The slow death of vocabulary: why I stopped letting AI write for me
Here's something I noticed recently that genuinely unsettled me, something that made me question the entire way I'd been approaching writing for the past two years.
I was writing an email to a colleague, a simple message of just a few sentences, and I found myself stuck on a single word—it was on the tip of my tongue, that perfect word that would express exactly what I meant with the right nuance and tone, and I simply couldn't retrieve it from my own brain. I paused, stared at the screen, and then instinctively opened a browser tab to type "another word for..." into the search bar.
That's when the realization hit me with uncomfortable clarity: I had been letting AI write for me for two years, and my vocabulary had quietly been shrinking the entire time without me noticing. I had gradually transformed from a writer who reached for words into a selector of AI-generated options, and I wasn't sure when that transition had happened or how I'd let it go so far.
The atrophy of vocabulary
It makes perfect sense when you think about it, even though it's uncomfortable to admit.
Vocabulary is like any other skill in that it needs to be used regularly or it atrophies, and writing is the primary way most of us maintain and grow our active vocabulary. When you write, you reach for words, you make choices between alternatives, you weigh nuances, and sometimes you struggle to find exactly the right word—and that struggle is the work itself. That struggle is what keeps your vocabulary alive, what forces you to maintain that mental dictionary of words you can actually summon when you need them.
When you let AI write for you, you stop doing that work entirely, and the consequences are predictable.
You give the AI a prompt, it generates text, you read it, you accept it, and you move on without ever reaching for a word or making a meaningful choice about language. Over time, your active vocabulary shrinks because the words you used to reach for become harder to retrieve, the nuances you used to express become harder to articulate, and you gradually transition from being a writer to being an editor who simply recognizes good options rather than generating them. And editors don't need rich vocabularies; they just need to recognize good options when they see them.
The clarity crisis
But the vocabulary issue, concerning as it is, is just the surface of something deeper.
Writing isn't merely about words; it's about thinking, and writing is fundamentally thinking made visible on the page. When you write, you're forcing yourself to clarify your thoughts, to build coherent arguments, to find structure in your ideas, and to make connections that weren't obvious before you started. This is hard work that takes time and effort, but it's also essential cognitive work that produces clarity of thought.
When you outsource the writing, you outsource the thinking that would have gone into it.
The AI produces text that looks clear and well-structured, grammatically correct and logically organized, but it's not your thinking and it's not helping you think better. You don't understand the material more deeply, you don't clarify your own position on the topic, you don't build the argument in your own mind—you simply review and approve what the AI has produced. The writing becomes a substitute for thinking rather than an expression of it, and that's a fundamental problem.
The three ways I write now
Over the past two years, I've developed three distinct approaches to writing with AI, each serving a different purpose, and I've come to realize that I need to be much more intentional about which approach I use for which job.
The first approach: full-draft generation.
Like everyone in 2024, I jumped on ChatGPT and Claude with enthusiasm because the promise was genuinely irresistible: give it a brief, get a full draft, polish it up, and publish. Simple.
It wasn't simple.
I found myself spending hours rewriting every single draft, removing em-dashes that the LLMs overused, cutting nonsensical sentences that looked correct but said nothing, humanizing generic prose that sounded like it was written by a committee of robots, and fact-checking hallucinations that I couldn't trust. I was doing the hardest part of the work—making it sound human—after the AI had done the easy part, and I was doing it every single time.
The second approach: agentic programmatic SEO.
I eventually realized that the full-draft approach was terrible at scale, so I built a pipeline where multiple AI agents work in sequence: one finds topics, another develops angles, a third writes research files, and a final one drafts the complete article. I can set gates to review at key stages or let it run fully automated, and suddenly I could publish twenty evergreen articles in the time it used to take me to finish one.
Technical guides, basic how-tos, news roundups—content that needed to exist but didn't need me specifically, and that was fine because these articles aren't personal or expressive. The catch is that my voice isn't in there at all, and that's acceptable because that's not what they're for.
The third approach: Levee.
I still needed something for articles where my voice actually matters—opinion pieces, personal reflections, articles where my perspective is the whole point—and I kept falling back to the broken full-draft method because nothing else existed. So I built Levee to fill that gap, and it doesn't generate full drafts but instead extracts my voice from a sample and then suggests words and sentences as I pause.
No generic prose to rewrite, no hallucinations to fact-check, just my voice expressed faster while I remain in control from the first word.
What I'm realizing now
The three approaches work well for different kinds of content, but I've been using the first two too much and the side effects are becoming impossible to ignore.
The vocabulary atrophy is real and noticeable when I try to write something personal or unassisted; the words don't come as easily as they used to, I reach for phrases that sound like AI, and I default to generic language because it's the path of least resistance. The clarity crisis is equally real because when I write without AI assistance, the struggle is harder and the thinking is muddier, the arguments are less structured, and I've simply gotten out of practice at doing my own cognitive work.
That's why I'm forcing myself to use Levee more often now, not just for opinion pieces but for more of my writing, because it makes me write rather than writing for me.
Why writing yourself matters
There are benefits to writing that go far beyond the final product, benefits that you completely miss when you outsource the work to AI.
Introspection is one of these benefits, because when you write you're forced to examine your own thoughts, you can't hide from them or avoid them, and you have to articulate them in a way that makes sense to others. This is how you understand what you actually believe, how you clarify your position on complex topics, and how you discover what you think—often surprising yourself in the process.
Deep work is another benefit because writing requires concentration, cognitive effort, and sustained attention in a way that few other activities do, and this is exactly the kind of work that produces genuine insight and understanding. The shallow work of selecting AI options doesn't produce the same benefits because it doesn't require the same cognitive engagement.
Idea generation happens through the writing process itself, because you start with a direction but the act of writing produces new thoughts—you write a sentence, that leads to the next sentence, and suddenly you've discovered a thought you didn't have before you started. When you let AI write for you, you skip this generative process entirely, getting the output without the insight that produced it.
Cognitive maintenance is the final benefit because writing maintains your cognitive abilities, keeping your vocabulary sharp, your thinking clear, and your ability to articulate ideas intact. It's exercise for the mind, and if you stop doing it, the abilities atrophy just like any other skill you don't use.
My commitment
I'm not giving up the AI approaches entirely because they still serve useful purposes—the agentic pipeline is still valuable for evergreen content, and the full-draft approach still works for one-off informational articles. But I'm using Levee much more now, because writing is thinking and I want to keep doing my own thinking.
The vocabulary atrophy is reversible, the clarity crisis is fixable, and the cognitive decline is preventable through consistent practice. The answer is simple: write more, write with Levee rather than instead of it, and keep exercising the cognitive muscles that make writing worth doing in the first place.
Try Levee free at levee.theleverage.tech—no card required. Write with AI, not instead of it. Keep your vocabulary. Keep your clarity. Keep your thinking.