Two years ago, I used AI to ensure my output was acceptable. And for a while, acceptable felt like enough.

Roughly 35–40 headlines generated and accepted without real pushback. These headlines went on ad copies or on blog titles or even as webinar titles.

Did any of them underperform significantly? Well, no. But, our growth had stalled. 

That is actually the scarier problem with using LLMs. Mediocre work quietly becomes your benchmark.

I didn't name that at the time. I was just focused on shipping.

Then my manager joined. He was AI-forward,  but more so to enable using AI as a thought partner. And he challenged a lot of my ways of working.

He showed me what better looked like.

Take the image below. My original suggestion was Made in Headspace with Headspace being our product name. My manager suggested adding Heaven with a strikethrough. Just that small addition but it adds so much drama.

That one crossed-out word makes such a massive difference to the copy and the overall impact. And when I saw it, I was pleasantly surprised. “Woah!” I thought, “I need to start rethinking how I create headlines.”

My main learning from this (any many other pushbacks) was to use LLMs as a guiding tool. A tool to help inspire me, and to push back when needed. But I cannot let them do the thinking for me. Otherwise the work quality will be flat and boring.

Which is exactly what it had become.

I pride myself on work quality. That is not a humble claim; it is something I genuinely believe about how I work. And for several months, I stopped checking the details. I didn't catch it myself. Someone else did.

Most LLMs cannot think about text visually. The way words sit on a page is a design and writing decision combined. Whether it is the impact of highlighting a word in a different color or using strikethrough on a word to create extra impact or using a large font size to embolden the impact of what you are saying. It can't feel the difference between a title that sits flat on a page and one that punches.

Those calls are human. And the moment I stopped making them, my work got flatter.

What I find most astonishing, looking back, is that I was honestly unaware. LLMs made me blind to alternative possibilities. I saw the suggestions and thought: “ah well, I will select the best of these options.” I didn't even consider alternatives. The option set became the whole universe of what was possible.

Being the sole person creating content and also tracking it made it extremely difficult to be objective. There was no one to tell me this could be different, this could improve. And so I used AI as a way to ensure my output was standard. Acceptable. Mediocre?

I was a bit flabbergasted at how much I had let it slip through. 

I still use AI every day. But I've stopped treating it as a thinking tool and started treating it as a thought partner.

What's the difference?

Thought partner: Helps you move faster on decisions you're already making. 

Thinking tool: Makes the decision for you.

Recently, I co-founded Spryngbase, where we help organizations figure out how to use AI well. And yes, I can see the irony. The person who let AI flatten her headlines for months is now telling individuals to be more deliberate about it.

But I think that is precisely the point. The skepticism is earned due to my previous experience.  

So. If you are somewhere in this story and are struggling with using AI but also being objective enough to push back (and I suspect some of you are) here are the three real options as I see them:

Option 1: Review what parts of AI you trust and the parts that are inadequate. Challenge the inadequate parts. Not once, but regularly, as your work evolves and your standards sharpen.

Option 2: Do everything manually. This is extremely time-consuming. In the world we live in now, it is also not realistic.

Option 3: :Step away from the laptop when you feel your brain getting tired. Collaborate with AI when you are ready again, or when you have something to actually say, and the judgment to evaluate the output AI gives you.

My manager just crossed one word out,  and suddenly, the actual headline was right there underneath it.

Words and design are not executional details you optimize after the strategy is done. They are the brand identity, they form the accumulation of small choices that tell people what kind of thinking lives inside your company. The impact a single word can have on your messaging is immense.

That is precisely why you need humans to guide AI and not vice-versa. 

Grow-th Architect is where I think out loud about brand-building, positioning, and the slow work underneath the visible stuff. If this resonated, subscribe, or share it with someone who needs to hear it.