Based on my years of reporting on AI and publishing, in this article I answer questions writers ask me most—about detection, disclosure, proof of authorship—and discuss where I see the most risk. Separately, I’ve addressed straightforward issues of fact, and where things stand legally, in AI and Publishing: FAQ for Writers.

The biggest questions I get right now about AI:

• How do I protect myself from an agent or publisher (or other publishing professional) using AI on my work?
• How do I protect myself from a false accusation of AI use?
• How can I prove that I wrote my work and/or how can I prove that my particular AI use is not a problem or doesn’t run afoul of agent/publisher guidelines?

Unfortunately, these questions don’t have straightforward answers.

How do writers protect themselves from publishing professionals using AI tools on their work without permission?

First things first, be sure to read any available policy from the agent, publisher or publishing professional about their AI use. Increasingly you’ll find a statement on their website, in their submission guidelines, or other documents they provide authors. If you can’t find one, ask. Generally you can trust them to keep their word, although I admit trust can be in short supply these days.

Over time, it will become difficult to find agents, editors, marketers, and publishers who don’t use AI at all. Every major publisher currently uses AI as part of normal business operations, even if it’s not used during the creative process by editors and designers. At least one Big Five house, Penguin Random House, has gone on the record as using AI detection software. My guess is that many publishers may be using it without disclosure, and large corporations don’t always know what their thousands of employees are doing. So the pragmatic and realistic answer is that you can’t protect yourself from this use—anyone with access to your work can decide to pop it into an AI tool without disclosure. Could you sue them for it? I don’t know; I see no clear legal standard for doing so unless they signed a contract that prohibits the behavior. Even if you do have legal recourse, I’m not sure it’s worth pursuing given the time, money, and effort that would be involved.

Some publishers and agents use AI systems that are closed or proprietary, meaning your work is not available beyond their own virtual desk and is not used for AI model training. Most publishing professionals care about that and don’t desire to use these tools to compromise you and your work. They’re using it to make their jobs easier. I know some writers will view that statement quite cynically, but it’s logical they’d look for efficiencies in a business that pays modestly while being time intensive. Most professionals I know want to save time and energy to focus on more desired and creative tasks, like working with authors and others in the business. (But whether they’re using AI skillfully or thoughtfully is another matter.)

Where I have the most concern: I think the temptation is high for service providers, among others, to use AI in ways that aren’t well disclosed to clients. E.g., maybe you hire a marketer or publicist to help you with a campaign, and you don’t realize a good part of it has been AI generated. Whether it’s a good plan will depend on the expertise of that marketer or publicist in using AI, evaluating its outputs, customizing it for your project, etc. Or maybe an editor uses AI to generate feedback without telling you. How would you know? Probably only by having a human conversation with them, with plenty of back-and-forth that would catch them out if they didn’t read or engage meaningfully with your work.

Any author partnering with publishers or hiring freelancers needs to be doubly diligent about doing their research, knowing what they want from the relationship, and taking steps to ensure that it is indeed a relationship, not just a transaction. AI is great at transactional work and if you just want boilerplate feedback, boilerplate campaigns, etc., ask one of the many AI tools out there for the average guidance one would receive.

How do writers protect themselves from AI companies training on their published books?

There is no realistic way to protect yourself from this. Sure, you can put a notice in your books that essentially says “no AI training.” But in the US, courts have so far found that AI training on copyrighted works is fair use, although that could change. Currently, AI companies are buying up print books, scanning them, and using them to train AI models. They’re also pursuing licensing agreements with big publishers and media companies.

If the courts find that AI training is not fair use, then you can rely on your publisher or the Authors Guild to litigate on your behalf. (They already are.) If you’re a self-published author, I would recommend signing up for one of the licensing services, like Created By Humans, that give AI companies a legal path to licensing your work. Still, if AI companies or others choose to train on your work without permission (e.g., by gaining access to pirated copies or scanning a print edition), there’s no good way to know on a case-by-case basis. And most authors don’t have the resources to file a lawsuit and will have to depend on class-action lawsuits, as they do now.

What if AI detection scores on my writing are wrong and cause me to lose a publishing opportunity?

Right now, I find people ask this question more often out of fear, not experience.

But, like everyone else, I see people using free AI detection tools, testing them with short excerpts of their own writing, and claiming false positives. They then declare the detection tool cannot be trusted under any circumstance.

It’s clear these tools aren’t 100 percent right all the time, but if used under appropriate conditions, they are a helpful tool when paired with human judgment.

When someone posts AI detection results online, look for the following information:

• What service is being used? There are many bad ones that can’t be relied on. Pangram is so far the industry leader, for good reason: studies have shown it’s reliable. Many people have been trying to say these tools aren’t reliable based on studies conducted 2–3 years ago. AI detection has advanced since then. If you’ve heard that it can’t correctly identify writing from neurodivergent writers or non-English speakers, you’re hearing about an older study.
• How much text was used? AI detection doesn’t work well on small amounts of text (fewer than a couple hundred words). For novel-length work, it’s far more accurate.

Almost every time I see a post showing a screenshot of an AI detection score, it’s a piece of clickbait trying to rile people up. Causing fear and panic is a good way to get attention these days.

What’s missing from the AI detection discussion is that, for the purposes of agents and publishers, these tools serve a real need, and it’s rarely a punitive one. Agents report that, since February of this year, about half of all submissions now involve AI use. As of today, I’d consider it part of a market analysis to know a work’s AI detection score. (Funny enough, some people who are against detection software seem to be using AI to write. Here’s an article on that.)

I see two primary uses of AI detection among agents and publishers:

• With contracted authors: Publishers use AI detection to help have conversations about submitted manuscripts, where there’s a relationship in place with the author. Educational and university publishers in particular feel it necessary to use AI detection because of the high stakes involved. It is exceedingly rare to see a public case of a contracted author getting canceled over AI use. (More on that below.) Professionals prefer to deal with these cases quietly and in conversation with the author, and put remedies in place to make the manuscript acceptable.
• With uncontracted authors submitting: Agents and publishers alike are beginning to use AI detection to help them quickly filter out work with very high scores. One agent and one editor told me once a manuscript scores above 20 or 25 percent likely AI-written, they start to hesitate, as that reflects more heavy AI assistance that could prove problematic.

I think many authors envision that agents and publishers have a zero-tolerance policy where they reject anyone for even a 1% AI score. No reasonable agent or publisher is doing that—and it’s not realistic to expect today’s writers to shun every single AI tool available to them, especially when major publishers are using such tools. That said, agents and publishers who are most avidly against AI are not likely using AI detection in the first place and don’t trust it. Those who are more accepting of AI use will have some tolerance and see it as an inevitability.

What about Jerry Falade? Agents who considered his submission (but did not offer to represent him), told me the manuscript had all the hallmarks of AI writing from the first paragraph. They did not need AI detection to tell them that. Why other agents didn’t see that for themselves is an interesting question, and there are theories, but that discussion is for another day. Suffice it to say, what sells in today’s market (both traditional and indie) includes both human-written work and AI-assisted work. Why Falade’s agent made any comment to the media about the situation is a mystery to me, but that agent didn’t use AI detection either. Some media outlets did.

What about Shy Girl? Readers believed her work had the hallmarks of AI writing before AI detection was used by bystanders. The publisher pulled the book so quickly when media scrutiny arrived, it feels like they were scared and caught unprepared. I have to wonder if today they would do the same thing. The environment is changing fast.

What about Daggermouth? Some academic researchers surfaced this book in a large-scale study using AI detection, and The Atlantic picked up on it and published the detection score. The author is being supported by her publisher and even signed another deal with them after The Atlantic’s article was published. This just goes to prove my earlier point: I don’t see agents and publishers on witch hunts, eager to reject anything touched by AI. The very online crowd who are trying to publicly shame authors for using AI, those are the ones to worry about. Publishers using AI detection punitively and/or to deliberately cause a public scandal are not the sort of people I consider professionals.

One agent confided in me that it’s common for writers to complain to her that, when doing their own private testing, their writing has been flagged positively for AI, even when they didn’t use it. After gentle questioning, she comes to find they did in fact use AI, sometimes via Grammarly, ProWritingAid, and other tools that have AI capabilities. Sometimes writers aren’t being honest with themselves about what AI writing is or they’re just being willfully obtuse. (More discussion about software below.)

What concerns me: I think some agents and publishers are rejecting work and citing AI as a reason when it’s not the reason, to shut down conversation and get rid of people. Who knows why. But these days, it’s something of a casual insult to tell someone “I think you used AI.” As I see it, the risk of an overly casual assumption that you used AI is greater with smaller outlets, less experienced outlets, and less professional outlets. (This would describe countless literary journals and online publications.) If I receive a submission that’s unacceptable for any reason (and that includes heavy reliance on AI), I reject it on editorial grounds. I don’t accuse the writer of using AI.

My submission was rejected because I used AI (and I admit to using AI). Now what?

Some agents and publishers ask you to disclose upfront if you used AI and may reject you if so, based on your disclosure. Frustratingly, it may be unclear what they mean by AI use. Does it include research? Spellcheck? Grammarly? Brainstorming? There’s no way to know unless they have a clear policy.

Regardless, there are any number of publishers and agents who have strict anti-AI policies and will tell you not to touch it at all. They might even say your work will be ineligible for copyright if you do (not true unless your work appears obviously AI written, but a good way to scare writers). As time goes on, I would consider these policies to be comparable to policies that rule out certain genres or categories of work, or certain themes or certain issues or events. If a prospective agent or publisher for your work announces, “No AI use ever, without exceptions,” but you regularly use AI for aspects of your creative process or research, then don’t submit to them. I would not try to convince them or fool them. If the latter, you’re destined to have a bad relationship. It is very hard to conceal consistent AI use over time.

I didn’t use AI in my work. How do I prove it?

I don’t think you can prove it unless you’re willing to rely in part on tools like Verify My Writing, which use the same technology as AI detection. The Authors Guild Human Certification does not prove anything, nor does any other certification method currently on the market that doesn’t use AI detection.

So far I have not found any universally agreed-upon method, system or score that’s recognized by every agent, editor, or publisher, that proves your work is human authored. You may find individuals laying out procedures or policies for how they determine something is human authored. But that doesn’t mean it will be shared by others.

I am regularly pitched by authors, tech people, and service providers who all think they have found “the way” to prove human authorship. You haven’t heard of them because none of them have achieved mainstream adoption. When one does, I promise you will hear it from me first. (If anyone tries to sell you on a tool or software like this, ask them what agency or publisher is using it as proof.)

The good news: It’s rare that agents or publishers demand proof of human authorship that you can’t provide. Right now authors’ desires to have proof seem to be fueled by anxiety over possible future demands or reader accusations. Unfortunately, just like Goodreads review bombing, you can’t stop readers from doing obnoxious and awful things. It may become part of what all authors experience. But as for agents and publishers, I simply don’t see anything yet that indicates they’re looking for any special proof of process that you wouldn’t normally be able to provide.

I am worried that using AI tools in Grammarly, ProWritingAid, Word, or XYZ software will make my work unacceptable to agents and publishers.

This worry is largely misplaced unless you are using these tools to generate or rewrite significant portions of your manuscript. I would not, under any circumstance, load up my entire manuscript into an AI-enabled editor and let it rip without reviewing every change. Would you let a freelance editor make countless changes without having them tracked in Word? Don’t let AI do it either. Fortunately, AI tools have become sophisticated in this regard—e.g., you can install Claude for Word and have it track changes for you.

Final thoughts

I believe writers are overly anxious about these issues, mostly because of misinformation online and some outlying, headline-grabbing cases, not because of anything they’ve experienced personally. Just like other issues in publishing where writers have little or no control, this can all be a grand distraction from getting your writing done.

That said, much as I try to bring clarity and calm to discussions about AI, none of these questions have good answers right now. It’s a world of uncertainty and frankly you just have to get on with your writing and your business as best you can.

Next week, I’m presenting a virtual talk at Goucher College, AI and Publishing: Where Things Stand, on Sept. 17 at 12 p.m. EDT. Free and open to anyone.