Readers say they want AI-assisted news labelled. Most publishers still do not.

What readers say

The Reuters Institute's 2025 Digital News Report asked audiences in 47 markets how comfortable they were with news produced with AI assistance. Most said they wanted it disclosed, and comfort rose sharply when a person was described as having checked the output. Pew's 2025 survey of US adults found the same shape: a majority wanted labels, and a majority did not trust the labels they had seen.

The two surveys agree on the direction. On the size of the gap they do not: Pew put the share of US adults who say they have noticed an AI label on a news article at under one in five^†, a figure no other outlet in this set reports, so it is attributed rather than stated.

What publishers print

Publisher policies are easier to find than labels. Most large publishers have a page describing when they use AI. Far fewer put a line on the article. Press Gazette's audit of 50 English-language news sites found a per-article AI disclosure on 9 of them^†. Where labels do appear they are usually a checkbox the writer ticks, which is exactly the kind of label readers say they do not trust.

A label you cannot tick

The alternative is a label nobody sets. Brenzuri derives "reviewed by a human: yes" from the approval record of that exact version. If the version changes, the label goes back to no until someone approves again. It cannot be ticked by a writer, an agent, an API key or staff. A label with that property is not a promise; it is a record, and a record is what readers said they wanted.

Frequently asked questions