Keeping a Human Voice with AI: What to Review, What to Rewrite

Check the facts, rewrite the opener, add a detail only you know: the method for making AI-written posts sound like you.

By the Oposto team · Published on October 9, 2026 · Content creation

To keep a human voice when you use AI, let it write the first draft and hold on to three jobs yourself: check the facts, rewrite the opening line and add one detail only you know. The rest is handled upstream, with a solid brief and a list of your words (the ones you use and the ones you never use), and downstream, by reading the post out loud before it goes live.

This guide covers what to review, what to rewrite and how to do it quickly, using Birchwick, a fictional coffee roaster, as the example. It builds on our guide to automating your social media: automating production only pays off if what comes out still sounds like you.

Why does AI-written copy sound like AI?

Because a language model with no specific direction produces the most likely version of a text: the most expected one, which makes it the most generic. When you read it, a handful of recurring tics give it away:

• One-size-fits-all openers: "In today's fast-paced world…", "Have you ever wondered…", "What if…"
• Adjectives in threes: "a rich, bold and refined coffee."
• Abstract words where a concrete thing belongs: "a one-of-a-kind experience," "unmatched craftsmanship."
• Mirror-image structures: "It's not X, it's Y," repeated post after post.
• A preachy wrap-up: "Because at the end of the day, what really matters is…"
• No verifiable detail at all: no name, no number, no place, no date.

None of these is off-limits on its own. It's the pile-up that makes a post feel like nobody wrote it. We break down the tics specific to LinkedIn in writing a LinkedIn post that doesn't sound like AI.

Do platforms penalize AI-written posts?

Not for being AI-written. Google explains in its guidance on AI-generated content that it judges content on quality and helpfulness, not on how it was produced. YouTube, on its page about disclosing altered or synthetic content, says that using AI for an outline, a script, a title or captions doesn't require any disclosure.

What needs to be disclosed is realistic content that could mislead: YouTube asks creators to disclose a video that makes a real person appear to say or do something they didn't, or shows a realistic scene that never happened. TikTok asks creators to label AI-generated content that shows realistic scenes, according to its help page, and Meta describes its "AI info" label on Facebook, Instagram and Threads.

So the real risk isn't a platform penalty. It's your followers tuning out a post that could have come from any brand.

What should you give the AI before it writes?

A human voice starts before anything is generated. Three inputs do most of the work:

1. A precise brief: the goal of the post, who it's for, what they should take away, the concrete detail to include. Our method to write a good brief for an AI content tool goes deeper.
2. Your written identity: how casual you are, sentence length, how much humor, topics you never touch. That's what we call your brand DNA.
3. A house word list: about ten words you actually use and ten you refuse to use.

For Birchwick, the list looks like this:

What should you review every single time?

Reviewing is first about never publishing something false. Five checks, every time:

1. Facts: prices, dates, hours, addresses, product names, ingredients. AI can fill a gap with information that sounds right and isn't.
2. Promises: "ships tomorrow," "best in town," "guaranteed." Keep only what you can deliver.
3. Numbers: each one should come from you or from a source you can cite.
4. Tone: no sudden slang, no unusual stiffness.
5. Sensitive topics: health, money, the news, competitors. A harmless-looking post can land badly on a particular day.

What should you rewrite, and how?

You don't have to redo everything. Three rewrites are enough to make a post sound human:

1. The first line

It's the one people read, and the one AI most often makes generic. Swap the mood-setting sentence for a fact, a scene or a specific question. "In today's fast-paced world" becomes "Tuesday, 7:12 a.m.: first batch of the Honduras just came out of the roaster."

2. One detail only you know

A first name, a number from your business, something that happened today, a mistake you fixed. That's what no model can make up without lying, and it's what proves a person is behind the account.

3. The words from your list

Replace every phrase from the "we never say" column with its concrete equivalent, and cut anything that says nothing. A shorter post almost always sounds more human.

What does a full rewrite look like?

First draft: "In today's fast-paced world, take time to savor the moment. Our new Honduras coffee offers a one-of-a-kind experience: rich, bold and refined. Stop by and discover it today!"

After review and rewrite: "Tuesday, 7:12 a.m., Maya pulled the first batch of the Honduras. In the cup: cocoa, a little apple, not much bitterness. 40 twelve-ounce bags this week, whole bean or ground for your drip machine. Come try it, there's always one on pour-over."

Same topic, roughly the same length. But the second version has a time, a name, three flavors, a quantity and an invitation that could only come from Birchwick. The numbers in this example are made up for illustration; in your posts, they need to be real.

How do you review quickly without losing your afternoon?

A ten-minute routine per post is usually enough:

1. Read the post out loud. Wherever you stumble or cringe, there's a sentence that doesn't sound like you.
2. Run through the five review checks.
3. Rewrite the first line.
4. Add one detail only you know.
5. Cut one sentence in five.

If several people on your team post for the brand, share the word list and the five checks with all of them. Consistency doesn't come from one talented writer; it comes from everyone reviewing against the same short list, so a post written on a Tuesday by one person sounds like the one written on Friday by someone else.

Over the weeks, keep track of the fixes you make most often and add them to your brief and word list: the AI will get it wrong less, and you'll review faster. To decide how much of the rest to hand off, see can AI run my social media for me.

A voice that stays yours with Oposto

Oposto starts from your website to learn your brand: voice, colors, products, audience. That's its content DNA, which you can add to and correct, and it guides every post, carousel or short video it drafts. Nothing goes out without your approval, and each post is analyzed before publishing, with a score and recommendations. You keep the final read; it saves you from the blank page. The free plan requires no credit card: try Oposto.

Frequently asked questions

Do you have to disclose that a post was written with AI?

For text, none of the platforms mentioned here require it: YouTube, for example, says help with writing a script or captions doesn't need to be disclosed. Labeling rules mainly target realistic images, audio and video that could mislead people.

How long should it take to review an AI-generated post?

About ten minutes is usually enough: read it out loud, check the facts, rewrite the first line, add a personal detail and make a few cuts.

How do you keep all your posts from sounding the same?

Give each brief a different concrete detail (a product, a person, a moment, a number) and vary your openers. A house word list also keeps the same stock phrases from creeping back in.

Can AI imitate my writing style?

It gets close if you give it samples of your writing, your word list and your tone rules. It can't invent your stories or your numbers: those are yours to add.