CNA built 20+ custom GPTs before calling itself an AI newsroom

The pain-point that changed the culture: Parliament AI

CNA's Editor-in-Chief Walter Fernandez pinpoints one use case as the turning point for newsroom buy-in: covering Parliament. Sittings are long and arduous, so the team built a tool that recognizes the faces of over 90 MPs, transcribes speeches, and generates searchable summaries.

This is a specific and repeatable pattern. Rather than mandating AI top-down, CNA asked journalists what their biggest pain point was, then built for the loudest one. The tool's value was legible immediately — reporters could see it doing work they otherwise did by hand.

When reporters saw AI solving a real challenge, everyone came on board. Then it became a matter of prioritizing which pain points to address next.Montana Labs

The lesson for applied teams: adoption followed a demonstrated win on a task people already dreaded, not a general pitch about capability. The backlog of subsequent pain points became the roadmap.

What the election coverage actually surfaced

During Singapore's recent General Election, CNA used ChatGPT two ways. First as a "second brain": internal GPTs seeded with verified information to give reporters context. Second, OpenAI's reasoning models analyzed campaigns, including manipulative social media activity.

The concrete result Fernandez cites is worth noting because it wasn't a prompted task. The model surfaced a link between two suspicious accounts that had changed their profile names during the campaign — an anomaly the newsroom hadn't asked about or even considered.

That distinction matters. The value wasn't summarization or drafting; it was pattern-finding across data at a scale the newsroom couldn't manually reach. Fernandez frames this as doing things "we simply couldn't do before" rather than doing familiar work faster.

A year of guardrails before scaling to 2,500 licenses

CNA began experimenting in 2019, before ChatGPT existed, but it spent roughly a year drafting AI guidelines, standing up cross-functional oversight, and enforcing human-in-the-loop processes before going wide. The bright lines are explicit: no cloned AI voices and no AI-generated footage in news coverage or documentaries.

Fernandez also rejects "vanity projects" — every deployment must solve a real problem. Notably, he refuses the "AI-first newsroom" label even while describing the organization as "all in."

Our North Star remains public service journalism, with AI as a tool to help us fulfill that mission.Montana Labs

The scale figures give the guardrails weight: more than 500 enterprise licenses across CNA and 2,000 more at group level, paired with basic and advanced training, hackathons, and cross-functional teams. The policy work preceded, rather than trailed, mass distribution.

The differentiator CNA is betting on: quality in a sea of slop

The specific implication of CNA's account is a claim about where competitive value moves when generation becomes free. Fernandez argues that when AI can produce infinite content and clone likenesses in minutes, the differentiator for newsrooms is no longer language, format, or medium — it is the quality and relevance of content amid what he calls "AI slop."

That reframes the twenty-plus custom GPTs, including the popular "Newsroom Buddy" that checks work against the CNA style guide. These tools aren't meant to raise output volume; they exist to free journalists for more ambitious reporting while the brand competes on trust and editorial judgment.

For teams building in adjacent industries, CNA's sequence is the useful artifact: identify a concrete pain point, ship a tool people can see working, codify the rules, then scale distribution and training. The strategy is aggressive on adoption and deliberately conservative on what the technology is allowed to fabricate.