We’ve got a whole small series in Texas on the different rhetoric from the attorney general, the governor, or other politicians on how their speech patterns have changed over the years and how they’ve gotten a little bit angrier, a little bit more extreme,” O’Rourke said.

Hearst has also done this for President Trump, comparing his inauguration speeches.

“It’s just another way to hold power to account and see what’s really going on behind the words that they’re saying,” he said.

3. Reader-facing experiences

Hearst is also using AI to create interactive tools that connect readers with journalism in intuitive ways.

Chow Bot, launched in San Francisco and San Antonio, is a restaurant recommendation engine built in collaboration with food critics. By structuring years of reviews and maintaining data accuracy, Chow Bot allows readers to explore local dining options through a conversational interface.

Another tool helps residents challenge property tax assessments. By inputting basic information, users receive a customised script to present before a review board, which makes an intimidating process more accessible. Hearst has received feedback from readers who saved thousands of dollars using the tool.

“We’ve already gotten feedback from real readers saying they’ve saved tens of thousands of dollars by using the tool and being able to have a easier way to go about doing this without having to contract with somebody,” O’Rourke said.

Most recently, Hearst published a public-facing version of its Assembly tool in Houston. Covering four school board meetings, the tool allows readers to search transcripts by keyword, track trends, and receive alerts on topics of interest. Summaries are reviewed by Hearst’s education team to ensure accuracy and complementing existing coverage.

Human-centred innovation

Throughout its AI journey, Hearst has prioritised human expertise, and reporters remain central to every tool and process.

“None of this is meant to be something that replaces [reporters’] work,” he emphasised. Instead, it improves efficiency whilst uncovering information that might have been overlooked.

“People like these tools,” O’Rourke noted. “Again, not everybody in the group obviously, but [we have] tonnes of buy-in and it’s really helping people save time and do better work.”

Hearst’s commitment to ethical AI use began early, with the DevHub team working closely with legal advisors to establish and publish a clear ethics policy.

Central to that policy is the requirement for human review of AI-assisted content — a standard the organisation continues to uphold. This cautious approach ensures that innovation never comes at the expense of editorial integrity or audience trust.

“We try to build around the expertise in our local newsrooms. That’s our value — not the tech,” he said. “But if we can replace old tech with new, easier tech, that’s a big win for everybody.”

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