How to write with AI without losing your voice

Most writing advice about AI is either a sales pitch or a panic. Neither helps you finish a piece. What follows is the practical position: AI is very good at a small number of writing tasks, actively harmful at a few others, and the difference matters more to your career than your opinion of the technology does.

BRYME tracks an AI policy field on opportunity listings for a reason. Publications have taken positions, those positions vary, and breaching one quietly is how writers lose relationships they spent years building.

What AI is genuinely good at

Interrogating your own draft. Paste your paragraph and ask what is unclear, what a hostile reader would object to, or what you have assumed the reader already knows. You are not asking it to write. You are using it as a fast, tireless, slightly dim first reader.

Breaking a blank page. Ask for fifteen angles on a subject. Twelve will be obvious and two will be wrong, but the thirteenth sometimes shows you the piece. Take the angle and throw the sentences away.

Mechanical transformation. Reformatting a reference list, turning notes into a rough outline, converting a bulleted list into prose you will then rewrite. This is typing you did not have to do.

Explaining things you do not know. Then verifying them somewhere else. Language models state wrong facts in exactly the same confident register as right ones, so anything load-bearing has to be checked against a real source.

What it is bad at, and why it matters

Your actual sentences. AI prose is fluent, symmetrical and empty. It reaches for "delve", "tapestry", "navigate the landscape" and "it's not just X, it's Y". Editors read hundreds of submissions a week and have become extremely good at spotting it — not because it is ungrammatical, but because it says nothing at a consistent, pleasant volume.

Anything specific to you. The reason your piece is worth publishing is the thing only you noticed. A model cannot produce that, because it does not have it. The moment you let it generate your central observation, you have removed the only part an editor wanted.

Facts, quotes, statistics and citations. Models fabricate references that look perfect — real journal, plausible authors, an article that does not exist. Never submit a citation you have not opened.

The disclosure question

Read the guideline before you use anything. Policies cluster into four positions:

• Prohibited outright. Most literary magazines. Using AI at all breaches the submission agreement.
• Permitted for research and editing, not generation. The most common professional position.
• Permitted with disclosure. You must say what you used and how.
• Silent. Assume the strictest reading. A policy that does not exist yet is not permission.

If you are unsure, ask the editor in one line before you file. Asking costs you nothing. Being found out after publication costs you the relationship and sometimes the fee.

A working method

1. Think and outline yourself. The structure is the argument. Hand that over and you have handed over the piece.
2. Draft yourself, badly and fast. Your bad draft contains your voice. An AI's good draft does not.
3. Use AI as an editor, not a writer. Ask what is unclear, what is repetitive, what a sceptic would push back on. Then fix it in your own words.
4. Verify every fact independently. Every one.
5. Read it aloud. Anything you would not say out loud is not yours. This single step catches most of it.
6. Disclose if required. In the covering note, plainly.

Common mistakes

Letting it write the introduction, which is where voice is established. Accepting its structure because the structure is reasonable. Trusting a citation because it is formatted correctly. Using it to sound more formal, which produces stiffness rather than authority. Assuming a silent policy means an open one. Pasting confidential or unpublished third-party material into a tool that trains on inputs.

Practical tips

• Keep a file of sentences you actually like. It is a better voice reference than any prompt.
• If a paragraph came back better, work out why and learn the move; do not just paste it.
• Ask "what have I not said that a reader needs?" rather than "improve this".
• Run a draft through BRYME's AI-writing checker to see which of your own habits already read as machine-made — the tells are learnable.