I often think of writing as a fairly straightforward act. For me, it’s a process of transfer and organization—a way to get ideas out of my head and onto the page.
But fascinating research suggests that how we externalize thought can change the very nature of the thoughts themselves, along with how well we remember them. That got me thinking—and writing.
This study, published in Consciousness and Cognition, compared two simple tasks that included writing words by hand and drawing pictures of those words. On the surface, both methods might seem equally effective for learning. But the researchers found something very interesting: Drawing consistently led to better memory performance than writing.
As you might guess, this got me thinking about LLMs and typing into that context window as a central tool to our “thinking” these days.
Drawing’s Cognitive Advantage
The research used a method called multidimensional experience sampling, which interrupts participants during a task to ask about the “quality and focus” of their ongoing thoughts. This let them look beyond the final written word or the sketch, and into the mental space where the work was happening.
When people drew, their thoughts tended to be more visual, more elaborate, and more connected. They weren’t just transcribing meaning but were building it in multiple modes at once in a triad of visual imagery, motor movements, and spatial relationships. These richer thought patterns, in turn, supported stronger recall later. Writing, by comparison, tended to stay more linear and verbal, activating a narrower slice of the mind’s capabilities.
From Pen and Paper to AI Keyboards
The study didn’t mention artificial intelligence, but its implications are hard to ignore in an age where more and more of our “writing” is co-authored with large language models. The leap from pen to keyboard may already strip away some of the sensory and motor loops that make learning durable—and the leap from keyboard to AI output may risk removing even more. At least, that’s my hypothesis based on the study’s findings.
With LLMs, we often bypass the slow, messy, generative phase of writing entirely. Instead of assembling thoughts through trial, error, and revision, we can feed the system a prompt and receive a mostly-finished block of text in seconds. Our role shifts from creator to selector, where scanning, tweaking, and approving drive content rather than building from the ground up.
The Danger of Skipping the “Making” Phase
The drawing study suggests that this shift matters. Those elaborative, multimodal thought patterns that strengthen memory emerge from making, not just evaluating.
When AI takes over the making, we’re left with thinner mental traces or paths. The words exist, but they live outside of us, stored in the machine’s output rather than embedded in our own cognitive architecture.
In my view, this isn’t an argument against using AI. It’s an argument for using it with more intention.
If drawing’s advantage comes from engaging the mind in multiple channels, then maybe our AI workflows should make room for that kind of engagement too. That could mean sketching an outline before prompting, diagramming relationships between ideas, or freewriting a rough draft before asking the model to refine it.
Designing for Cognitive Richness
The key is to preserve or maybe even to amplify the generative friction that AI so efficiently removes. That friction is where the deeper connections form. It’s where ideas get pushed, pulled, and sometimes broken before they’re rebuilt into a final form. Without it, we risk producing work that’s clean on the page but rather vague in the mind.
Of course, we’ve long known that the tools we use change the way we think. The quill, the typewriter, and the word processor have all left their marks on our mental habits. And today, LLMs are just the latest tool. But they arrive with a speed and fluency that might short-circuit the mental processes that make knowledge stick.
My sense is that this study offers an important reminder that the process matters as much as the product. If we want AI to serve our thinking rather than replace it, then we need to design our interactions so that we remain active participants, not just passive recipients of results. That means keeping our hands and our minds in the work, even when the machine can do it faster and (in some instances) better.
And the key point here is that memory and understanding aren’t built in the output; they’re built in the making.
