Imagine this: You’re editing a manuscript, getting into your groove. At first, everything seems fine but soon your spidey senses are tingling. Something feels off, though you’re not sure what. The manuscript seems a bit flat, maybe a bit clichéd.
You read closer. The phrasing is rather generic and there’s no real context. It doesn’t sound like your author, either.
You start to wonder, “Did AI write this manuscript?”
While there may be specific situations where an AI-written manuscript is okay (e.g., internal documentation), usually the unspoken agreement with a published work is that a human being wrote it. That turns editors throughout the publishing process into AI detectors.
Where do you begin? How do you spot AI writing? (The presence of em dashes is not sufficient.) And, most important, how do you talk to your author about it?
Talk About AI at the Start
At this year’s Editors Canada conference, editor and educator Christina M. Frey led a session on AI and ethics. The best time to handle undisclosed AI use, said Frey, is before you start editing. She recommended that at the start of the project editors share their own AI policy (here’s our AI policy) and ask clients to share theirs, if they have one. Consider reminding your client of your policy when you return the edits to them, she said, especially if you’ll do subsequent editing rounds.
To that I’d add that if you can share your policy in a neutral way, you’ll make it easier for the client to talk about whether they used AI or have a policy. In an email, you might try something like:
I’ve attached my AI policy for your review. Please let me know if there’s any AI use we should discuss. And if you have your own policy, could you share it? It will help me to do my best editing for you.
Understandably, there’s a lot of judgement surrounding AI use and your client may feel ashamed or defensive about their usage. Creating a neutral space for them to talk about AI will help them to be honest with you.
Focus on the Writing, Not the Writer
Frey also suggested that when we have difficult editorial conversations, we should focus on the writing, not the writer. This is a principle that should be familiar to all editors, and one we can apply to suspected AI writing.
Remember: The editor’s job is to serve the text and the reader. Staying anchored to that will help keep the conversation professional and productive. It can also help keep the relationship with your author intact, even when things get uncomfortable.
Don’t Accuse
If an author didn’t disclose AI use but you suspect they should have, you might be tempted to call them out on it. But editors must tread carefully with suspected AI use.
I’ve seen my share of public accusations that an author has used AI to write their piece. Such accusations never go well. Like accusations of plagiarism, accusations of AI writing can end careers.
AI detection tools are far from reliable, frequently returning false positives or questionable positives. It has to do with how these tools work, measuring things like “perplexity” and “burstiness,” which relate to a writing style that is clear, formal, and structured. If someone writes like that, then their manuscripts are likely to be flagged as written by AI.
Even if your suspicion is correct, making accusations tends to backfire. The situation will quickly become adversarial and put your author on the defense. Once you accuse someone, they’re less likely to honestly share their process with you. Instead of educating your author on appropriate AI use, you may find you lose a client.
Be Curious Instead
Rather than confronting a client with what you think happened, ask open questions about their process. Try:
- Can you tell me about your process for this piece?
- What tools did you use?
- Could you share your prompts and outputs with me?
These questions invite transparency without triggering defensiveness. They signal that you’re engaged with the work and interested in collaboration, not in building a case against anyone.
And here’s a tip from Frey: Have these conversations over the phone or on a video call. Hard conversations often go better this way. With written communications, it’s too easy to misread tone and to distance yourself from the other person. A phone or video call makes you both more human and makes tone easier to distinguish.
No, Really, Don’t Accuse
If you reach a point where you genuinely believe undisclosed AI use has crossed into a breach of your agreement and you’re considering saying so explicitly, collect proof first. Not a hunch, not a detector score. Proof.
Proof is hard to come by:
- Detection scores are problematic.
- Fabricated citations are proof of fabrication, not of who fabricated them.
- A mix of writing styles might indicate AI use or plagiarism or something else.
- Generic phrasing, emotionless writing, and a lack of context might be AI or simply bad writing.
Unless the author tells you they used AI or you see their prompts and the AI’s output, you can’t truly be sure they used AI. And if you’re not truly sure, defaulting to curiosity is your best bet.
As editors, our job isn’t to be the AI police. It’s to notice when something seems off in the manuscript, no matter what the reason. We have to determine how it feels off and how to address that. Sometimes that means having hard conversations. When we remain curious rather than accusatory, we can work with our authors to make their writing truly theirs and the best it can be.
Want to read more about AI and editing? Explore all of our AI content on The Writing Resource.
