Photo by Google DeepMind

Today’s guest post is by editor Mary Kole.

Having worked on thousands of manuscripts as a freelance editor, I’ve developed a pretty reliable instinct for what makes a project difficult to edit. That said, I have never priced my services according to a manuscript’s overall perceived “quality” or the level of “effort” required to offer feedback (as these judgments are extremely subjective).

Recently, AI-generated writing has entered my inbox, and this has made me rethink what editors are being asked to do, how, and how to charge for operating in this new world.

Earlier this year, I unknowingly accepted two AI-assisted novels back-to-back. Why? Well, I didn’t ask … and the clients didn’t disclose their chatbot use. Now, this article isn’t an ethical argument about AI. Writers will make their own decisions about where and whether these tools belong in their creative process, as will the industry at large. This is, however, an article about how editors should protect themselves accordingly from AI “scope creep” with new contracts and even fee schedules.

The two projects in question weren’t bad by any means. At first glance, both looked very polished. But the more I read, the more I realized I wasn’t performing the same kind of editorial work I usually do. My goal is always to understand a client’s intentions, then offer feedback and craft teaching to help them execute their vision in the most marketable way possible, depending on their overall publishing goals.

Instead of having an editorial conversation with a human creator in these cases, though, I was trying to find the humanity in work generated by AI, which added a new layer of nuance and complexity. And I was still being paid the same, though the rules had changed.

Editing versus humanizing or outright revision

Most editors price services around a fairly consistent assumption: a client has written a manuscript, and they have approached a professional with the belief that it can be improved. But inherent in this assumption is the idea that the underlying material belongs to the author. AI-assisted manuscripts complicate this premise.

Josh Bernoff, who recently proposed AI guidelines for ghostwriters and editors, says:

Authors think the job of writing is to create a logical, neat, error-free manuscript. In fact, the job is to express your views cogently in a way that is unique to you. AI can’t do that. AI-generated manuscripts are clean, free of grammatical errors, and smooth and easy to read. They are also soulless. A good editor attempts to allow the author’s voice and ideas to shine through by fixing writing problems. In this case, there is no author’s voice. This makes it very hard for the editor to do the job—we have no evidence of what the author truly believes or how they express themselves.

The two projects I was asked to simply “edit” took me far longer because I was also forced to make hundreds of creative decisions that ordinarily belong to the originating writer. In essence, the clients who’d hired me had outsourced their creative expression first to AI, and then to me, the human implicitly tasked with bridging the gap. Instead of simply evaluating execution, I found myself making authorial decisions: reshaping emotional arcs, deepening characterization, stripping back thematic overstatements, and rebuilding character interiority so it sounded like a distinct voice rather than a statistically probable Everyman.

This is authorship work, not editorial. Messy, human first drafts from organic brains with something to say frequently reveal a writer’s instincts before they reveal their underlying grasp of craft. The bot-written manuscripts I encountered actually inverted that equation. The craft was polished … but the humanity was difficult to find.

Bernoff has responded by treating AI-assisted manuscripts as their own category of work, which approaches ghost revision or outright ghostwriting. He says:

It’s become clear to me that ‘editing’ AI generated text sits in a middle ground between ghostwriting and ordinary editing. It is easier than ghostwriting, since the source material is already there and neatly organized. It is harder than normal editing since it requires a complete rewrite.

Editors who also ghostwrite or ghost revise will immediately recognize the difference. Stepping into a writer’s shoes entails a different investment of engagement, responsibility, and time. As such, Bernoff suggests editors charge a 50% premium for editing AI-generated material, and argues:

As a human editor, I cannot concentrate for as long on the mind-numbing soulless prose generated by AI. Worse yet, I need to rewrite every word of AI-generated text to ensure that the result is eligible for copyright. That’s at least 50% more work than editing a human-generated manuscript. It’s also a signal to clients that there is a cost to their cutting corners with AI. If everyone in the editorial business charged 50% extra like I do, clients would rapidly realize that they’re not going to save time and money by substituting AI for their own writing.

I wholeheartedly agree and have included this premium in my client agreement, though I haven’t actually had to invoke the upcharge yet. I have no doubt I will.

Questions of disclosure and detection

After enacting my own AI-generated writing policies, I’m less interested in whether writers are using AI and more interested in whether they are forthcoming about it. Disclosure isn’t about catching someone “cheating” or playing AI Police. It’s about accurately defining the project before editorial work begins. I don’t think most clients are actively trying to deceive editors, literary agents, or publishers—in fact, I suspect many genuinely believe that generative AI use has made their manuscript “better.”

But Bernoff puts it bluntly when asked whether editors should require disclosure about AI use: “Any client who’s been paying attention knows that they’ve created a problem for the editor in helping to craft a publishable manuscript.” For those still confused about the broad principles, Bernoff points them to the Gotham Ghostwriters AI Guidelines for Ghostwriting, which he authored.

Disclosure of AI-generated writing in a manuscript should change the scope of the editorial engagement. If I know that substantial portions of a project are non-human-created before I send a proposal, I can set expectations and account for my time accordingly.

Of course, disclosure only works if clients are willing to disclose. This naturally raises another question: Should editors attempt to verify AI use themselves? Well, the current generation of tools at our disposal remains controversial.

Derek Newton, founder of Verify My Writing, an AI-detection tool (full disclosure: I have used this and believe it’s very effective), acknowledges that not all software is trustworthy. “Most AI detection services are scams … But there are three or four really good technology products that work.” He also adds, about his firm, “We think of detection kind of upside-down—not as catching AI, but as recognizing and rewarding human work and storytelling.”

Whether editors choose to use detection software is almost beside the point. Even Newton is careful about what these tools, including his, actually measure. A detector can estimate whether text is likely to have been generated by AI. It cannot tell you whether the writing succeeds creatively. He adds: “AI detection can tell if a robot painted a painting. It cannot tell if the painting is any good.”

This brings me back to how editors interface with projects. In our role, we’re already evaluating manuscripts. We don’t need software to tell us whether a scene feels emotionally flat or whether a reconciliation between rivals feels earned. What’s left is for us to enact clearer workflows and boundaries around our time for dealing with suspected, assumed, or proven projects in this space, no matter how we arrive at this conclusion.

Policy, not judgment

The new AI addendum in my client agreement isn’t a moral statement but a scope-of-work one. It requires disclosure, defines “AI-Generated Content,” and explains why AI-assisted manuscripts require additional time and a different pricing structure. It also clarifies where my editorial responsibility begins and ends.

Whether or not editors adopt a similar framework, I think our industry needs to rethink how we estimate projects, define deliverables, discuss revision, and approach contractual liability. Bernoff’s advice is practical: “Protect yourself legally … write into your contract that you are not responsible for factual errors, AI-generated uncopyrightable text in the source material, or inaccurate quotes … And block off more time.”

Even in this new landscape—especially in this new landscape—I also believe human editors are becoming more important than ever. The value of the editorial profession has never rested on our ability to identify weak grammar. Spell check and Grammarly can help you there. It has instead hinged on our ability to recognize authentic communication … and help writers achieve more of it.

Bernoff has this to say:

Our job is to help human writers express themselves in a unique and powerful way. That’s even more important in a world flooded by soulless inhuman slop. We are on the front lines of the battle to preserve the unique qualities of the human voice, and like all front-line soldiers, our job is crucially important.

My new clauses don’t keep AI out of my business. They attempt to ensure that my pricing and policies accurately reflect the effort needed to do my job to the same standard I’ve always applied to human-created work. For your consideration and use, I’m sharing the language I developed and added, using Bernoff’s original blog post as a springboard. (This phrasing retains some conventions specific to my client agreement. You are welcome to adapt it to your needs and style.)