Mediacorp, Singapore’s national media network, created a GPT that eases its staff’s workload, helps train them — and brings in revenue. It has led to rapid growth in social video views — with absolutely no increase in output.

The journey began after their audience growth team talked to 100 or so journalists about the importance of presenting for social video. 

“And then everybody wanted to be a creator. We started getting pitched, all the time,” Lyn-Yi Chung, chief editor for growth and innovation, told participants of the Newsroom Innovation seminar in Berlin.

“That means we were being enthusiastically pitched to, but it was to the point of harassment. The ideas came late into the night, over the weekends, and people wanted to know, ‘Why aren’t you taking my idea?’”

To take some pressure off their editors, Chung’s team created “SnackBot” (so named because it helps them create “snackable” content with bots). They downloaded all our audience data for social video and trained a GPT on their top 500 best-performing TikTok and Instagram videos when measured by views and engagement.

“And then we decided, OK, this will answer whether it’s a good idea,” Chung said. “It assesses journalists’ pitches, but it also coaches them. The key thing is that we didn’t want to dampen people’s enthusiasm. So it takes you end to end. More importantly, it teaches you how to film yourself; it gives you production guidance.”

The team also uploaded some of the scripts that really did well, Chung said: “Again, because people came with very abstract ideas, it helped people sharpen their angle.” 

Mediacorp was “very transparent” when explaining the criteria used in assessing the pitches and helping reporters flesh them out. 

“If you give a not-so-good idea, it doesn’t dismiss you. It tells you maybe you should do this instead.”

For example, a reporter pitched a video on the Beijing Auto Show, typically an industry-oriented event that is “not exciting to somebody on TikTok.” The GPT explained why the pitch didn’t work for TikTok and offered visual direction on what to film and how to sharpen the hook for social video. The end result was an attention-grabbing video on new and interesting vehicles at the show. 

SnackBot has also smoothed relationships in the newsroom, she said.

“Somehow, when ‘computer says no’ as opposed to an editor on my team saying, ‘No, I don’t think that will work,’ it somehow helps a lot,” Chung said. “People feel, ‘OK, this is grounded in audience data. It’s evidence-based, so I’m not going to feel upset. I will try again.’” 

It has also led to results that are significant to the broadcaster’s business: Mediacorp is now regularly producing at least one viral video daily that brings in more than a million views a day.

This applies even to topics that would typically be considered “very, very dry — like parliament,” Chung said. The GPT helps reporters understand which questions in parliament have high potential (for example, those dealing with the cost of living), which have average potential, and which ones to avoid entirely (the government’s R&D expenditure, for example). 

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

Banner art: Adobe stock image by Gorodenkoff.

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