Your 'Make This Human' Prompt Isn't Working. Here's What to Actually Tell AI.

Say you're a freelance content writer. A client asks for a blog post, you draft it with ChatGPT's help, then — out of caution — you paste it back in and type: Make this sound more human. ChatGPT obliges. You hand it in. The client runs it through an AI detector anyway. It still flags.

The problem isn't that AI can't humanize text. The problem is that how you ask matters enormously — and most people are asking wrong. Here's a Q&A with a content strategist who works with writers navigating this exact issue every week.

Why Does "Make This Sound Human" Usually Fail?

Q: I keep telling ChatGPT to humanize my text and it still gets flagged. What's going wrong?

A: The phrase "make this sound human" is too vague. It doesn't tell the model which features to change — and AI detectors aren't checking for "humanness" in some general sense. They're looking for specific statistical patterns: uniform sentence length, predictable structure, overuse of transitional phrases, and a kind of relentless logical flow that human writers rarely maintain.

When you just say "humanize this," the model might swap a few words or add some hedging language. But the underlying architecture of the text often stays intact. You can see exactly what detectors are scanning for in this breakdown of how AI detectors work.

What Should I Actually Say to Get AI to Humanize Text?

Q: Okay, so what's a better prompt?

A: You need to give the model a checklist of specific structural changes — not a vague style goal. Think of it as handing an editor a list of notes. Here's what tends to produce more useful results:

• Vary sentence length aggressively. Ask it to mix very short sentences (under eight words) with longer, more complex ones. AI text gravitates toward a suspicious middle ground.
• Break the paragraph rhythm. Tell it to occasionally start a new paragraph with a single sentence, or even a fragment. Like this one.
• Cut the transition words. Phrases like "in addition," "it is worth noting," and "this is important because" are AI fingerprints. Ask it to remove them and restructure the logic without them.
• Add a concrete detail. Ask it to insert one example, analogy, or observation that grounds an abstract point. Human writers do this instinctively; AI often skips it.
• Shift the voice. Ask it to write as if the author is a skeptic who had to be convinced, rather than an authority presenting conclusions.

Even with all that, results vary. The model is still producing output based on its own patterns, and sometimes the rewritten version just trades one AI fingerprint for another.

Can AI Actually Humanize Its Own Text?

Q: That's what I keep wondering — can AI ever really humanize AI text?

A: It can shift things meaningfully, but there's a ceiling. With specific instructions, the output does change. But you're still asking a language model to disguise the statistical properties that language models produce. The model doesn't get tired, doesn't make the small missteps that characterize human writing, and doesn't have genuine opinions that color its word choices.

That's also part of why AI detection false positives affect real human writers too — detectors are pattern-matching, not intent-detecting. The reverse holds as well: AI text can sometimes survive detection not because it's truly human-sounding, but because the patterns happen to fall outside a particular detector's training distribution.

When Should I Stop Prompting and Use a Dedicated Tool?

Q: At what point is it worth just using a purpose-built tool?

A: When the stakes are high and you can't afford to guess. Prompting ChatGPT to rewrite itself is manual and trial-and-error. You might run it three times and still have no idea whether it'll pass. A dedicated humanizer like WriteMask is built specifically to modify the distributional features that detectors target — that's a fundamentally different intervention than asking a general-purpose model to "sound more human."

WriteMask reports a 93% pass rate across major AI detectors. You can also run your text through the free AI detector before and after to see whether your changes are actually moving the needle — or whether you're just spinning your wheels.

For a step-by-step walkthrough of this process with ChatGPT-generated text, the guide on how to humanize ChatGPT for Turnitin covers it in practical detail.

Any Final Advice?

Q: What's the one thing most people get wrong here?

A: They treat "humanize" as a single command rather than a set of targeted edits. Specific prompts do better than vague ones. And if you're under time pressure or dealing with a high-stakes submission, a purpose-built tool skips the guesswork entirely. There's no prize for doing it the hard way.