The easiest way to sound unlike yourself with AI is to use it at the exact moment you feel most uncertain. Uncertainty makes polished output feel especially persuasive — and persuasive output is easy to mistake for better writing. I've done this enough to recognise the pattern quickly now. The paragraph arrives clean and competent. It also sands away the edges that made the idea feel personally held. That's the trade many writers fear when they think about AI assistance, and it's a reasonable fear.

The good news is that the trade isn't inevitable. Voice can survive AI help if you design the workflow carefully. In my experience, that means using the tool for expansion, challenge, and cleanup while reserving the emotional and rhythmic centre of the piece for yourself. Voice isn't only vocabulary — it's pacing, specificity, honesty, and what you choose to admit.

Preserve Your First Natural Language

Before asking AI for help on a piece, I try to write at least a rough opening or a set of notes in my natural language. Not polished language — just mine. This gives me an anchor for cadence and emphasis. Once I have that, the tool can support structure or revision without completely resetting the tone of the piece.

When I skip this step, I notice my final draft drifting toward generic eloquence. It becomes harder to hear the difference between clarity and flattening.

Ask the Tool to Critique, Not Impress

A reliable way to protect voice is to ask AI for critique rather than performance. Where does the writing sound vague? What would a sceptical reader question? Which parts feel like clichés? These prompts help me improve the draft without replacing its character. By contrast, asking the tool to rewrite whole sections often introduces a polished accent that's difficult to fully remove later.

Critique-based use keeps me in the position of author and editor. That distinction matters more than it first appears.

Reinsert Specific Human Material

Whenever a draft starts sounding too smooth, I add back concrete human material. A scene. A detail from a recent day. A sentence that admits uncertainty. An observation that's slightly awkward but true. These additions restore texture. AI tends to be very good at producing acceptable generalities. It's less naturally grounded in your lived specifics unless you deliberately put them back.

Readers feel this difference immediately. Specificity creates trust because it signals contact with reality, not just fluency.

Read the Draft Aloud Before Publishing

My favourite final test is reading the draft aloud. Voice becomes much easier to judge with the ear than with the eye. If a sentence sounds like something I'd never actually say, I change it. If transitions feel too engineered, I loosen them. If the confidence feels inflated, I add honesty. Reading aloud catches the subtle ways machine-influenced writing can drift into a register that's technically fine and personally unconvincing.

The ear is still one of the best defences against accidental genericness. It hears artificial smoothness faster than the eye does.

AI can support writing without erasing voice if you keep the first natural language, ask for critique more often than full rewrites, deliberately add back specific human material, and read the final piece aloud. The goal isn't to reject assistance. It's to keep authorship audible.