aceAI-reporters as the future of journalism? Investigating recipients’ credibility evaluations of AI-authored news
AI-authored news articles are becoming increasingly relevant, but little is known about how recipients evaluate news credibility attributed to an AI (artificial intelligence) author compared to a human journalist author. Referring to the machine heuristic ( Sundar, 2008 ) and Expectancy Violations Theory ( Burgoon, 2015 ), this study compared articles with AI-author labels to articles with human-author labels and without author labels. It further included AI trust and attitudinal congruence as relevant aspects t…
In brief
Researchers ran an online between-subjects experiment in which 844 German respondents read an article on cannabis legalization. The article was framed as pro or contra and carried an AI author label, a human author label, or no label. Credibility was then compared across these conditions.
The finding matters for news organizations deciding whether to disclose AI authorship, since the label lowered perceived source credibility relative to a human byline. The effect did not appear against unlabelled articles and did not extend to message credibility, so the practical consequences remain unclear.
Main points
- The study used a 2 (pro vs contra arguments) x 3 (AI, human, or no author label) between-subjects design.
- AI-labelled sources were rated less credible than human-labelled sources but not significantly differently from unlabelled ones.
- Author labels did not change how credible the message itself was judged to be.
- Trust in AI directly affected both message and source credibility but did not interact with author labels as expected.
- Attitudinal congruence with the article strongly shaped credibility evaluations.
The problem
Among 844 German respondents, news sources labelled as AI-authored were judged less credible than sources labelled as human journalists, which may disadvantage newsrooms that disclose AI authorship.
The rundown
The experiment drew on the machine heuristic and Expectancy Violations Theory to explain how audiences might react to an AI author. It tested these ideas using a news article about cannabis legalization, with attitudinal congruence controlled in a MANCOVA.
Source credibility was lower for AI-labelled articles than for human-labelled ones, but the comparison with articles lacking any author label showed no significant difference. Message credibility was unaffected by the author label.
Individual attitudes mattered more than labels in some respects: people's trust in AI affected credibility judgments directly, and agreement with the article's position strongly influenced them. The authors discuss theoretical and practical implications.
Sources
- Peer-reviewedJournalism2025-09-25
Blue Planet III features cloudier footage to prove authenticity amid AI boom
The Guardian
Made for $1.5M, China's first AI film, Sanxingdui: Future Past, finds no takers with just $10K gross in pre-sales
Hindustan Times
Anthropic, music publishers square off on ‘fair use’ defence in AI copyright case
The Straits Times
The debate