The Prompt That Never Works

One of the prompts I use most often is also the one I’ve started to question the most.

“Make it sound more human.”

The more I worked with AI, the more I realised that’s an impossible instruction, because there is no such thing as a human voice. There are billions of them, each shaped by different experiences, biases, habits and imperfections, which means that every time I ask AI to sound more human, what I’m really asking is for it to sound more like me.

The irony is that I spent years learning not to do that. Like many people working in technology, I learned to value consistency over personality, polish over rough edges, and clarity over character. We built design systems, brand guidelines and writing conventions to make our work feel coherent. Then AI became exceptionally good at producing exactly that style, and suddenly I found myself deleting perfectly good sentences because they felt too polished to belong to anyone.

That shift changed the way I prompt. I stopped treating AI like a writer and started treating it like a collaborator that needed better context than “be more human.”

A few things made a surprising difference.

Research before you generate

AI’s greatest strength isn’t writing. It’s synthesis. Asking it to research a topic, compare perspectives or identify patterns before generating an answer produces work that feels more informed and less generic.

Give it context, not adjectives

“Human” is vague. Context isn’t. Share examples of your writing, explain who you’re speaking to, describe the feeling you’re trying to create, or even build a mood board of references. The richer the context, the less the model has to guess.

Pay attention to your fingerprints

Every edit you make reveals something about your voice. The phrases you always keep, the transitions you always delete, the ideas you instinctively return to are far more valuable than a list of adjectives describing your style. Those patterns become the best prompt you’ll ever write.

Build a council, not a critic

Instead of asking one model whether something is good, ask several perspectives why it might not be. One can challenge the argument, another can edit for clarity, another can question the assumptions. The reasoning is usually more valuable than the verdict.

Building Tansei→ https://tansei.io

As my prompting evolved, so did my workflow. I realised my creative process no longer lived inside a document. It lived between prompts, copied snippets, screenshots, and ideas I wasn’t ready to throw away. That observation eventually became Tansei: not another clipboard manager, but a place to preserve context while thoughts were still taking shape.

The biggest lesson, though, had nothing to do with prompting.

AI didn’t force me to become a better writer. It forced me to answer a much harder question that I’d somehow avoided for years.

What does my own voice actually sound like when I’m no longer trying to sound like everyone else?