I want to be upfront about something before I get into this: I was skeptical.
I'd seen too many "AI will transform your marketing" takes from people who'd never actually tried to produce enterprise B2B content at scale. Content that has to pass legal review. Content for four different markets. Content that a subject matter expert who has been in the industry for 30 years will read and immediately know if it's generic.
So when I started building AI into my content workflow, I wasn't looking for magic. I was looking for anything that could take the grind out of the process without taking the quality with it. Here's what actually worked.
The shift that changed everything
Most people use AI wrong for content. They open ChatGPT, type "write me a blog post about demand generation," get back 800 words of bland prose, and either use it as-is (bad) or throw it away (wasteful).
The insight that changed my workflow: AI is a research and structure tool, not a writing tool.
I stopped asking it to write. I started asking it to think with me. Before I write anything now, I use Claude to pressure-test the argument. I'll paste in my rough thesis and ask: what's the strongest objection to this? What am I missing? That process — 10 minutes of back-and-forth — used to take me a full day.
The actual workflow
Step 1: Brief development (AI-assisted) — I start with a prompt that includes the audience, the specific problem they have, my take, and one concrete example from my own experience. I ask Claude to identify gaps in my reasoning. 20 minutes instead of two hours.
Step 2: First draft (human) — I write it. This is non-negotiable. The voice, the examples, the specific things only I know — those can't come from a model. If you're not writing the first draft, you're producing content that sounds like everyone else.
Step 3: Structural edit (AI-assisted) — I paste the draft and ask: does this argument hold together? Is the structure clear? Two minutes of feedback that used to require a trusted editor or a day of distance.
Step 4: Distribution variants (AI-assisted) — One piece becomes a LinkedIn post, email snippet, short-form version, and quote pull. Used to take an afternoon. Now takes 20 minutes.
The guardrails that keep it from going sideways
Every piece of AI-assisted content gets a full read-through with one question in mind: does this sound like something I would actually say? If there's a sentence I wouldn't use in conversation, it gets rewritten.
I don't let AI generate statistics or data points. It will confidently cite things that don't exist. I source all numbers myself.
I keep a "voice file" — a running document of phrases, analogies, and ways I tend to explain things — that I paste into context windows when doing heavy AI collaboration.
The honest number
Across the content function I run, AI integration reduced production time by about 60% on average. The quality didn't drop. In a few areas it actually got better, because I was spending less time on process and more time on argument.
But it took about three months of iteration to get there. The first month was mediocre. The second was better. By month three, the workflow was locked in. If you're starting, don't expect instant results. Expect a learning curve. It's worth it.
— Erik R. Miller