Why AI won’t make copy editors obsolete

I do developmental editing. That means I edit text for everything from meaning and connotation to rhetorical consistency to subject-verb agreement. Obviously (he wrote arrogantly) that requires judgment, and cannot be replaced by a machine.

But what about copy editors (or as they are known in the UK, subeditors)? Copy editors check text for spelling errors, grammatical errors, consistency, and in some cases factual errors. Surely that’s a purely mechanical process that will eventually be replaced by a piece of AI software, right?

A good copy editor is never going to be replaced

Software (like Grammarly) already finds spelling and grammar errors. With AI improvements, I expect it will soon become better at finding factual errors as well. Why couldn’t it eventually get smart enough to replace a copy editor?

Because, just like developmental editing, copy editing requires judgment.

For example:

• Should a passive voice sentence be changed to active voice? The answer depends on the audience (academics use far more passive voice than lay audiences), how many other sentences surrounding it are also in passive voice, and whether the change would actually clarify the text or reduce an intentional emphasis on the object of the verb.
• Consider a word like “yup.” It belongs in some contexts (like an informal blog post) and not in others (such as a grant application). It’s neither right nor wrong. It’s context dependent, and the context isn’t necessarily easy to determine.
• When should a list be written out as bullets? When it’s clearer to read that way. That’s a judgment call.
• When is a clearly false statement intended to be sarcasm? If it is sarcasm, will readers recognize it as such? Of all the things humans can do, recognizing sarcasm appropriately will be one of the last that machines will learn to do consistently.
• Copy editors should flag some frequently repeated words — if your manuscript has 100 instances of the word “strategic” in 3000 words, it’s probably a problem. But other repeated words are fine; in an article about water resources, the word “conservation” might be used in half the paragraphs. The question of when a repeated word is inappropriate requires judgment about which are common words, which words have specific meanings in context, and which words the writer is leaning on to avoid nailing down their specific meaning, or because they think it makes them sound important.
• When is it necessary to enforce consistency? A shift from present tense to past tense might be an error, or it might be an intentional choice the author made to create an emotional response. Text that goes from formal to informal might reflect sloppy word choices, or it might be an attempt to persuade the reader of the inescapably human nature of an argument.
• If a text includes a reference to “Megan Thee Stallion,” is the audience going to understand who that is? If the audience is a bunch of 50- and 60-year old lawyers, maybe not. If it’s a 30-year-old marketing copy writers, it’s probably fine. A copy editor know the difference.
• I just read this passage in a SubStack by Paul Krugman: “First, look at the chart. The second line claims that it shows ‘medium income’ — a term unknown to economics. Clearly it was supposed to say median income. OK, speling misteaks hapen. But not, usually, in charts prepared for a presentation by the President of the United States.” A smart copy editor knows not to correct “speling misteaks hapen.” But could an AI make that judgment?

I could go on, but I don’t have to.

AI-powered copy editing software is going to catch a lot of errors. But it will also catch some problems that are not actually errors, and will miss subtle problems that only an intelligent human familiar with prose, audiences, and context would catch.

Just as with writing and research, AI can help writing professionals. But it can’t replace them.