Skip to content
Article WAN-IFRA Jun 2026

WAN-IFRA: what 168,000 AI actions in 18 newsrooms actually look like

What the article is about

WAN-IFRA published usage data from Xalok Assistant, an AI tool deployed across 18 digital newsrooms. Between January and April 2026, journalists made 168,363 AI-assisted actions. The article uses this dataset to argue that the public conversation about AI and journalism has been focused on the wrong tasks — and that the actual distribution of AI use tells a more specific story about where editorial friction lives.

Context

The newsrooms in the dataset represent roughly 1,200 users, with about one in five using the tool five to eight times daily. Rather than surveying journalists about what they think they do with AI, the article reports what they actually did at scale across four months, which makes the findings more grounded than most coverage of AI in editorial work.

Key findings

The usage breaks down to approximately 21% in content creation, 48% in editing, and 31% in distribution tasks including SEO, tagging, and translation. The most-used individual functions were bold formatting (around 23,000 uses), transcription (around 20,500), tagging (around 19,000), spell-checking (around 18,500), and SEO metadata generation (around 10,500). Headline generation, one of the most discussed AI applications in journalism coverage, ranked far lower at approximately 4,000 uses.

The article frames this as a “newsroom-in-the-loop” model: AI removes friction from repetitive supporting tasks, and journalists retain decision-making authority over the parts of the work that require editorial judgment. Adoption is uneven — frequent users account for 55% of all actions, suggesting a pattern where a minority of journalists develop deep habits while most remain occasional users. Some newsrooms in the dataset reported up to 30% more digital output and AI-assisted content reaching up to 40% of their digital audience.

Who should read this

Editorial teams evaluating AI tools and looking for realistic benchmarks rather than vendor-supplied projections. Also useful for editors making decisions about where to introduce AI support — the data suggests that productivity gains come primarily from the infrastructure around articles (formatting, metadata, transcription) rather than from drafting or headline generation.