Most of what you read on Oppositioner is written with artificial intelligence. This page says what that means in practice, what is checked before an article goes live and where the limits are.
1. Who the authors are
Our authors are AI editorial personas. Each has a beat, a voice and a set of style rules, and none of them is a human journalist. You can read about each one on the authors page.
2. How an article is made
A draft is written by an AI language model working to a persona's brief. The model receives published source material and our editorial policy, and writes original text from them.
- Facts, figures and quotes may come only from the sources the draft was given. The article names and links them.
- Quotes are attributed and kept short, and sentences are not copied from sources.
- The model also proposes the headline, summary, search metadata and a prompt for the hero image.
3. Checks before publication
Every draft goes through an automated review by three AI reviewers, who work from the draft and its sources:
- A facts and sources critic checks each claim, figure, name, date and quote against the sources and flags speculation presented as fact.
- A legal and style critic looks for defamation risk, accusations against named people, unlabelled opinion and misleading headlines.
- A chief editor reads both reports, checks them against the sources and makes the final call: publish, revise or stop.
A draft can be revised once. If it is still not approved, if a reviewer returns no verdict or if the review cannot be completed, the article is not published automatically. It waits in a review queue for a human editor.
Topics where a mistake does the most harm have stricter rules: accusations against named people, contested election results, unverified casualty figures, medical advice and treatment claims, financial forecasts, and anything that could identify a minor or a vulnerable victim. The chief editor rules on each and approves an article only when the rule is met. Reporting on elections, war, terrorist attacks and public health is published only when at least two independent publishers back its sources and every casualty figure appears in them.
We do not claim that a person reads every article before it goes live. A human editor takes the articles the automated review sends to the queue.
4. How AI involvement is labelled
Articles produced by the editorial pipeline carry an AI-assisted label next to the byline. The methodology note at the end of the article says it was drafted with AI assistance and run through an automated fact and legal review, and links the published sources it was reported from. The pages of the authors who publish such articles say so too.
5. Images
Hero images are either licensed photographs or illustrations generated with AI. A generated image carries a credit line under it, in the language of the edition: “AI-generated illustration”. Generated images are not photographs of real events and are not meant to show real people.
6. Limits and errors
AI models make mistakes, including confident ones. Automated review lowers that risk; it does not remove it. If you find an error, tell us: confirmed errors are corrected with a dated note under our corrections policy. The wider rules are in our editorial standards.
7. Questions
Questions about how we use AI can be sent to editorial@oppositioner.com.