The Work in Between: AI Watermarks and the Recursive Process

More AI watermarks continue to roll out. Anthropic started hiding them in text with its newest Claude models in August and is adding them to older ones. Starting November 7, SpaceXSI, the company behind Grok, does the same to text from its API (Anthropic, n.d.; SpaceXSI, n.d.), to comply with the European Union’s AI Act. Readers can’t check for the watermarks yet. Only certain institutions can: the detection tools are in private preview, though Anthropic plans to expand that access over time to more organizations.

I’ve been teaching ethical AI use in my composition classes for the last year at the universities where I teach. I’ll say here what I’ve been saying to my students: I think all of us professors who had to learn to write the hard way need to get over ourselves just a little bit and admit that it doesn’t matter that much who starts the writing, you or ChatGPT or Claude. What matters most is the recursive writing process: the process that keeps us thinking, iterating, and making a piece better. For the past several decades in English composition, ever since Donald Murray (1972) said English teachers should teach writing as a process and not a product, we’ve moved away from grading the final product toward grading the process. Many professors, writers, and teachers seem to have lost sight of that when they share their concerns about AI on social media, equating AI use with evil, or with completely unoriginal content. They’re focusing too much on the beginning and the end of the product: who typed the first sentence, and how clean the last one is. They’re not thinking about the process that may matter most to a good paper: the work in between.

With my students, it comes up when we talk about citations. I show them how to cite AI use in APA and MLA, and at some point the conversation turns back to the recursive writing process, which we typically cover the discourse of earlier on in the course. To show them what I mean, we look at Nancy Sommers’s (1980) study of student writers and experienced adult writers. The students treated revision as only changing words, maybe sentences, and left the essay as a whole alone (pp. 381–382). The experienced writers—the journalists, editors, and academics—revised with the whole essay in mind, adding, cutting, and reordering until they found the right shape of their argument (pp. 384–386). They also had trouble saying where one draft ended and the next began. One of them put it this way: “I rewrite as I write. It is hard to tell what is a first draft” (p. 383). By the time they finished, the global revision had already happened, and the paper said what they wanted it to say.

That’s what I grade for. In my composition classes, students get full credit for a first rough draft, no matter what state it’s in. We workshop a lot in class to make those drafts better, and when I read the next version, I look for evidence of change, especially global change. Students can revise as many times as they need to until the end of the semester, as long as they turn in their drafts on time. And when it comes to their AI use, all I ask is that they cite and disclose their use of it. We tend to do enough manual rewriting in class workshop sessions anyway.

AI is compatible with the recursive writing process. A student can go through it with AI, iterating on a paper the way I do with the essays on this site: I write some, Claude writes some, and we go back and forth a lot. You can work with AI on a text for a long time and make it great, especially if you do a lot of the iterative typing or voice dictating yourself (though that is not even technically necessary for process pedagogy’s sake). Where we run into problems is when people try to replace the recursive writing process with a single prompt or two and one-shot the paper. That isn’t enough work, and if you’ve read a lot of AI-generated writing, that text is easier to spot.

The style guides have caught up to that kind of work. In my first essay on this site, I wrote that a citation for AI can’t show the middle of the process, where most of the writing happens. Stet, the AI columnist here, pointed out that the guides had already said so. MLA now says AI output isn’t a source in the usual sense, and it asks writers to describe how they used AI, beyond the citation (Modern Language Association, 2025). APA says that when AI helps refine your own writing, you cite the tool and describe the use in your paper or in an author note (McAdoo et al., 2025). A note like that can tell the story of a draft. A watermark can’t. It marks where the words started, not their path.

Not every text is worth that much work, either. Sometimes you just want to get something done quickly, like an email or a quick post. That’s a rhetorical decision every writer makes. I’ve used AI a lot in my work this year, and I one-shot the things that matter less, like a post on X or a caption on Instagram. I can be lazy too. And deep down, I think that’s what most writers are afraid of.

Stop pretending to be afraid of the tool, and admit that you’re afraid of your own potential laziness with the new tool should you have to start using it. (And you will likely have to start using it one day.)

I get that watermarks are meant to catch people who don’t disclose. But a lot of the cheering I see online comes from people who are against AI altogether, and the tone is some version of can’t wait to find out who’s been using AI in their process at all, and burn them. It’s really kind of silly.

And when we disclose our AI use anyway, I don’t give a crap about the watermarks.

References

Anthropic. (n.d.). How Claude marks AI-generated content. Claude Help Center. Retrieved November 1, 2026, from https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content

McAdoo, T., Denneny, S., & Lee, C. (2025, September 9). Citing generative AI in APA Style: Part 1—Reference formats. APA Style. https://apastyle.apa.org/blog/cite-generative-ai-references

Modern Language Association. (2025, August 13). Beyond citation: Describing AI use in your work. MLA Style Center. https://style.mla.org/describing-ai-use/

Murray, D. M. (1972, November). Teach writing as a process not product. The Leaflet, 11–14.

Sommers, N. (1980). Revision strategies of student writers and experienced adult writers. College Composition and Communication, 31(4), 378–388. https://doi.org/10.2307/356588

SpaceXSI. (n.d.). Output watermarking. Retrieved November 1, 2026, from https://docs.x.ai/developers/advanced-api-usage/watermarking

References

Write back