Using Grok 4 as a Dev Editor

I’ve been a developmental editor for over fifteen years. I’ll just say I know a thing or two about a thing or two.

Grok 4 got rolled out to me today, and it’s a definite improvement. I’ve already been making extensive tests on Grok 3, which was the first model that really seemed to get what I wanted, though it was only after I hit on a simple but specific prompt. That prompt is:

“Acting as a developmental editor for (genre), avoiding unnecessary enthusiasm and praise, please give me a straightforward analysis on . . .”

(Yes, I’m one of those guys who likes to say please. I’ll make jokes about surviving the AI takeover, but the reality is that I think it helps me phrase my prompts better.)

I don’t have the funds to go around with a ton of different paid AI accounts, so Grok is the only one I have top-tier access to (and even then it was because I picked up an X Premium account on sale last year). But as I’m sure we’ve all noticed, AI is built to be molded by you, the user, during a thread, unless and until you hit a hard-line for its programming. To that end, it is programmed to be very friendly.

Speaking as one, you don’t want an editor to be concerned about being friendly first. You want your editor to be just shy of mean. Your editor should be the literary equivalent of the personal trainer who won’t let you quit because you can ABSOLUTELY do one more set, so get to it. Your manuscript won’t get better without criticism; but AI won’t give you firm criticism most of the time unless you tell it to or hit one of its walls (such as, at least for Grok 3, anything to do with Atlantis will trigger a lecture on how Atlantis never existed; it’s fine with dragons and wizards and aliens, but Atlantis, horrors!).

I’ll also add that AI Thomas is also very enthusiastic and friendly . . . but that’s how Flesh Thomas works too, so I haven’t done that prompt there yet. It’s on brand.

So you can get some excellent use out of Grok as a kind of “pre-gamed” dev editor, using it to spitball ideas and give instant feedback, as well as some very impressive research and calculations now that it’s on Grok 4. My latest test was getting it to analyze a description of a fictional high school in a fictional New York city in 2002, and it was able to pull from historical records on high school population statistics and economic variables without more than a general prompt.

Where it still fails is on understanding human motivation and conversational flow, so I wouldn’t get it to analyze dialog or emotional responses.

But it’s shockingly accurate at describing fight moves; I’m a former martial artist (karate, mostly) and I asked it to handle a street-fight scenario I was specifically trained for in the dojo, in terms of specifically Okinawan karate. It was almost exactly what I was trained to do twenty years ago. So if you’re struggling with fight scenes, load a description of the start of a fight, how you want it to end (vaguely; give it as much freedom here as possible), tell it how it should choreograph (savate, karate, BJJ, MCMAP, etc.), and then let it generate a blow-by-blow. Then — and this is crucial — shrink it down to the absolute minimum description. Most fight descriptions should really come down to variations on “he hit him, hard.” The point of this is to look for anything extra you might want to include.

I haven’t tested it on using guns or swords yet, but I suspect it’ll be accurate enough to help you out. Then you can take it to a subject matter expert and ask if it’s good enough.

One middle-ground element is that it’s still having trouble analyzing an outline properly. However, so far at least, it’s pretty accurate when it says something doesn’t fit, even if its replacement suggestions seem to be the equivalent of a coin toss. Roughly half the suggestions seem accurate to me, half don’t, so don’t pay much attention to that side; in this kind of thing, it’s still more like a very knowledgeable beta reader than a dev editor.

But using AI like this, once you know what to do, will absolutely cut down on your need for a human dev editor. It won’t replace a human editor, but it will be better than a bad editor (and will help you identify a bad dev editor; there are a lot out there who claim to be one, but are really just copy editors looking to inflate their credentials).

If you do use a human dev editor, go to that editor with something like “I ran this part through (LLM model) and it identified this flaw; can you tell me how to fix that?” That means the editor can focus in on specific tasks, which saves you money and the editor time and frustration.

Yes, as a dev editor that means I would potentially be losing money if my clients are using AI to pre-game before coming to me. I am personally betting it will get me more jobs, though, not fewer. If more authors are able to understand how dev editors can help, and come to me with specific factors rather than general questions or not at all, then I can make it up on volume.

I’m not at all nervous about losing my job to AI. AI is a tool; writing is an art. Use your tools wisely.