The Quackery Gazette

Mince a shallot by hand, mince one in a food processor, and set them side by side. Most people cannot tell you which is which. Nearly everyone can tell you that something is off. The processor's blade does not part the cell walls so much as burst them, and what comes out is wetter, sharper, and faintly acrid — a difference with no name and no address, which you taste in the finished dish without ever locating it.

We are opening with that because the same is true of writing, and because we are not going to pretend otherwise. There is no test we can hand you. There is only telling you, so here is the telling.

This policy covers generative artificial intelligence specifically: large language models, image and video generators, voice synthesis. It does not cover spell-check, search, layout software, or the statistical machinery inside a camera. The line we draw is not between machines and no machines. It is between a machine that formats what a person wrote and a machine that produces what a person did not.

Issue I of this paper was written and designed with substantial use of a large language model — Claude, made by Anthropic. Drafting, editing, page layout, and typesetting all involved it. What did not: the decisions about what the paper is for, what runs, what gets cut, and whether a thing is true. Those are a person's, and so is the responsibility for them.

We do not invent quotations. No words appear inside quotation marks attributed to a real, named person unless that person said them and we can show where. This holds in satire exactly as firmly as it holds in reporting. It is the one rule here with no exception.

Every photograph in this paper was taken by a person and shows something that was there. Where an image is a drawing, an engraving, or a diagram, it looks like one.

Most of the bylines in this paper belong to animals. A rooster covers foreign policy. A rope covers education. That is satire, and it is not a claim about who typed. A fictional byline does not mean a machine wrote the piece, and a human byline does not mean one didn't. The two questions are unrelated and we would rather you had both answers.

We cannot detect generative text in anything submitted to us. Neither can anyone selling a tool that says otherwise. If you send us work, we take your word for it. If we find out we were wrong, that runs as a correction, not as a quiet deletion.

This policy will be longer than it needs to be for as long as the practice is unsettled. When there is a norm worth pointing at, we will point at it instead.

This is not a consent form and there is nothing here to accept. It is a list of what happens.

This site is hosted by Vercel. Every time you load a page, Vercel's servers record your IP address, the time, which page you asked for, your browser and operating system, and the page you came from if you arrived by a link. We did not choose this and cannot switch it off. We do not read those logs. They exist anyway.

The typefaces on this page load from Google's servers rather than ours. That means Google is told your IP address and browser details every time you open the site, whether or not you have any relationship with Google and whether or not you are signed into anything. This is a thing we could fix by hosting the fonts ourselves, and have not yet.

We run no analytics. No Google Analytics, no tracking pixel, no heatmap, no session recorder. We do not know how many of you there are, which pages you read, or how long you stayed. This is not modesty about the numbers. We do not have them.

This site sets no cookies of its own.

If you write to the subscribe address, your message arrives in a Gmail account. Google holds it, scans it, and applies its own policies, over which we have no control and into which we have no visibility.

A setting on our hosting account permits our host to share this site's code and configuration with AI model providers for training. That concerns the site's source rather than anything about you. We are listing it because a policy that promises to be complete and then omits the inconvenient item is worth nothing.

The domain is registered to a person rather than a company. Registrar privacy hides the registrant's name and address from public lookups. It does not hide them from the registrar, from ICANN, or from anyone holding a subpoena.

The paper itself collects nothing. There is no way for us to know that you picked one up, no way to know whether you read it, and no way to reach you afterward. It is the least surveilled thing we make. That is not incidental to why the writing is in the paper and not on this page.

If any of the above stops being true, this page changes and the change is dated.