AI Writing Assistant — Tone-Adaptive Copy in Seconds

The system prompt is doing most of the work

The difference between this and typing the same request into a general chatbot is a tuned system prompt — and knowing what it forbids tells you exactly where to look when editing.

It bans specific cliché openers, including "In today's fast-paced world". It forbids connective filler such as "in conclusion" and "it is important to note". It requires every paragraph to earn its place, so a sentence that can be cut without loss gets cut. It pushes for concrete examples and numbers over vague assertion. And it demands the tone be matched exactly rather than approximately.

Those bans exist because they are the default failure modes — the things a model reaches for when left alone, and the things that make generated prose instantly recognisable. So if a cliché survives into your draft anyway, that is the first sentence to delete.

Tone is the input that changes output most

Five options — professional, casual, persuasive, inspirational and funny — and they are genuinely different registers rather than cosmetic labels. Professional is precise without stiffness. Casual uses contractions and shorter sentences. Persuasive builds toward a point. Inspirational lifts. Funny puts the observation before the punchline.

A mismatch between the tone you selected and the audience you actually have is the most common reason a draft feels wrong in a way you cannot name. Choosing deliberately at the start is worth more than rewriting afterwards.

The specificity problem

The model can only be as concrete as your prompt. Ask for something about "my career change" and you will get competent generalities, because nothing else was available to it.

Give it the actual situation, the real numbers and the specific claim you want to make, and the output improves out of all proportion to the effort. Vague in, vague out — and the fluency of the result disguises how little it actually said, which is the trap.

What to do with the draft

Read it as a structure rather than as prose. Keep the shape; replace the general claims with your own specifics; cut the opening paragraph if it is warming up rather than starting.

Then check anything factual. A model will produce a plausible statistic with complete confidence and no source, and confidence is not evidence. Note too that requested length is approximate — asking for 300 words gets you roughly that, not exactly.

Before you paste anything in

What you enter is sent to this site's server and on to a third-party model provider. It does not run in your browser and it is not private, so avoid pasting anything confidential, anything under NDA, or personal details belonging to other people.

For particular formats, the email writer and grammar fixer are tuned more narrowly.