
Thirty-seven per cent. That is how many people across 48 markets say they trust most news most of the time, according to the Reuters Institute Digital News Report 2026, the lowest figure since the institute began measuring in 2015. In the same survey, 10% of people now use AI chatbots for news, up from 7% a year earlier.
Those two numbers pull in different directions. Readers trust news less, and more of them meet it secondhand, summarised by a machine. What does a newsroom do with that?
Why provenance became a product feature
Trust in AI chatbot answers sits at 20% in the Reuters data, well below trust in news itself. Yet 42% of people who use chatbots for news say they always or often click through to the original source. They are checking. When they land on your story, the question they bring is simple: who says so, and how do they know?
A byline used to carry that weight alone. It no longer does. Readers now look for the date a fact was last confirmed, the source behind it, and a visible record when something was corrected. None of that can be bolted on at the template level if the content model never stored it.
What one newsroom built
Track where every claim came from and when.
That instruction comes from Code & Theory's editorial system for RealClear Crime, described in a Sanity write-up in August. An editor has to mark each stage reviewed before the next one unlocks. Every piece is labelled as AI-assisted and edited by a named human. The friction is deliberate: the team asked for it, because a fast system that nobody stands behind is worth little to a publisher selling credibility.
Four things the CMS should hold for every story
Whatever platform a newsroom runs, these belong in the content model as fields, so they can be queried, displayed and audited.
- The source of each factual claim, with the date it was checked, so re-verification is quick when a source moves.
- The review stages a piece passed, and who signed off each one.
- Corrections as their own records, linked to the story, so every page and feed that shows the story shows the correction too.
- A clear note on how the piece was produced, including where AI tools were used and which editor is responsible.
Corrections without a developer ticket
A correction that waits in an engineering queue is a correction readers do not see. Pagepro, a Next.js and Sanity migration specialist, replatformed GP Notebook, a 100,000+ subpage medical reference used by UK GPs, from a legacy custom CMS to Next.js and Sanity in 2022. Its editorial team now publishes independently of developers, and organic traffic doubled within 2 years of the relaunch.
Medicine and news share the same weakness: a fact goes stale, and every copy of it has to change at once. Structured content makes that a single edit. The team describes how it approaches these builds at https://pagepro.co/services/sanity-development.
Questions for your next CMS review
Can an editor publish a correction alone?
If the answer involves a developer, the correction will lag the error, sometimes by days.
Is the source of each fact stored, or only written into the text?
Text cannot be queried. A field can, which means you can find every story that cites a source the moment that source is retracted.
Would a reader see how the story was made?
Labels for AI assistance and a named responsible editor only work if the system requires them before publication.
So here is the question to take into the next planning meeting: if a key source in last month's biggest story turned out to be wrong tomorrow, how many minutes would it take your team to find every page, app card and newsletter that repeats it? A good answer is a number, and a small one.