Search "humanize AI writing" and you will find a hundred tools promising to make your text undetectable. Beat the detector. Slip past the scanner. Sound human enough that nobody can prove otherwise. That is the whole pitch, and it is the wrong goal.
The question was never whether a detector catches you. The question is whether you would put your name on the thing you are about to send.
People miss something when they copy from a chatbot and paste it straight into an email, a proposal, or a LinkedIn post. The moment it leaves your hands, it is not the bot's writing anymore. It is yours. The bot has no reputation to lose. You do.
Raw AI has an accent, and everyone can hear it
Large language models write in patterns. They reach for the same words, the same sentence shapes, the same tidy little structures, far more often than any person would. Once you learn to hear it, you cannot unhear it.
The em dash sprinkled through every other sentence. The phrase "it's important to note." Everything arriving in groups of three. Headings in Title Case. That breezy "Let's dive in" opener. A closing paragraph about how the future is bright. None of these are wrong on their own. Stacked together at machine frequency, they read like a uniform.
There is even a running catalog of these tells. Wikipedia editors maintain a page called Signs of AI writing, built from the cleanup work of spotting machine text in real articles. It is a useful mirror. Read it once and you will start catching the accent in your own drafts.
I know the list well, because I built the same discipline into my own publishing pipeline. Every article I run goes through a scan against a couple dozen of these patterns before it goes anywhere near a reader. It works. And it still is not enough on its own, which is the part worth sitting with.
The reflection is on you, not the machine
A clean scan tells you the text is free of the obvious tells. It does not tell you the text sounds like you.
Even after a scan comes back clean, I read the draft out loud, start to finish. That read still catches things no automated check can. A sentence that is grammatically perfect but that I would simply never say. A word that is a little too formal for how I actually talk. A rhythm that is smooth in a way real speech never is. The scanner sees patterns. Your ear hears whether it sounds like a person, and specifically whether it sounds like you.
Take a real example. I was cleaning up a batch of articles recently, and the scan came back clean on a sentence like "a good many of our members head north for the season." Grammatically perfect. Not a single AI tell in it. But say it out loud and something is off, because nobody actually talks like that. "A lot of us head north" is what a real person says. The scanner could not hear the difference. My ear could, and so will your reader's.
That gap is the whole point. When you skip the review, you are not shipping the bot's work. You are shipping your judgment, and the reader is judging it. A prospect reading a proposal that reeks of ChatGPT does not think the tool was lazy. They think you were.
This is already playing out in the open. Forbes recently covered how AI-written work is forcing a reckoning in academic publishing, and the smartest takeaway there is the same one that applies to your inbox. The fight over detecting AI is giving way to a simpler standard. Verify the substance, and own the words.
A tool worth knowing: blader's humanizer
If you work with AI agents, there is an open-source project worth adding to your kit. It is called humanizer, built by a developer who goes by blader, and it has picked up more than thirty-three thousand stars on GitHub. That is a lot of people agreeing this problem is real.
It is an agent skill, which means it plugs into tools like Claude Code and rewrites text on request. Under the hood it checks against thirty-three of the patterns I mentioned, the same lineage as the Wikipedia list. It runs the rewrite in two passes, then audits its own output for anything it missed.
One detail tells you the author actually understands the problem. The rewrite follows a strict no-fabrication rule. It never adds a fact, a name, a date, or a citation that was not already in your text. That single rule is the difference between humanizing and faking. A good humanizer makes your real point sound like you said it. It does not invent a more impressive point. The tools promising to make AI "undetectable" have that backwards, and it shows.
The project is MIT licensed and free. If you are technical, npx skills add blader/humanizer gets you started. If you are not, the principle still applies to you, and that is what the checklist below is for.
The pre-send checklist
You do not need a tool to do this well. You need a habit. Before anything an AI helped you write goes out with your name on it, run these eight steps. They take a few minutes and they save you from the version of yourself that hit send too fast.
- Read the whole thing out loud. Your ear catches what your eye skims. This one step does more than the other seven combined.
- Delete the throat-clearing. Any sentence that opens with "In today's," "It's important to note," or "At its core" can start one clause later and lose nothing.
- Hunt down the em dashes. AI loves them. Swap most of them for a comma, a period, or a set of parentheses.
- Verify every fact, name, number, and link by hand. This is the big one. AI states wrong things with total confidence. If you cannot confirm it, cut it.
- Break the rhythm. If every paragraph runs the same length, it looks generated. Add a short punch. Let one longer thought stretch out.
- Question the lists of three. AI groups things in threes on reflex. Ask whether the third item earns its place or is just filling out the pattern.
- Put the important parts in your own words. Find the generic phrasing and replace it with how you would actually say it to a colleague.
- Ask the real question. Would you put your name on this in front of the person whose opinion you care about most? If the answer is no, you are not finished.
Use the tools, keep the judgment
None of this is an argument against writing with AI. I use it every day, and so should you. It is fast, it is a genuinely good first-draft partner, and pretending otherwise is its own kind of denial.
The argument is against the paste-and-send reflex. AI writes the draft. The judgment stays yours, because the name at the bottom is yours. That is the trade, and it has not changed just because the drafting got easier. If anything it matters more now, when everyone has the same tool and the only thing separating your work from the generic version is the fifteen minutes you spent making it sound like a person wrote it.
The people who understand this are pulling ahead already. Not because they refuse to use AI, but because they refuse to send its first draft. I wrote a while back about why AI won't make your degree worthless, and this is the same idea from a different angle. The tool does not replace your judgment. It raises the price of not having any.
So run the checklist. Read it out loud. Then send it, and mean it.