How my news podcast checks its own facts

Early episodes went vague whenever the morning's articles left something out. Now the writer lists what it doesn't know, and a separate step looks each question up before the hosts say a word.

My Dutch news podcast is written from that morning’s articles. Early on I kept hearing the same weakness. When the articles left something out, the hosts went vague. A story could say three Amsterdam restaurants had won an award and never name them, because the one article it was based on didn’t.

Worse, the script sometimes talked about its own homework, with lines about what “the write-up” or “our sources” did or didn’t say. Nobody wants to hear that on a news show.

The idea: write down what’s missing

My first instinct was to tell the writer to be more careful. That only produces more hedging. What worked was giving it a different job: for every story, list the open questions. An open question is a specific fact a curious listener would want next, or a point where two outlets disagree.

I ask for questions that can be searched. “Which three Amsterdam restaurants won?” is useful. “Is more information available?” isn’t. Anything that can’t be known yet, like a verdict that hasn’t been given, is left off.

What happens to the questions

WriteThe briefing is written from the full articles. Each story comes with its open questions, and the text itself doesn't hedge.
Look upA fact-checker runs once per story, all stories at the same time. It searches the web, favouring the primary source and established outlets, often in Dutch.
Answer or leave blankEvery answer needs a page the checker actually read. If it can't confirm something, the answer stays empty.
ReviseThe answers are folded into the story as normal news copy. The script is written from that version.

Here is what came back for three stories on the first test day:

Open question → what the checker found

Which three Amsterdam restaurants won a Bib Gourmand?All three named, with what each one cooks.Answered

Were any hospital systems actually breached, or were the shutdowns a precaution?A precaution. No breaches had been reported.Answered

How serious is the striker's injury, and will he play on Thursday?Out for both remaining matches, confirmed by the football association.Answered

Did any Amsterdam restaurants lose their award?Couldn't confirm it, so the story doesn't say.Left blank

On that first test it answered 13 of 14 questions.

What I changed along the way

It can’t break the episode. Each lookup has a four-minute limit. If one fails or runs out of time, that story runs as it was written. The worst a bad morning can do is delay the episode by a few minutes.
A disagreement isn’t an error. On the test day two outlets seemed to contradict each other about whether two Dutch staff abroad had been sent home. The checker found both were partly right: they were allowed to stay in the country but had to stop working at that post. So the instruction is to read the original reporting and give the full picture, not to pick a winner.
Watch the date. The articles are from early morning. If a minister has responded since, the checker includes that and says it’s the latest.
Never mention the research. The hosts don’t say “we checked”. The script is banned from talking about the write-up or its sources at all.
Follow-ups stay short. When a story continues from an earlier episode, the checker keeps to what’s new and doesn’t re-research the part listeners already heard.

What still isn’t right

The answers vary from run to run, the same way the rest of the pipeline does. Some of the pages it cites are weaker than I’d like, such as Wikipedia or an outlet with an obvious slant. The briefing I publish alongside each episode shows every question with its answer and links, so I can at least see where a fact came from when something sounds off.