How to Tune a Humanizer Prompt to Your Own Voice

Generic humanizer prompts do one job well: they strip the most obvious AI patterns from a draft. What they don't do is replace those patterns with your patterns. The result is prose that no longer sounds like a bot but doesn't quite sound like you either. It sounds like a competent stranger.

This guide is about closing that gap. You'll learn how to build a voice guide that any model can actually use, how to feed it into a humanizer prompt without making the instructions a mess, and how to test whether it's working.

Why "write in my voice" never works on its own

Adding "write in my voice" to a prompt is almost useless without specifics. The model has no voice file to pull from. It guesses, which means it defaults to the statistical center of everything it was trained on: medium-length sentences, three supporting details, an upbeat sign-off, and transitions like "furthermore."

What the model actually needs is a reference document. Not instructions telling it to be casual or formal. A document showing it what you sound like in practice, with real examples from things you've already written.

This is what prompt engineers call a voice guide, and once you have one, you can use it across tools, models, and sessions without rebuilding from scratch each time.

Building your voice guide

A voice guide is a short document (usually 200-400 words) you write once and reuse. It has three parts.

Part 1: Sentence rhythm examples

Pull 3-5 sentences from your own writing that you think sound distinctly like you. Don't pick your most polished lines. Pick the ones that feel natural when you read them aloud. Paste them verbatim under a heading like "My sentence patterns."

The model will infer rhythm from these. If your sentences tend to run long and then cut short, it will notice. If you habitually start sentences with "Look" or "Here's the thing" or drop the subject in short commands, it picks that up too.

Part 2: Vocabulary preferences

List 10-15 words or phrases you actually use, and 5-10 you never use. Be honest about both. If you always write "I think" instead of "I believe," note that. If you'd never write "it's worth noting" or "this allows us to," that goes in the avoid column.

Keep this section tight. A long word list gives the model too many signals to weigh and you'll get inconsistent results.

Part 3: Structural habits

Do you open with a question? Use short paragraphs? Avoid bullet lists? Prefer to make one claim at a time rather than stacking three? Write these down as plain observations, not rules. Something like: "I rarely use bullet lists. When I do, the items are usually short fragments, not full sentences."

One or two sentences per habit is enough.

How to feed the voice guide into a humanizer prompt

You have two options.

Option A: Inline it. Paste the voice guide directly into your humanizer prompt above the main instructions. This works well if you're using a chat interface. The downside is that longer guides eat into your context window and can dilute the editing instructions.

Option B: Use a system prompt. If you have access to a system prompt (in Claude, in the API, in a custom GPT), put the voice guide there and keep the humanizer prompt focused on editing tasks. The model treats system-prompt content as persistent context rather than one-off instructions, which typically produces more consistent results.

The free humanizer prompt at /humanizer-prompt is designed to accept a voice guide in either position. You can paste it in above the prompt text or reference it in a separate system message.

Before/after: what voice tuning actually changes

Here is the same paragraph run through a generic humanizer prompt, then through a voice-tuned version. The source content is identical; only the instructions differ.

---

Generic humanizer output:

When writing emails to clients, it's important to maintain a professional yet approachable tone. This ensures that your message is well-received and fosters a positive working relationship. Consider your audience and adjust your language accordingly.

---

Voice-tuned output (with a guide specifying: short sentences preferred, "I think/find" over passive constructions, no "it's important to" framing, direct second person):

Client emails don't need to be formal to be professional. I find a short, direct message lands better than a careful one. Just say the thing, then give them one clear next step.

---

The tuned version isn't objectively better. It's better for that specific writer. Someone whose voice runs more formal and thorough would configure the guide differently and get a different result. That's the point.

For more on the underlying principles, see the best prompt to make AI writing sound human and how to write a system prompt that strips out AI tells.

Testing whether your voice guide is working

Run the same source paragraph through three times. If the outputs are structurally similar and sound like the same person wrote them, the guide is giving the model enough signal. If each output sounds different or defaults to generic patterns, the guide needs more specifics.

A useful test: show the output to someone who knows your writing well. Don't tell them it's AI-assisted. Ask if it sounds like you. If they hesitate, go back to the voice guide and look for what's missing.

One thing to check specifically: does the model drop your voice guide partway through longer pieces? This happens when the guide is buried in a long prompt and the model loses the thread. The fix is to shorten the guide (prioritize what's most distinctive) or move it to a system prompt where it stays in scope.

You can also look at the article a ChatGPT prompt that makes copy sound like you wrote it for a parallel approach if you work primarily in ChatGPT.

Maintaining and updating your guide

Your voice guide is a living document. If you start noticing that AI-assisted drafts don't match how you're actually writing now, the guide is probably stale. Writers change. Audiences change. What sounded like you two years ago might not anymore.

A useful signal: when you find yourself editing large portions of the AI output rather than small patches, the guide has drifted from your current voice. That level of editing is the guide's job, not yours.

A simple habit: after any piece you're genuinely happy with, take five minutes to check whether your guide still fits. If you wrote three things that contradict it, update the guide rather than fighting the instinct. The whole point of personalizing the prompt is to reduce editing time. If you're still spending an hour fixing the output, something in the setup isn't calibrated right.

Some writers keep two versions of their guide: one for professional contexts (reports, client copy, formal proposals) and one for personal or editorial writing. The distinction can be as small as a note about register and sentence formality. You don't need a completely different document, just a few lines noting how the tone shifts.

---

FAQ

How long should my voice guide be?

Short enough that you can read it in under a minute. That usually means 200-350 words. Longer guides introduce contradictions the model has to resolve, and it will resolve them in ways you don't expect. If you're finding it hard to keep it short, prioritize the three things most distinctive about your writing and cut everything else.

Can I use my voice guide with any AI model?

Yes, though results vary. Models with longer context windows handle voice guides more consistently because they don't lose the reference mid-generation. If you're using a model with a shorter window or a free tier with token limits, keep the guide as tight as possible.

Does a voice guide work for different content types?

To a point. A guide built from your email writing will produce decent emails but might not translate cleanly to long-form articles or social posts. If you write across very different formats, consider keeping separate guides or adding a short format-specific note at the top of each prompt.

What if I don't have enough of my own writing to sample?

Start with what you have, even if it's a handful of emails or a few short posts. The voice guide doesn't need to be comprehensive to be useful. Even three strong sentence examples will move the output closer to your natural patterns than a prompt with no examples at all.

Will this guarantee that AI detectors won't flag my content?

No. A well-tuned voice guide improves the naturalness of AI-assisted writing, but we can't make predictions about any specific detector's algorithms or thresholds. The goal here is copy that reads well to actual humans, not one that games any particular tool.