My policy on AI is simple: let’s talk about it.

AI has ushered us into a strange moment when it comes to writing and publishing (among other things). Some writers eschew AI as much as possible, others use it poorly, while still others are very sophisticated users.

The largest issue, in my experience, lies with the group that is curious but doesn’t know how to use AI well. Many aspiring authors use AI in ways that lead to formulaic writing and that can risk their ownership of the work’s copyright. (If you want to learn more about the copyright issue, check out industry analyst Jane Friedman’s webpage about this, especially the section titled “Copyright for AI-assisted and AI-generated works.”)

On top of all that, AI capabilities are changing incredibly fast, so it’s best that we talk about it. Here’s my starting point and my request to you:

• I won’t use AI in our process in any way without talking it through with you first.

  If you never want AI anywhere near your book project, that’s perfectly fine with me. Most of the books I’ve edited or collaboratively written, including all 85 of them at Shambhala Publications, were done with precisely zero LLM involvement, and I’m happy to work that way.

  That said, I have a Claude Pro account and am able to work much more quickly on certain tasks with its help, while in other areas I find it’s detrimental to the process. For example, Claude is exponentially faster than me at spotting large-scale issues of flow in a manuscript (try as I might, I can’t read tens of thousands of words per minute). Whereas its work as a line editor makes the prose sound worse than if I do it without its input—it seems to nudge every piece of writing toward sounding promotional, revelatory, and self-summarizing.

• If you’ve used AI on your manuscript, let me know how.

  In most cases where clients send me AI-generated or AI-modified text, I’m able to tell based on certain types of sentence structure, rhythm, and word choice favored by LLMs. And in many cases, due to AI’s sycophantic tendencies, writers have prompted it to edit and may not register that it’s done something closer to a rewrite.

  Once we’re on the same page about how you’ve used it or how you intend to use it, we can make a plan that fits with your publishing goals. That will help me help you to avoid unwanted surprises down the line (for example, it’s increasingly common for traditional publishers to run manuscripts through an AI checker such as Pangram).

Lastly, you can read more of my thinking about AI on my Substack. For example, check out my post “Implications of the Face Hugging Incident,” and subscribe to see more.