You can pry em-dashes from my cold, cramped hands — even though they often signal that a LinkedIn post, for example, was written by AI. 

So how do you use AI in a way that’s actually useful?

Here’s how Hilary Gridley, head of core product and AI at Whoop, answered that question this week on the podcast. 

Hilary Gridley’s Anti-Slop Playbook 

• Build a rubric from your own edits. Gridley suggests taking 10 - 20 examples of work you've edited. Put the originals in column A and your edits in column B. Upload it to ChatGPT (or your AI tool of choice) and ask: “What patterns do you notice in the edits I make over and over?" Then ask it to turn those patterns into five pass/fail criteria. Now you have a rubric you can paste into any custom GPT or Claude project as a system prompt that anybody on your team can use. As an editor, I particularly like this one, because it systematizes my standards without relinquishing control of edits. As a writer, I like it, because it helps me understand what my editor is looking for.

• When you see slop, don't call it bad, just call it a first draft. Think of your AI workflow as draft → feedback → improve. If obvious AI slop comes across your desk, tell your marketer that it sounds AI-generated and ask them to add their own spin to it before shipping it. 

• Build small, specific tools, not a second brain. Unless you get a lot of concussions (and I really hope you don’t), you don’t need a second brain. So don’t build an AI tool that captures everything you know. Instead, build a dozen tiny, hyper-specific ones — one for exec emails, one for A/B test design, one for problem statements. The more specific the tool, the more useful and accurate the feedback. Gridley’s made dozens of these, each scoped to one exact use case.