Skip to content
Article BestAIFor Jun 2026

BestAIFor: agentic AI in newsrooms — where drafting saves time and where it doesn't

This article, published June 15, 2026 by journalist Claire Beaudoin on BestAIFor, examines how agentic AI is working in newsrooms in practice. It draws on Newsquest — the UK regional publisher — as a case study, alongside Reuters Institute survey data from 280 senior executives across 51 countries.

Newsquest runs 30 “AI-assisted reporters” using a tool called News Creator, which generates up to 30 story drafts per day. A human reviews every draft before publication. Beaudoin notes that Newsquest doesn’t publicly disclose how often those drafts are substantially rewritten — which is the figure that would indicate whether the tool is saving real time or simply shifting effort from one task to another.

The Reuters Institute data gives a wider picture. Sixty-seven percent of publishers say AI hasn’t saved any jobs yet. Only 44% describe their results as promising, while 42% call them “limited so far.” Seventy-five percent expect significant impact over time. The pattern across these responses is that efficiency gains tend to be absorbed into higher output volumes rather than cost reductions.

The article focuses on a failure mode specific to AI-generated text: data transposition. This includes swapped digits (a jersey number becoming another), unit errors ($1.2 billion becoming $1.2 million), and similar character-level mistakes that read naturally but are factually wrong. Catching these requires dedicated character-level verification against source data, not ordinary proofreading. Beaudoin argues this is why verification, not content generation, becomes the bottleneck in AI newsroom workflows.

A table in the article rates story types by time saved against verification load. Structured content — earnings summaries, sports statistics, local government voting records — gives the clearest net benefit. Council meeting summaries offer modest gains. Features, investigations, and breaking news typically break even or cost time. Beaudoin recommends avoiding AI drafting entirely for fast-moving breaking news, source-dependent stories relying on off-record context, and content where errors could harm readers.

The practical takeaway for editors is to start with the most structured beat, build a dedicated verification checklist before launch, and track how often drafts require substantial rewriting to judge whether the tool is actually saving time.