When Machines Guess My Mind: The Strange Art of Transcribing Handwritten Thoughts

I love writing with a fountain pen. There’s something deeply grounding about starting my day with pen on paper, right after my morning chores. It feels like watching raw thoughts slowly take shape—untidy, personal, and alive. Over time, many of these thoughts turn into concepts about life itself. Naturally, I began wanting to preserve them—store them safely on the cloud and occasionally share a few publicly.

That’s when I started using ChatGPT to transcribe my handwritten notes into editable text.

In theory, it sounded perfect. In practice, it turned into a year-long experiment in humility.

Despite all my efforts, I still haven’t reached where I want to be with my handwriting—neither in legibility nor in beauty. The sloping headline on my palm from a palmistry angle, if one were to believe, does not allow me to be “unartistic” even if I try. So I end up with writing with ornate strokes. To me, form is just as important as function. And so, when I do something as direct with my hand as handwriting—I automatically slip into styling certain letters making it hard for a machine to read my writing. 

That’s part of the problem.

I also write in a slant style, not in carefully rounded, upright letters. Add to this my long entry and exit strokes, and suddenly my handwriting becomes far less friendly to machines. For a tool like ChatGPT, which relies on matching patterns and templates, some of my letters simply go out of bounds. When that happens, it starts guessing.

And here’s where things get truly fascinating—and frustrating.

If the system thinks it has correctly understood one phrase, it often confidently guesses the rest of the sentence in that same direction. Instead of getting a few wrong words, I sometimes end up with a completely different sentence altogether. The chain of thought breaks. Meaning shifts. Logic bends.

Ironically, this is not just a handwriting problem—it’s a lesson in how artificial intelligence actually “understands.”

If I truly wanted flawless transcription from handwritten notes, the solution is brutally simple:

• Write in clean, large letters
• Leave generous space between words
• Use round shapes
• Keep the writing upright
• And completely abandon ornamentation

But that would mean abandoning myself.

There’s another factor I hadn’t considered deeply enough earlier—contrast. Since all this transcription is based on optical reading, contrast is everything. This time, I accidentally used a single-ruled notebook, and that introduced a new problem. With larger letters, the long strokes from one line occasionally drift close to the strokes from the next line. To a machine, those overlaps blur contrast and cause misreads—even though my own eyes can separate them effortlessly.

Of course, there is a near-perfect workaround.

I can simply read my notes aloud and let ChatGPT transcribe my voice. And when I do that, accuracy is excellent. But I’ll admit it—I’m lazy. And sometimes, embarrassingly, I can’t even read my own handwriting very well.

What can I say.

Still, today’s deliberate attempt—slow writing, larger letters, controlled slant—proved something important to me:
It is possible to meet machines halfway without surrendering your personal style entirely.

And maybe that’s the real lesson in this dance between pen and algorithm.