GenAI is a helpful tool except when it makes you look (and sound) like a tool because you didn't revise it. Here are the not-so-subtle tells you didn't put in all the effort necessary.
I got a 12-page, 4,253-word report to edit the other day. It started well enough with a good summation of the issue and a quick preview of what was to come. The author was someone I've not edited before, so I didn't have a sense of whether this was their voice or not, but it seemed pretty in line with the company tone. Cool.
But as I read, I noticed the familiar patterns that AI had done most of the work. Here's what tipped me off:
Ideas & Support
I've edited a lot of papers for this company and have never once in the five years I've been doing it have I seen a textbook cited in their reports. It's just not how this firm does things. This report cited two textbooks in one section. OK, maybe that's the author's thing (they're new here). But they really stand out as the only two across the dozen pages. So I looked them up. One is a current textbook for a 101-level class in a related but different discipline. The second was worse: an outdated edition of a textbook about personal finance (not at all cogent to the topic of this paper, but having the same key words). I flagged both, pointing out the style break.
Fix: Verify that sources delivered by AI are appropriate for your audience, including reading reviews for books or other works.
Organization & Logic
Each section followed the same structure, probably because each section's topic was used as a prompt. Same number of paragraphs (maybe that was in the prompt, too?) and often slightly out of logical order and frequntly buried the lede. I admit that this one's more of a feeling that you get after reading several sections.
Fix: Reorganize to create a more logical flow.
Sentence Fluency & Structure
Almost every section included at least two sentences that included a semicolon. We love to bash the em dash (undeservedly), but for me, the giveaway is the semicolon sentence. I use semicolon sentences sometimes — more often introduce a bullet or numbered list — but not in every section of a long-ass paper. This is a pattern I see a lot more since the rise of AI. The machines learned it somewhere, but like the em dash, they overuse it, living out the curse of too much of a good thing. (There were a few em dashes on the scene, but they were correctly deployed.)
There was another recurring pattern, that's not an x problem, that's a y problem, but that's so niche that I think it's probably the author's doing (it's used as the paper's title.) It might have been an effective device, referencing the title as it does, but with the other repeated pattern, it didn't feel natural or clever.
Fix: Look for repeated patterns, including sentence length and structure, and rework them to reduce repetition, improve fluency and increase impact.
Word Choice & Vocabulary
One thing people keep telling me they love about genAI is how it helps them write in a style that suits the intended audience, because you can bake that into your prompt. Unless you don't. Or you don't accurately assess them. One of those two things appears to have happened on this project. There were a few words in each section that were too technical for the audience this company addresses. Now, that's fine if there's enough context to guide the reader to understanding, or if you explain the term in line. But neither of those was done here, so the concepts were beyond the audience.
Fix: Review technical terms and complex comments, replacing or explaining to keep readers engaged while they're learning new things. You could also include a glossary.
I don't begrudge the author using AI to draft this report. But as I've said on here before and will keep saying, I am begging you to revise that draft. I've shared a bunch of tried-and-true strategies for you on this blog. Want more detailed and personalized help? I got you. Sign up for one-time and recurring coaching or a group training session for a deeper dive that will make you a stronger writer in the bargain.