How to Add Voice and Opinion to Flat AI Copy

Most AI copy isn't wrong. It's just empty. Sentences arrive in the correct order, transitions are logical, the paragraphs cover the topic. And yet something is missing. You can feel it. The text could have been written by anyone about anything, and nobody takes responsibility for a single word.

That gap between grammatically correct and actually readable is where voice lives. The good news is you can add it back in, even to a draft that's already been generated. This guide shows you how.

What "voice" actually means in practice

People throw around "voice" as though it's a mystical quality. It isn't. Voice is the accumulation of small choices: how long a sentence runs before it stops, whether you use "it's important to note" or just say the thing, whether you're willing to take a position or hedge everything into meaninglessness.

AI models default to the hedge. They've been trained on enormous amounts of corporate communication, academic writing, and content-farm prose (all genres that reward caution over clarity). So the output sounds like a committee wrote it, because the training data contains thousands of committees.

Here are the concrete things that make writing feel like a person produced it:

• Sentence variety that's actually varied. Not just mixing long and short, but varying the structure. Sometimes a fragment. Sometimes a question. Sometimes a sentence that starts with "And."
• Positions, not observations. "There are many perspectives on this" is an observation. "This approach tends to backfire" is a position. Readers trust positions.
• A recognizable register. Formal, casual, dry, warm: pick one and hold it. AI writing often slips between registers mid-paragraph, which is why it feels slightly off even when you can't name why.
• Specificity over breadth. The first thing AI does when it doesn't know something is generalize. Real writers narrow down; they give you the one example that proves the point instead of gesturing at a category.

The most common reason AI copy feels flat

It's not vocabulary or grammar. It's the absence of a perspective.

When you ask an AI to write about, say, remote work productivity, you get a balanced, even-handed summary of what various people think. When a real person writes about remote work productivity, they have a take. Maybe they think async communication is overrated. Maybe they've found that shorter work days help more than any tool. That take is what makes the piece worth reading.

AI copy is flat because it has no stake in the subject. It doesn't prefer one approach over another. It isn't trying to change your mind.

Your job when editing is to supply the stake. What do you actually think about this topic? Where would you push back on conventional wisdom? Even a mild opinion, clearly stated, transforms the reading experience.

A before/after example

Here's a real pattern you'll see in AI drafts. The topic is email newsletters.

Before (AI default):

Email newsletters can be an effective way to communicate with your audience. Many businesses use them to share updates, promote products, and build relationships with customers. It's important to send newsletters on a consistent schedule and to provide value in each edition. Subject lines should be compelling to improve open rates. Including a clear call to action helps readers know what to do next.

Nobody said anything there. It's five sentences of things anyone could have said, in the order anyone would have said them. There's no personality, no preference, no reason to keep reading.

After (adding voice and opinion):

If your email newsletter reads like a company announcement, your subscribers are deleting it. People subscribe to newsletters because they want a shortcut to someone else's thinking. They're hoping you've already sorted through the noise and can just tell them what matters. That means your job isn't to "provide value" on a schedule. It's to have a point of view and share it, consistently, in your own voice. Subject lines, CTAs, send frequency: all of that is secondary.

Same topic. Roughly the same length. But now there's a person in the room with an opinion about what the reader is probably doing wrong.

If you want a shortcut for this kind of transformation, the /humanizer-prompt gives you a structured prompt you can paste directly into your AI tool to push for more opinionated rewrites.

Practical techniques to add personality to any draft

Replace "it's important to" with the actual point

"It's important to consider your audience" is a non-sentence. It doesn't say anything specific, and it wastes the reader's time. Delete the setup and write the point: "Your audience has probably heard every generic productivity tip. Skip those."

Cut the throat-clearing

AI drafts often open with a sentence that explains what the article is about to do, instead of doing it. "In this article, we will explore..." starts about 40% of AI-generated pieces. Delete that sentence. Start with the substance.

Add a "this is what I actually think" sentence

At some point in the piece, write a sentence that starts with "I" and takes a position. If you're editing for a brand or client, translate it into brand voice. But the instinct is the same: stake a claim somewhere, even a small one.

Use more specific numbers and examples

"Many people struggle with this" becomes "Most first-time managers I've talked to make this mistake in their first six months." Specificity isn't just clearer. It feels more honest. When a writer uses specific numbers or examples, the reader's brain registers: this person has been somewhere.

Let sentences end

One of the most reliable tells in AI writing is the sentence that keeps going with "-ing" clauses after the main thought is already done. "You can improve your copy by focusing on clarity, making sure each sentence adds value, and avoiding unnecessary filler that pads the word count." Cut it after the first clause. "You can improve your copy by focusing on clarity. Each sentence should earn its place."

For a deeper look at this and the other structural patterns that give AI writing away, see a simple checklist for removing AI tells from any draft.

How to add voice when you're editing for someone else

This is the harder problem. You can't inject your own opinions into a client's copy. The voice has to be theirs, even if they handed you an AI draft to fix.

A few approaches that actually work:

Ask them to react to a position. Send the client the draft and ask: "Is there anything here you'd push back on, or a place where you'd say it differently?" Their answer is where the voice lives. Pull it in verbatim if you can.

Listen to how they talk. If you've had calls with the client, pull up your notes. People speak with far more personality than they write. The phrases they use in conversation are almost always better than anything in the AI draft.

Find one place to take a stand. You don't need the whole piece to be opinionated. One paragraph where the client's perspective is clearly stated does more work than five paragraphs of balanced hedging.

Match their vocabulary, not the AI's. If your client says "screw it, just ship it" in Slack, that's a voice. The AI draft probably says "it is often preferable to prioritize action over perfection." Translate back.

For a more detailed walkthrough of the editing process, see how to edit an AI draft so it reads like a human wrote it.

When to rewrite from scratch instead of editing

Sometimes the flat AI draft isn't salvageable with line edits. The structure is wrong, the examples are generic, and there's no one position you can locate and amplify. In those cases, editing is slower than starting over.

The sign you've hit this point: you're rewriting more than half the sentences, and the piece still isn't working. That's not editing; that's drafting with extra steps.

When this happens, use the AI output as research notes. Pull the facts and examples you want to keep, then write your own version from scratch. This is faster than it sounds once you have a usable outline. See how to rewrite a robotic AI paragraph by hand for what that process looks like at the paragraph level.

---

FAQ

How much should I change an AI draft to give it voice?

There's no percentage, but the goal is readable, not original. If a sentence already says what you'd say, in roughly the way you'd say it, leave it alone. Focus your edits on the places that feel weightless: where the writing covers the topic without taking any position on it.

Can I ask the AI to write with more voice upfront?

Yes, and it helps. Prompting the AI to write from a specific perspective, or to take a clear stance on a question, produces better raw material than asking it to write a "comprehensive" or "balanced" overview. The /humanizer-prompt is built around this principle: give the model constraints that force it toward a position.

Why does AI writing always feel balanced and neutral?

Reinforcement learning from human feedback (RLHF) rewards responses that feel safe and acceptable to a broad audience. Opinions and strong takes tend to generate disagreement, which looks like a negative signal during training. The model learns to hedge. This is useful in many contexts, but it makes the writing feel like no one is home.

Is adding voice the same as adding personality?

Mostly. "Voice" describes the pattern of choices across a piece of writing. "Personality" is more about tone. You can have a dry, minimal voice or a warm, conversational one; both are legitimate. The problem with AI copy isn't that it has the wrong personality. It's that it has no observable preferences at all.

Does adding opinion make writing less credible?

It makes it more credible, not less. Hedged writing signals uncertainty; clear positions signal that someone has thought the topic through and arrived somewhere. Readers sense the difference. The caveat is obvious: the opinions need to be defensible and accurate. But credibility doesn't come from avoiding positions. It comes from taking ones you can back up.