There is exactly one category of image on this site that does not carry the words AI-GENERATED IMAGE — OBVIOUSLY FAKE, and it is the cartoons, because a person drew those and stamping them would be its own small untruth. Everything else is generated, and everything else is labeled.
That decision looks like caution. It is not. It is the product.
The generated pictures here are the most persuasive thing on the page. They are more persuasive than the headlines, which are absurd on purpose, and far more persuasive than the articles, which are made of qualified sentences and citations. A photograph does not argue. It arrives already believed, and the work of doubting it happens afterward, if it happens at all. That asymmetry is the entire reason synthetic imagery is a problem worth caring about, and it does not go away because the person publishing it had good intentions.
A disclosure that only survives on the page where you published it is not a disclosure. It is a footnote waiting to be cropped off.
So we made three commitments about it, and the interesting one is the third.
The first is the caption. Every generated image carries the label in plain type directly beneath it, not in a tooltip, not on hover, not in a modal a reader has to open. The second is the alt text. Every generated image begins its alternative text with the words AI-generated satirical image and ends it by saying it is not a real photograph — because a reader using a screen reader is entitled to exactly the disclosure a sighted reader gets, and stripping it out would mean the caption was decoration for people who could see it.
The third is the one that took an argument to settle. The label travels. It is not applied by the page template; it is attached to the image record itself, so the same words follow the picture into the card on the front page, the thumbnail in the archive, the social preview, and the print edition. The reason is that a disclosure which only exists on the article page is not a disclosure — it is a footnote waiting to be cropped off by the first person who screenshots the image and posts it somewhere we do not control.
We cannot stop that person. Nobody can. What we can do is make sure the crop has to be deliberate: that removing the label is an act, rather than a side effect of the picture leaving our page.
None of this makes the images honest. They are still fabrications, they still work on you, and a reader in a hurry will still take the impression and skip the words underneath. We know that, and we are not going to claim the caption cancels the picture.
What the caption does is make the transaction legible. You are being shown something that did not happen, by people who are telling you it did not happen, in a place that keeps a numbered record of every such thing it has ever shown you. Whether that is worth doing at all is a fair argument, and it is one we are having in public, in this section, under a signature.
We would rather have it here than not have it.