Why Lianhe Zaobao is rethinking how stories are framed
“When we try to write an article, we pack all angles in it. That becomes very difficult for us to identify user needs,” said Tan Lee Chin, Digital Product Editor, SPH Media.
The newsroom used a custom GPT model developed with SPH Media’s data scientists to analyse a month’s worth of editorial output – 6,006 articles – as part of WAN-IFRA’s User Needs Bootcamp.
What began as an effort to better understand audience needs also prompted broader questions about story framing and how editorial value is measured.
The newsroom had already begun exploring user needs before joining the programme, but wanted a clearer structure for implementation.
“For our journey in user needs, we identified this as one of the goals that we’re going to achieve in our newsroom,” said Tan.
The bootcamp, she added, helped the team “put things together” and map out a roadmap for implementation.
Looking beyond ‘Update Me’
The baseline analysis revealed patterns that felt familiar. Like many publishers, Zaobao found it was heavily producing “Update Me” stories while allocating fewer resources to other user needs categories.
“A lot of publishers, what we have noticed, is we actually overproduce or allocate a lot of resources on Update Me articles,” Tan said.
But the analysis also suggested that some underserved story types could have a stronger impact with readers. Emotional stories, in particular, stood out as an area where the newsroom might rethink resource allocation.
The team also broke audiences into three broad groups:
- General readers,
- registered users, and
- subscribers
They found that different segments appeared to have different content needs.
“When we zoomed into our subscriber space, we noticed that actually actionable articles might be more useful if we can allocate more resources in terms of that,” she added.
The discussion also moved beyond traffic metrics.
Earlier exploratory work had focused largely on reach and pageviews. During the bootcamp, however, discussions increasingly turned toward engagement and attention.
The goal, Tan added, is to focus on “the quality of attention, not just purely on the volume of traffic.”
When stories became hard to classify
Some of the most revealing findings emerged while the newsroom was testing the AI classifications themselves.
The team ran blind tests with human reviewers to verify whether the custom GPT was categorising stories accurately.
In some cases, the AI assigned two or three user needs to a single article. But there were also stories where even editors struggled to determine the dominant need.
“We noticed that the custom GPT, when they actually classified two needs or three needs for a particular article, there were also articles where in our current setup, we couldn’t identify even as human,” Tan noted.
The finding also raised questions about how stories were being framed in the first place.
Rather than producing stories built around a clearly defined user need, the newsroom often approached articles by trying to include multiple angles at once. The result, Tan suggested, was journalism that could become harder to classify clearly within a user needs framework.
“That was a very interesting learning,” she said.
The newsroom is now considering whether stories should be tied to a single primary user need, or whether multiple needs can coexist within one article.
One possibility under discussion is assigning a dominant need for analytical purposes while still allowing secondary needs to exist editorially.
“That is something that we are trying to work out as part of our strategy,” she added.
Building a shared newsroom language
The process also exposed another challenge: translating the user needs framework into something editors and journalists could use consistently inside the newsroom.
“We realised that what we really need was a framework for ourselves,” Tan noted.
Part of that work involves adapting the framework into Chinese and aligning it more closely with the Zaobao brand and newsroom practices.
“We need to translate the user need itself into Chinese language and something whereby our newsroom leaders or our editors could actually tell their journalists how it works, especially during the assignment,” she said.
Later this month, the newsroom plans to bring editors and newsroom leaders together to review past coverage and discuss which approaches best reflect the Zaobao brand.
A shared framework could help standardise tagging and make collaboration across sections more consistent.
The newsroom wants editors and journalists to develop a common understanding of what different user needs actually mean in practice.
“It makes us share a common language,” she said.
Rather than adopting the framework in its standard form, Zaobao plans to adapt it into its own internal model.
Testing the framework section by section
From here, the newsroom plans to focus on three areas:
- Defining a Zaobao-specific model,
- piloting the framework in selected sections, and
- establishing metrics to evaluate results.
They have already identified two sections for initial trials. According to Tan, both showed signs that audience needs were currently underserved.
“We will be working with two sections that we have identified and run some trials and test the hypothesis,” she said.
They also plan to track engagement-focused indicators as part of the experiments.
“The framework will only be effective if we can see results,” Tan said.
For now, the rollout remains deliberately gradual.
The process begins with defining needs, testing approaches in a limited number of sections, and iterating before any wider expansion across the newsroom.
“We feel that this is a much more structured and phased approach,” she added.
The larger challenge, she suggested, will only become clearer once the framework moves from discussion into day-to-day newsroom practice.
“People have some ideas about it,” noted Tan. “But when we start to work on it, it might be a very different challenge.”
For now, the newsroom is proceeding cautiously.
“We are taking one step at a time,” she said.
