Make AI text sound like a person wrote it

It's rarely one thing. It's a stack of small habits models fall into, and once you can name them you can ask for them gone.

Throat-clearing openers. "In today's fast-paced world," "It's important to note that," "Let's dive in." Whole sentences that carry no information and exist only to warm up.

Relentless symmetry. Three examples every time. Every paragraph the same length. Every list exactly parallel. Real writing is lumpy — a two-word sentence next to a long one.

Hedging on everything. "Can often be," "may potentially help," "generally considered." Confidence is a voice; endless qualification reads as nobody home.

Vocabulary nobody says out loud. Delve, leverage, robust, tapestry, testament, navigate the landscape. None of these are crimes on their own. Six in a row is a signature.

Summary endings. The closing paragraph that restates what you just read and adds "ultimately, the key is balance."

No specifics. This is the big one. AI text describes categories; people describe Tuesday. Swap "improved engagement significantly" for "replies went from two a week to eleven" and half the problem solves itself.

The chat above will point these out on your own text if you ask it to — "what gives away that AI wrote this?" is a genuinely useful prompt.

Pasting text with "make this sound human" gets you a mild improvement. Pasting it with instructions gets you something you'd actually send.

1. Give it your voice to copy. Paste two or three paragraphs you wrote yourself — an old email, a Slack message, anything unedited — and say "match this voice." This single step does more than every other instruction combined, because now it's imitating a real person instead of guessing at one.

2. Ban your least favorite words. "Don't use delve, leverage, robust, or journey. No sentence may start with 'Moreover'." Models follow a banned list well.

3. Ask for specifics, not just style. "Where this is vague, ask me for the real number or example instead of inventing one." You'll get a short list of questions back, and answering them is what turns generic text into yours.

4. Set a reading level and a length. "Eighth-grade vocabulary, under 200 words, some sentences short." Constraints break the default rhythm.

5. Read it out loud. The last pass is yours. Anything you'd never say, change. That's the part no tool does for you — and it's also what makes the text honestly yours.

More on writing the instructions well: how to write better AI prompts, and the email-specific version in how to use AI to write better emails.

Worth being straight about, because it's why a lot of people land on pages like this one.

AI detectors are unreliable in both directions. They flag human writing — non-native English speakers especially — and they miss AI writing constantly. Nobody, including the companies selling detection, can give you a dependable score. So "will this pass" is a question with no honest answer, and any tool promising otherwise is selling you something.

What we'd actually suggest: treat this as a voice and clarity tool, not a laundering one. Rewriting a draft so it reads like you, adding real specifics, and doing a final human pass produces better writing and happens to remove most of the machine signature. That's a good deal either way.

And where AI use is disclosed or disallowed — a graded assignment, a journal, a workplace policy — follow the rule. This tool is for making your writing better, not for getting around someone's terms.