Journalism and AI: What Is Left When a Machine Writes the News?

In 2015 the Associated Press was publishing about 300 reports on company earnings in a quarter. It then automated them, drawing on structured data supplied by Zacks Investment Research, and the number rose to about 3,700. One of the agency's executives said at the time that this freed about 20% of staff time. The question that remains after a full decade: what can a machine not take from the journalist?

What was actually automated

In that experiment the machine wrote no investigation and no analysis. It wrote earnings reports, which are near-fixed templates in which the figures and company names change. The pattern recurs: the more a text is a template and the more structured the data, the easier it is to automate. The more a job needs a question with no ready answer, a source who hesitates to speak, and a judgement about what deserves publication, the more it stays human. That is the editor's analysis, not a measured result.

The public asks the machine

Readers have changed too. The Reuters Institute's report (June 2026) finds that 10% of respondents worldwide use chatbots for news every week, against 7% the year before, and that only 1% call them their main source of news. The share rises to 17% in the youngest age group and falls to 5% in the oldest. Some 20% worldwide trust news from these chatbots, compared with 37% who trust most news most of the time. Across 27 markets, only 4% said they often or always click through to the original source, against 19% when using search. These figures are self-reported, so they may understate or overstate real behaviour.

What does the machine say about the news?

A study coordinated by the European Broadcasting Union and led by the BBC (October 2025) examined more than 3,000 answers from four assistants (ChatGPT, Copilot, Gemini and Perplexity) in 14 languages, across 22 public-service media organisations in 18 countries. 45% of the answers had at least one significant problem, 31% had serious sourcing problems, and 20% had major accuracy errors such as invented details or outdated information. The study said results had improved slightly since an earlier one in February 2025, but errors remained frequent.

What is left for the journalist?

What cannot be reduced to a template: being on the ground, checking a story before publishing it, and answering for what is published. It is a trust before it is a profession, since readers rely on journalists for what they know about people and events.

A qualitative Reuters Institute study (2024) of 45 users in Mexico, the United Kingdom and the United States found that most saw human oversight as essential to the use of AI in journalism, and cared more about disclosure when AI produces new content than when it does behind-the-scenes tasks. The sample is small and does not represent the public as a whole.

Three commitments

1. For the news organisation: tell readers where the machine was used, and leave the final decision with a person who bears responsibility.
2. For the journalist: treat the machine's output as a draft to be checked against the source, not a finished story.
3. For the reader: click through to the source, and verify any story that matters to you before passing it on; verification is part of Islamic manners and of our social ethics.

The question no machine answers remains: why does this story matter at all, and to whom?