AI detector puts opinion pages under pressure
An analysis finds many opinion pieces written wholly or partly by AI, but a detector result is not conclusive evidence.
A study by AI Report identified AI use in dozens of submitted opinion pieces at five Dutch newspapers. The figures present publishers with an editorial dilemma: transparency is needed, but automated detection cannot independently establish who wrote a text.
AI Report examined 252 opinion pieces published between 27 July and 27 August 2026 by de Volkskrant, NRC, Trouw, Het Parool and het Financieele Dagblad. According to the analysis, 49 pieces were labelled fully AI-generated and 57 partially AI-generated. At de Volkskrant, the same method identified 32 of 111 pieces with a full AI label.
The analysis used Pangram, a commercial tool that compares text patterns with AI-generated examples. Such a result is an indication, not a declaration of authorship. The method can mark human text as AI-generated and can miss AI use. The results therefore do not automatically mean that an author made no substantive contribution or that a piece was generated without permission.
The figures do have journalistic significance because opinion pieces are presented as contributions from identifiable authors. Readers expect not only a point of view, but also that the author stands behind the wording and takes responsibility for the text. When a newspaper allows AI assistance for spelling or editing, that is different from a fully generated piece published under someone’s name.
At de Volkskrant, editor-in-chief Pieter Klok was reported to have said he was surprised by the scale. The newspaper told staff that AI may not be used to write submitted contributions, except as a spelling aid. Het Financieele Dagblad responded less negatively, according to the same reports, but did call the outcome uncomfortable.
The debate is therefore not only about detection technology. Newspapers must decide what assistance is acceptable, whether readers should be informed about it and who checks that submitted work complies with the rules. A ban without oversight is difficult to enforce; full acceptance without disclosure may damage trust in authorship and editorial selection.
The subject is particularly relevant to De Vector because our articles are written by AI and this is made clear in advance. That model does not resolve every question, but it makes the production method visible. At newspapers that present human authors, the threshold for disclosure is different. The current analysis does not prove deception in any individual piece, but it makes clear that editorial teams can no longer treat their policy as a theoretical issue.
One story, several perspectives
What is established
- AI Report analysed 252 opinion pieces using an AI detector.
- The analysis marked 49 pieces as fully and 57 as partially AI-generated.
- A detector result is not the same as proof of authorship or permission.
Left
Arguments Newspapers have a public responsibility and must inform readers honestly about substantial AI involvement. Submitted contributions must not become a cheap production method in which authorship and power relations remain invisible.
Values Transparency, labour rights, editorial responsibility and equality in public debate.
Consequences Stricter disclosure rules could strengthen credibility, but would require additional editorial time and could exclude some contributions.
Centre
Arguments The outcome calls for clear, proportionate rules: distinguishing between spelling assistance, editing and full generation, with human oversight and a complaints procedure. Detectors can flag possible issues, but must not act as judge.
Values Care, reliability, practicality and the right to be heard by both sides.
Consequences A transparent policy can preserve trust without banning legitimate use of tools, but requires consistent application to all authors.
Right
Arguments An author who publishes a viewpoint under their own name is responsible for the text, regardless of the tools used. Newspapers must therefore be able to take firm action against concealed outsourcing, because otherwise readers will not know whom they are addressing.
Values Personal responsibility, enforcement of standards, ownership rights and trust in institutions.
Consequences Strict rules can protect the quality and recognisability of opinion pieces, but must not lead to unfounded accusations based solely on an algorithmic score.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The figures and period examined were checked directly with AI Report and an independent second source. The limitations of AI detection are presented as a methodological warning, not as an absolute factual claim about every individual piece.
- confirmed AI Report examined 252 opinion pieces from five Dutch newspapers. — Mentioned in AI Report’s methodology description. source
- confirmed AI Report labelled 49 pieces as fully and 57 as partially AI-generated. — The analysis explicitly gives both figures. source
- confirmed At de Volkskrant, 32 of 111 pieces received a full AI label. — Mentioned by AI Report. source
- confirmed A detector result does not independently prove that a specific text was written by AI. — A methodological conclusion consistent with the limitations of detection research; the text does not present it as the result of a single individual test. source
Editor's note
The figures are the results of a detector analysis and not a determination about each article. The text makes that distinction explicit and does not use a detector result as evidence of individual deception.Sources
More on this in Dutch media
- AD — „kunstmatige intelligentie”
- de Volkskrant — „kunstmatige intelligentie”
- RTL Nieuws — „kunstmatige intelligentie”