It’s not all about identifying AI tells. You have to know what (and what not) to do with them.
A while ago, I wrote a blog post about AI tells. Since then, many of the tells I mentioned in that post have become common knowledge. I’ve thought about writing a sequel to that post, but first, I’ve put together this list of things you should do when asked to edit AI-assisted text — and some things you should never, ever do.
The big one first: DON’T accuse your author of using AI. Ever. Even if it’s patently obvious. Don’t even point out that some readers may decide it’s AI based on the writing. It doesn’t do any good, and authors often say they’re being unfairly accused. Besides, your job is to edit the text, not speculate about its origin. Concentrate on making it better.
Another don’t: DON’T edit every single AI tell you find. Remember that humans write such things too, all the time. That’s where AI learned them. Seeing them once or twice isn’t going to make a reader suspicious. It’s when they get repetitive and overused that the reader may start to go “Hmm,” and that’s when your judicious edits become crucial.
One final DON’T: Don’t cut things just because they’re part of AI tells. You may have a series of three items that you feel should be cut down to two. But if you’re unsure whether any of those three is important or relevant to the topic (as can be the case when editing certain technical documents), err on the side of leaving it. You can always flag repeated tells as necessary (see below).
DO investigate text that feels irrelevant or unfamiliar. If you come across a phrase that seems unnatural, do your due diligence: Use a search engine to see whether it’s a term you simply haven’t encountered before. If you come up empty, ask the writer to review.
DO think about whether cutting something might eliminate intended meaning or tone. For example, one of my biggest reflexes is to cut the “It isn’t X” before “It’s Y.” It’s often appropriate to do so for conciseness, but make sure you’ve thought about it first: Is it important to mention that it isn’t X? Do people often mistake it for being X? Did your writer previously allude to it being X? If the answer is “yes” for any of those questions, X should stay.
DO flag repeated AI tells for the author, but not because they’re tells. The real issue with tells is that they are becoming cliché. When we say we dislike AI writing, we’re really reacting to language patterns we’ve seen too often. Maybe your query to the author looks like “Consider varying language” or “Consider varying number of items in lists throughout.” You know your author better than anyone else, so make sure your query is constructive and helpful.
The truth is, nobody needs to accumulate a huge list of tells. A sequel to the AI tells blog post might be interesting, but it won’t help you edit. As an editor, your job is to make language better and more correct — whether it’s written by a human, a bot, or some combination of both.
