In 2019, I wrote a story for BBC Travel about how Iceland helped humans reach the Moon.
It was an unusual piece of journalism. The BBC team described its multimedia format at the time as groundbreaking. It combined reporting, photography, video and narrative to tell the story of how NASA astronauts trained in Iceland’s lunar-like landscape before the Apollo missions.
This week (seven years later), I went looking for it again. I typed “BBC Travel Iceland moon landing” into Google.
Instead of my story, the first thing I encountered was an AI Overview telling me the answer. Stranger still, the most visible attribution was to Facebook.
For a journalist, that was mildly depressing. So I tried something else.
I removed the BBC from my question and simply asked Google: “How did Iceland help humans reach the Moon?”
This time something interesting happened. Google’s AI found my BBC story. It used it repeatedly as a source for its answer.
Then I asked another question: “Can you recommend the best long-form article where I can read more about this story?”
It recommended mine.
That changed what I thought this experiment was going to tell me.
We’re increasingly being told that publishers need to make journalism discoverable and understandable by AI. Articles need clear structure, identifiable facts, explicit attribution and information that machines can confidently extract and cite.
All sensible. But my BBC story was written in 2019. I wasn’t writing for generative AI. Nobody was.
I was writing for people.
Yet seven years later, an AI could understand the facts, extract them, attribute them and still recognise that the original story offered something worth reading.
Perhaps there’s a lesson in that. Publishing has been here before.
When Google became the dominant gateway to journalism, we learnt to write for search. Some of that improved journalism. Clear headlines and better structure helped readers too.
But we also created SEO journalism: keyword-heavy headlines, repetitive explainers and articles written partly to satisfy an algorithm rather than the person reading them.
We shouldn’t make the same mistake twice.
Yes, journalism needs to be legible to machines. Sources should be explicit. Facts should be clearly attributed. Names, dates and context should be unambiguous.
But that doesn’t mean journalism should be written for machines.
Because the things AI can extract aren’t necessarily the things that make someone want to read.
Narrative. Curiosity. Surprise. Judgement. Personality. Great photography. A beautifully told story.
Those remain stubbornly human.
Over this series I’ve written about disappearing homepages, AI products, changing habits and the battle for trust. All of them point towards a radically different relationship between publishers, readers and technology.
Yet perhaps the final lesson is reassuringly old-fashioned. Make the facts easy for machines to understand. Make the journalism worth reading for humans.
And let the machines work out the rest.