Jaemark Tordecilla started his fellowship at Harvard University shortly after the rollout of ChatGPT, and he quickly noticed that “… everyone on campus was talking about AI, but hardly anyone on campus was talking about AI in the context of journalism.”

With nearly a decade of digital experience in journalism behind him, Tordecilla decided it “sort of fell on me to figure that out.”

During this week’s Webinar, Tordecilla — a journalist and technologist who specialises in building AI tools for newsrooms in the Philippines — shared some of his discoveries and provided an overview of how AI is affecting the industry and how news media companies can derive the most value from it.

A terrible journalist

One of Tordecilla’s first projects was a reporting tool that could process reports with massive amounts of data. He developed a custom GPT and gave it a persona, which was to pretend it was a journalist and approach the information as a journalist would.

That led to an important discovery that informed his work going forward, he said: “It turns out that ChatGPT was a terrible journalist,” Tordecilla said. “It wasn’t finding any insights for me. I realised I was a better journalist than AI. So why was I asking it to do my job for me?”

Using AI, he learned, required understanding what machines are good at and what requires human insight: “You need to figure out what you bring to the process. So in this case, what I was good at [was] investigative journalism [and] editorial judgement,” he said.

“AI is good at processing large volumes of text and at generating summaries fast.”

He created a tool that would help reporters sift through massive amounts of text and data, then provide headlines or summaries. It was immediately useful for experienced journalists on the beat because, just by looking at the headlines, they could determine what was most significant about the report and whether it was worth a deeper dive.

“It helps them make the decision on whether they need to invest more time on a particular report or to move on,” Tordecilla explained.

A game changer for data discovery

His next tool took on a true Goliath: the Philippines’ budget.

Although the budget is technically available to the public, it has a daunting 700,000 line items, all presented in an Excel file. Just downloading it was a time-consuming effort, and trying to find or filter information was a nightmare for humans, but perfect for technology: “AI is good at investigating even data sets that big,” Tordecilla noted.

He created a tool that allowed journalists to “chat” with the budget and ask questions about specific areas, such as how much is allocated for Internet expenses at every school. “Just by chatting with the budget, I've been able to find a lot of stories that are interesting,” he said.

For instance, he was able to map every public works project for 2025 and found that most of the money went to road work and flood control. Since basketball is a popular sport in the Philippines, he asked the bot how many basketball courts were being built in a three-year span. One rural engineering district immediately stood out, as it had more than 100 basketball court projects slated for just 30 villages. This red flag required some human investigation.  

“I wouldn’t have found this information if I had to do deeper data analysis, if I had to programme it,” Tordecilla said. “But by chatting with it using AI, I found it easily. And it’s a good example of where the job of the AI ends and where the job of the reporters begin — because now you can go to the people who grew up the budget and ask them, what’s up with this?”

Other interesting discoveries included a look at advertising budgets for the U.S. House of Representatives, which had a larger ad spend than the health and tourism departments combined. Digging deeper, they discovered the budget ballooned in 2024 — right after a new House speaker, whose family owns media companies, took office.

“I’m not accusing anyone of anything, but it’s sort of shady when you install a new House speaker who happens to own media organisations and then suddenly the media budget of the agency that he oversees blows up,” Tordecilla said. “Now it’s up to the reporters to go to the House of Representatives, inspect that budget, and ask for details about where this money is going.”

AI gives a voice to news stories

Aside from assisting reporters in scouring data, Tordecilla is using AI to help tell stories in new ways.

According to the latest Reuters Institute Digital News Report, he said, the Philippines has the highest proportion of audiences who prefer watching the news instead of reading it. Tordecilla used this information as an incentive to launch yet another experiment, and turned a 4,000-word long-form investigative feature into a 33-minute video using AI for all the illustrations, animations, and even the voiceover and music.

The video took just two days to create — a fraction of the time it would have taken humans, he noted.  

“Most people who saw the video found the images compelling, especially because the illustrations fit in perfectly with such a graphic story, which was about an assassin who worked for the former president when he was mayor of a city in the Philippines,” Tordecilla said.

Although most people received it warmly, he said there was pushback from artists who objected to the use of AI-generated images.

The voiceover also received mixed reviews. Although people in the United States thought it sounded too robotic, those in the Philippines — including the original author of the story — found it nearly indistinguishable from a real human voice.

He attributed the different reactions to the fact that the voiceover had a “slight hint” of a Filipino accent, which is how every college-educated Filipino speaks English. And that produced a new takeaway for him: “I think it also underscores a need to have representation inside these data sets.”  

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