Humanize Writing: The Skill That Makes AI Text Sound Like You

I can spot AI-written text in about four seconds. So can your clients, your boss, and everyone scrolling past your LinkedIn post. It's not one thing that gives it away — it's a cluster of small habits. The word "delve." The word "leverage" used as a verb three times in one paragraph. Sentences that all run the same length, like a metronome. Lists of exactly three things, every single time. A closing paragraph that starts with "In today's fast-paced world."

The frustrating part is that the underlying draft is usually fine. The facts are right. The structure makes sense. It just doesn't sound like a person — and it definitely doesn't sound like you.

That's the problem this skill solves. Humanize Writing is a small, self-contained instruction file you drop into your AI assistant. From then on, whenever you paste in a stiff draft — AI-generated or written-by-a-tired-human corporate — the assistant rewrites it in your actual voice, keeps every fact intact, and hands you a short change log explaining what it fixed and why.

That last part matters more than people expect. The change log turns every rewrite into a tiny writing lesson. After a few weeks of seeing "removed rule-of-three chain," "cut hedging from the opener," "shortened three sentences in a row," you start writing cleaner first drafts yourself. The skill quietly works itself out of a job. I consider that a feature.

What the skill actually does

Here's the workflow, step by step:

1. It asks for a voice sample. Before rewriting anything, the skill asks you to paste 200–500 words of writing that genuinely sounds like you — an email you were proud of, a post that got real replies, a Slack message where you explained something well. If you don't have one handy, it infers a reasonable professional voice from context and tells you it's doing so, so there's no silent guessing.
2. It extracts voice traits. From the sample, it builds a short profile: your typical sentence rhythm (do you write short and punchy, or long and layered?), your vocabulary register (plain, technical, playful), your warmth level, and how direct you are. This profile drives the rewrite — it's not a generic "make it casual" pass.
3. It strips the AI tells. The skill carries a banned list: delve, leverage, landscape, tapestry, crucial, moreover, furthermore, "in today's fast-paced world," excessive em-dashes, rule-of-three chains ("faster, smarter, and more efficient"), and hedging that says nothing ("it's worth noting that," "arguably"). Each one gets removed or replaced with something a person would say.
4. It preserves every fact. Names, numbers, dates, claims, commitments — nothing gets invented, softened, or dropped. This is a rewrite skill, not a summarizer. If a fact can't survive the rewrite intact, the skill flags it instead of quietly changing it.
5. It varies sentence length on purpose. Human writing has rhythm. Short sentence. Then a longer one that takes its time and lands somewhere specific. The skill checks for monotone rhythm and breaks it up.
6. It outputs three things: the rewrite, a short change log, and two alternate versions of the opening line — because the opener is where AI voice is most obvious and where you'll most often want a choice.

Who this is for

Professionals who use AI to draft but can't afford to sound like they use AI to draft. In practice that's:

• People writing work emails — where a stiff, over-formal AI draft reads as cold or evasive, and a humanized one gets an actual reply.
• Anyone posting on LinkedIn — where the audience is actively allergic to AI voice, and one "delve" in the first line kills the post.
• Consultants and managers writing reports — where the facts must survive untouched but the prose needs to sound like a person stands behind it.
• Founders and freelancers writing website copy — where "we leverage cutting-edge solutions" is indistinguishable from ten thousand competitors, and your actual voice is the only thing they can't copy.

If you write anything professionally and AI is somewhere in your drafting loop, this is for you.

Before and after

Here's a real-shaped example. The "before" is the kind of paragraph every AI assistant produces by default:

Before: In today's fast-paced business landscape, it's crucial for organizations to leverage effective communication strategies. Moreover, our team has delved into the quarterly results and identified several key insights. Furthermore, we believe these findings are important, valuable, and actionable — and it's worth noting that they could arguably transform how we approach Q3 planning.

Sixty words. Zero information a reader can hold onto, and at least seven AI tells: "fast-paced landscape," "crucial," "leverage," "moreover," "delved," "furthermore," a rule-of-three chain, and two hedges stacked at the end.

After: Quick summary of the Q3 numbers. Revenue came in 8% over plan, mostly from the two enterprise renewals we closed in May. Churn ticked up in the SMB tier — I think that's the pricing change, and I've put a one-page fix proposal in the deck. Worth 20 minutes at Thursday's meeting.

Fifty-two words, shorter than the original, and it actually says something. Notice what changed: the throat-clearing is gone, the vague "key insights" became specific numbers, the hedging became one honest "I think," and the sentences vary in length. That's what the skill does — and its change log would tell you each of those edits explicitly.

How to install it

1. Click the Download button on this page to get SKILL.md.
2. Create a folder for it in your AI tool's skills directory. For Claude Code, that's:

~/.claude/skills/humanize-writing/SKILL.md

3. That's it. Next time you say "humanize this" or paste a stiff draft and ask for a rewrite in your voice, the skill loads and runs the full workflow — voice sample, tells sweep, rewrite, change log, alternate openers.

The same file works in any tool that supports the skills format; only the folder path changes. No dependencies, no API keys, no setup beyond copying one file.

Make it a weekly workflow

A skill you use once is a toy. Here's how I actually run this one, and how I'd suggest you do too:

• Monday, 15 minutes: Batch-draft the week's outbound writing with AI — the LinkedIn post, the client update, the two emails you've been putting off. Don't polish. Just get drafts.
• Same session: Run each draft through humanize-writing. Because your voice profile is already established (save it in a note and paste it in, or keep it in your project memory), each rewrite takes under a minute.
• Read the change logs. This is the step people skip and shouldn't. Two minutes of reading "what got cut and why" per week compounds into genuinely better first drafts within a month.
• Refresh your voice sample monthly. Your writing evolves. Once a month, swap in a recent piece you're proud of so the profile tracks the real you, not the you from January.

Total cost: maybe 30 minutes a week. What you get back is every piece of writing you ship sounding like it came from you — because, after the skill's pass and your final read, it did.

FAQ

Does it work on text I wrote myself, not AI? Yes, and it's one of the best uses. Stiff corporate writing by humans has most of the same tells — hedging, filler, monotone rhythm. The skill doesn't care who wrote the draft.

Will it change my facts or numbers? No. Fact preservation is a hard rule in the skill, and the quality checklist forces a verification pass before output. If something factual is ambiguous, it flags it rather than guessing.

What if I don't have a good voice sample? The skill will infer a clean, plain professional voice and tell you it's inferring. It works — but a real sample works better. Even 200 words of your genuine writing beats no sample.

Is a humanized draft "undetectable" as AI? Wrong goal. The goal isn't to fool detectors; it's to write things your readers actually want to read, with your name honestly on them. You should always do a final read and own every word you publish. The skill gets you 90% there; the last 10% is you.

Can I edit the banned-word list? Please do. It's a plain markdown table inside SKILL.md. If your industry has its own zombie words ("synergy," "circle back," "best-in-class"), add them. The skill is a starting point, not a sealed box.

Does it handle languages other than English? The rewrite mechanics (rhythm, fact preservation, change log) carry over, but the banned list is English-specific. For other languages, replace the tells table with local equivalents.

Resources & links

• The humanize-AI-writing prompt, explained — the original prompt this skill grew out of, with the reasoning behind each rule.
• Building an AI marketing team in Claude — where humanize-writing slots in as the final polish step of a content pipeline.
• How I write SEO articles with AI — long-form drafting workflow that pairs naturally with this skill.
• A one-person business with AI in 30 days — the bigger picture: repeatable AI workflows for solo professionals.
• SEO Longform Writer skill — drafts the article; humanize-writing makes it sound like you.
• AI Marketing Team skill — a multi-agent content system that uses voice rules like these.
• Cover Letter Writer skill — tailored, human-sounding cover letters under 300 words.
• Context Engineering skill — how to feed AI the right context, including voice samples, so outputs improve everywhere.

If this kind of practical, no-hype AI workflow is useful to you, I send one like it every week: subscribe on Substack.