This is the story of a financial newspaper that answers its readers' questions from its own reporting. I'm showing it as The Meridian, a paper I made up.
Search on a news site gives you headlines by date. To answer a real question, readers opened story after story and pieced the answer together themselves.
Trust was the whole question. The Reuters Institute found that only about a third of people are comfortable with news made by journalists with AI's help, and fewer than one in five with news made mostly by AI. So an AI answer is only worth reading if you can see the reporting behind it.
Google came to the paper with an AI summary of every search, and the paper didn't like it at all. Who would set the tone of an editorial answer? How would they control it? Would their journalists become redundant, replaced by AI? So I proposed something else: pull verbatim quotes from the editors and writers instead. They loved it. It used technology to lift up the people, not replace them. Every insight shows which articles, videos and editors it came from.
The business case was simple: we wanted people to read more than one article. And when they were done, we'd recommend stories from the group's other titles, so they'd stay in the ecosystem. A subscription to a paper like this is a big deal. You pay for great journalism, not AI filler. So the little things were about feeling seen and wanting more: an audio overview to catch up on the go, and a short note on why the paper picked this read for you. The newsroom gained too. Finding the right photo used to mean digging through an archive of odd file names and bad metadata; now we could suggest pairings. And charts took seconds, almost in real time, which matters for breaking news, or a last-minute story that needs numbers behind it.
The CEO and CTO of the paper's media group were in the room. I presented to them, backed by Google Cloud's excellent sales and engineering teams.
The idea: ask the paper a question, get suggestions to sharpen it, then an answer in short insights, each linked to the articles, videos and editors behind it.
The flow goes from the first question to insights, sources and charts, and then to saving the search and getting summary emails as the story develops.
The vision: the paper answers wherever you read it. On the desktop, on a phone, where a lot of reading happens, and in your inbox as the story moves.
The most important thing to launch first was conversational search. The paper's search matched keywords exactly, so we wanted to move to something you could talk to, even by voice. Insights, saved searches and summary emails would build on top of it.
We built a working prototype from the first question to a saved search, so the team could try that future instead of imagining it. I sent it over, the client went on to sign a multi-year deal, and I moved on to the next client. From there, timelines and priorities belong to the client's own engineering team. The original, in the real paper's branding, is behind the lock.
A question for my students from this one: what changes when a source can talk back, and what does it have to show so people can still trust it? Thanks for watching.