We love a good “here’s a headline explaining why the sky isn’t falling” post. When it comes to AI content, one of our favorite marketing minds just did exactly that.
Our very own Mickey Mellen, over at his personal blog MickMel, recently ran an experiment worth stealing. He answered 100 questions from Claude about how he thinks and writes, then let the AI draft part of the post based on his answers. The idea came from marketer Ruben Hassid, who put it bluntly: most people can’t articulate their own taste, so they can’t hand it off to anything, human or AI. That’s the gap that turns AI writing into wallpaper.
Why generic content is the real enemy
We’ve said it in our own brand guide for years: our goal is to be clear, practical, and personal, never confusing or generic. Hand a client brief to five different agencies, and you’ll get five nearly identical paragraphs about “leveraging synergies” and “driving growth.” None of it sounds like an actual business run by actual people.
AI doesn’t cause that problem. It just makes it faster. A tool like Claude or ChatGPT, given a vague prompt like “write a friendly blog post about our services,” will default to the statistical middle of everything it’s ever read. Fluent, confident, and about nobody in particular.
The fix isn’t avoiding AI. Plenty of good writing today starts as an AI draft and gets shaped by a human editor before it goes anywhere. The fix is giving the tool something specific to sound like, instead of a one-line tone instruction.
The 100 questions, and why they work
Hassid’s approach treats an AI like an interviewer, not a typist. Instead of describing your voice in adjectives (friendly! bold! authentic!), you get asked pointed questions across a few categories:
- Beliefs you hold that most people in your industry would push back on
- Writing mechanics, like how you open a piece, how you close it, and which words you overuse
- Pet peeves, the specific phrases that make you distrust a piece of content
- Structure, including how you feel about bullet points (we clearly have opinions)
- Hard nos, the topics or approaches you’ll never touch
- Red flags, the signals that tell you someone doesn’t actually know their subject
The genius of the exercise isn’t the questions themselves. It’s that answering them forces you to notice contradictions. Mickey mentioned in his post that he thought of himself as someone who hedges his claims, then realized his actual interview answers were flat and declarative. Claude caught the mismatch and asked him about it directly. That’s the kind of self-knowledge a one-paragraph “our brand voice” document never surfaces.
This isn’t really new. It’s just faster.
If this sounds familiar, it should. It’s the same reason we push clients to define their voice before we write a single headline. Specificity is the whole game. “We’re friendly” tells a writer, or an AI, nothing useful. “We never say ‘circle back,’ and we’d rather explain something three times in plain English than once with a buzzword” tells them everything.
What AI changes is the speed and depth of that process. You can go from a vague sense of your voice to a detailed reference document in an afternoon, instead of workshopping a tone-of-voice slide for a month. Once that document exists, it’s portable. Hand it to a new team member, a freelance writer, or your own next AI session, and the output gets closer to actually sounding like you.
How to try a smaller version yourself
You don’t need the full 100 questions to get value here. A shorter, honest version of the exercise works for most small business owners who don’t have hours to spare:
- Ask yourself what you believe that most people in your industry would roll their eyes at. This is the question Hassid recommends starting with, and it’s a good one. Strong opinions are the fastest route to a voice that doesn’t sound like everyone else’s.
- List the words and phrases you’d never use. Jargon, hedging language, whatever makes you cringe when you read it in someone else’s marketing.
- Pull three pieces of your own writing and look for patterns. How do you open? How do you close? Where do you use humor, and where do you get serious?
- Write it all down in one document, then hand that document to whatever AI tool you’re using before you ask it to draft anything.
The output still needs a human to review it. That part doesn’t change, no matter how detailed the voice profile gets. But a well-built voice document is the difference between a draft you have to rewrite from scratch and one you just need to polish.
Where this fits into your marketing
This is exactly the kind of small habit that separates content people remember from content people scroll past. If your team is drowning in the “what should we even post” problem, or every draft that comes back from a freelancer or an AI tool reads as if it could belong to any business in your industry, this is worth an afternoon of your time.
To give the full exercise a shot, go to Ruben’s post, specifically at the section titled “Step 2: The Interview.” Read this section, or watch his video tutorial on how to copy/paste his prompt into an AI of your choice, and then start a conversation with your AI task. Once you’ve completed the conversation, save this conversation as an .md file and tell your AI to read this file first before writing something on your behalf (Ruben goes into more details about this in “Step 3: How to Use It”).
It sounds time-consuming, but in the end, you will teach your AI system how to write like you and no one else.
If you have any questions about using AI for your writing or want to share your experience with this exercise, reach out to GreenMellen today.
