I don’t usually turn this on myself, but here we are.

Lately I’ve been refining the Content Lab — my multi-step process for working with AI on these articles, built to lean on my strengths and help with my weaknesses. But something is off, and so I put together some field notes.

What I wouldn’t do to be a kid again. Not because sixty isn’t nifty (how cliché), but because you just did things and didn’t overthink for perfection. Reckless abandon was my childhood. There were no refined edges. Eat the sugar and run off the calories. Jump the bike ramp without thinking about the broken bones. Wear the old, crusty sweatshirt, not caring what anyone else thinks.

We all had those moments. Some more curiosity-driven than others.

And there was nothing wrong with any of it. My edges don’t need to be refined to perfection. My spirit animal is the mess that comes with everything I do. I say what I say, and while sometimes it comes out a little wonky, people who know me get that it’s just me. It’s how I think and act and learn.

Which is exactly what the Content Lab started smoothing away — that’s the “offness” I keep coming back to. I start every session with a brain dump — the good, the bad, and the indifferent. Throw-spaghetti-at-the-wall moments, and AI organizes the information, runs an interview, and pulls out what’s missing. Tells me if there are too many directions. Helps me work out the kinks.

And it works. Or that’s what I thought. Because I looked at the numbers and the data to confirm what my gut has been telling me for far too long. Something is off. The favorite sweatshirt is a little too stiff and ironed out. And so were these articles.

Today, I’ll be showing you what came out of looking closer at a process I built to help me refine my own writing, what was actually underneath the feeling I kept pushing aside, and what that might help you notice in your own work with refining processes before you decide everything is fine.

✳︎ ✳︎ ✳︎

The guardrails were there for good reasons

The Content Lab, a multi-step process for helping me with articles, was never meant to replace me or my voice. It was built to work with me.

And over time, as I kept refining it, I added guardrails to each of the phases. They were supposed to:

• Make sure the story had a flow and gave it a real beginning and end instead of letting it trail off on a tangent.

• Add something actionable when a piece called for it.

• Stop me from overthinking or overanalyzing a decision I’d already made.

• Keep me from repeating myself once I’d already made the case.

• Give me a simple, refined message.

That was the bottom line: a nice, neat, tidy, refined message. Every one of those reasons still holds true. I wasn’t trying to sound like everybody else, or have AI write for me, or even hide behind it. I didn’t want to produce something generic, but I also didn’t want slop. I was trying to fix real weaknesses of mine:

• The tendency to circle through a thought six times before it actually landed.

• The drafts that went nowhere because there was never really an ending.

What I didn’t build in was a way to protect the parts of the process that gave me the real big picture of what I was working through.

The numbers made the feeling harder to ignore

A feeling by itself is easy to talk yourself out of, especially when the thing sitting in front of you technically works. What changed, though, was the one thing that never lies: the numbers.

of GenAI Unplugged and the Subflow connection into Substack GenAI Unplugged and the Subflow connection into Substack made surfacing those numbers easy enough to find. I used Claude to pull the information, have the conversation, and then sit with it — noodle through where the numbers were actually coming from instead of just looking at a dashboard of facts and figures. What I found, comparing my last four posts against a stretch in the spring:

• Views were down by about 25% compared to the spring range.

• Open rate was down to 18–20% recently, versus 24–25% in March and April.

• The top performers in the batch weren’t the polished pieces. They were the practical, work-driven posts about things I’d actually built, messy middle parts included.

Now, I need to be honest about what this does and doesn’t prove. New subscribers came in over that stretch, and new subscribers can quietly pull an average open rate down without anyone disengaging from anything. Topic, timing, and who’s actually in the audience right now can be doing some of that work too — and this is only from subscriber information. The numbers didn’t hand me the cause of the discomfort.

The numbers showed something was off, and that was enough to stop me from explaining the feeling away.

There was relief in it too — not because I liked what I was seeing, but because the numbers were starting to make sense of something I’d been sitting with for a while. It was a justification of the feeling, not in the “am I right or wrong” sense, but the sense that something wasn’t working, and the numbers proved it.

The relationship being edited out

Now, these numbers didn’t tell me what was actually missing. Going back and rereading the stronger pieces next to the recent ones did. The early articles were more me — messy in a very specific way. They kept:

• the wrong turns

• the questions I asked two, three, four times because the first answer didn’t hold

• the thing I got wrong before I got it closer to right

The newer pieces kept the conclusion. They just didn’t keep the walk to get me there. The messiness, what I did wrong, not just the perfect tied-in-a-bow end piece, was being taken out of the equation.

The relationship between me showing the whole process and you watching it happen was what was being edited out.

That’s the unfinished thinking that I was struggling with and that I noticed many of you struggle with. AI can summarize what we’ve learned. It’s very good at that. What it doesn’t do on its own is show how the learning happened. That’s the part that had quietly disappeared.

Not every rough edge was a flaw

For a while, I thought it was the “voice.” It was the word everyone reaches for when something written with AI feels off. But even that seemed too vague to actually diagnose anything, so I needed to sit with it a little longer.

To me, refinement is smoothing away the rough edges. I think I shot myself in the foot a little, because refinement is not my natural state. Those edges are what keep my curiosity flowing.

It’s that never-ending feeling of not knowing what I don’t know, and somehow it feels like my favorite sweatshirt on a cool summer evening — comfortable, broken in, not something I want to try to fix or replace.

It’s like when you were a kid. You didn’t ask “will this work?” — you just did it. The scraped knees, the broken arms, the knocks to the ego. But also the “hey, that was cool, we should do that again” moments — the things that actually happened and turned out to be kind of cool anyway.

Not every rough edge in my writing was a problem I wanted the Content Lab to fix.

• Worth smoothing: confusion, repetition that didn’t serve anything, missing context that made a piece hard to follow.

• Not worth smoothing: the uncertainty itself, the reversals, the place where you’re still figuring something out.

That second kind isn’t a flaw sitting in the draft or brain dump. That’s the part that lets you know someone else feels the same thing.

I don’t have a clean name for the line between those two yet, and I’m not sure I even want to figure one out. Naming it too early is how you end up with a rule that flattens the very thing you’re trying to protect.

I know where to start. I don't know the fix yet.

What I decided was to go back through every guardrail I added to the Content Lab, phase by phase, because something in there is where I think I’m getting edited out of my own process. I’m not sure that’s actually the case. It might be that the guardrails themselves got over-refined. It might be that I’ve been defining refinement wrong the whole time. It might be something else in the workflow that I haven’t quite isolated yet.

Here’s what I’m not doing:

• Stripping every guardrail out at once.

• Bolting on some new rule that tells the process to sound more “authentic” or more like me.

• Rebuilding the whole workflow off one signal before I actually know what’s causing it.

What I am doing is using this article as the test:

• Starting from the real material of the week I document, instead of a topic that’s randomly picked.

• Keeping the raw conversation intact instead of summarizing it way too early.

• Building a record of what actually happened before I let anything interpret it.

• Carrying the exact words and uncertainty forward instead of smoothing them out before the first draft.

• Watching, honestly, whether the finished piece still holds the process I went through, or whether it quietly disappears by the time you’re reading it.

Right now I have a starting point. I don’t have a reworked or repaired Content Lab. I don’t know yet which guardrail is responsible, or how much of what I saw in the numbers is even about all of this.

And if you’re running into something similar — your gut telling you something’s off — start with the numbers. Pick one part of it. You can even start like I did:

Something in my gut has been telling me [this part of my work] is off, and now I’ve got some numbers or evidence that back that feeling up: [describe what you found]. I don’t want to tear the whole thing down. Where would you start figuring out what’s actually going on?

Refining should make the work more mine

Something is off. That’s still true. Not because the Content Lab failed — it did exactly what I built it to do. What it revealed is that my idea of “working” only ever asked one question: did this get clearer? It never asked the other one: did I stay inside it?

I’m not walking away from refinement. I still want the flow, the ending, the restraint. I’m just changing what I check for before I call the process working — clearer isn’t the whole test anymore. Whether I’m still recognizably in it is the other half.

Somewhere in all of this is still the kid who’d rather create the new next ending than accept the one the story already gave her. I’d rather leave that part a little unfinished than iron it into something tidier than it actually is right now.

What did your process improve — and what might it have quietly smoothed away?