How to ground content in real research, draft in your actual voice, and audit out the tells so your writing sounds human and passes detection.
You can feel it within two sentences. The same punctuation tic in every paragraph. The "it's not just X, it's Y" construction, three times on one page. The throat-clearing intro that begins "In today's fast-paced world." The rule-of-three that repeats under every heading. None of it is wrong, exactly. It's just generic, and readers have learned to recognize generic as a signal that nobody really wrote this.
That recognition is the AI writing trap, and it's costing people trust, rankings, and replies.
Here's the reframe most advice misses: AI writing sounds like AI because it writes like everyone. Ask a model for "a blog post about X" with no other context and it returns the statistical average of everything ever written about X, which, by definition, sounds like no one in particular. The fix isn't a magic "humanize" button. It's giving the AI the specifics it's missing: real sources, your actual voice, and a final pass that strips the tells.
Why AI Writing Gets Detected (and Distrusted)
Before the fixes, it helps to name the actual tells, because "sound more human" is useless as an instruction. Detectors and discerning readers flag the same things:
Vocabulary tics. "Delve," "leverage," "robust," "tapestry," "testament to," "navigate the landscape."
Structural uniformity. Every paragraph the same length, every section the same shape, the rule-of-three on repeat.
Formatting habits. The same punctuation overused everywhere, bold abuse, "it's not just X, it's Y."
Vague attribution. "Studies show," "experts agree," "many believe" with no actual source.
Zero specificity. No real numbers, no named examples, no opinions a person would actually risk having.
Notice that the last two aren't style problems. They're substance problems. The most human-sounding writing is specific: it cites a real study, names a real example, takes a real position. Which is exactly why research and voice come before the cleanup pass.
Step 1: Ground It in Research (Specificity Is the First Human Tell)
Generic content reads as AI because it's built on nothing. "Remote work has changed how teams operate" is the kind of sentence a model produces when it has no facts to work with, and a reader skims right past it. Real writers anchor claims to specifics: a number, a study, a named company, a dated event. That specificity is the single biggest difference between writing that sounds informed and writing that sounds auto-generated.
A research-first writing skill builds the article on a foundation of verified sources. It researches the topic, pulls current statistics and expert perspectives, and weaves cited sources into the narrative so the draft makes specific, defensible claims instead of vague gestures.
"Research and write an article about the future of remote work. Use current statistics and real expert perspectives, cite every claim with a credible source, and replace any vague generalization with a specific, attributed fact. I want named examples, not 'studies show.'"
The discipline this enforces: never let a claim go in without a source behind it. When the substance is specific, half the AI tells disappear on their own, because "experts agree" becomes "a 2025 Stanford study of 1,200 hybrid workers found," and no detector mistakes that for filler.
Best for: content marketers, journalists, analysts, founders, researchers.
Step 2: Draft in Your Voice (Teach the AI How You Write)
This is the step that separates "teach the AI how you write" from "let the AI write for you." The default model voice is an average. Your voice is a specific set of choices: sentence rhythm, the jokes you make, the words you'd never use, how much you hedge, where you get blunt.
A writing skill captures those choices so the draft starts in your register instead of starting generic and getting edited toward it. The leverage comes from how you configure it. Tell it your tone, give it a sample of your past writing to match, name the phrases you ban, and the draft arrives sounding like you rather than like a content mill.
"Write a blog post about intermittent fasting. Match the voice in the two sample posts I'm pasting below: direct, slightly skeptical, short paragraphs, no throat-clearing intros. Audience: busy professionals who've tried diets before. Banned words: delve, leverage, robust, journey. Open with a specific scene, not a generalization."
The principle is the same across long-form articles, landing pages, ad variants, and social posts: the more of your specifics you give the skill up front, the less it falls back on the generic average. Voice isn't something you add at the end. It's an input you supply at the start.
Before: "Write a blog post about X." You get the statistical average of the internet: correct, generic, unmistakably AI.
After: A draft in your rhythm, with your banned words gone and your angle baked in. 80% of the way to publish, in your voice from the first line.
Best for: bloggers, content marketers, founders, developer advocates, growth marketers.
Step 3: Audit Out the Tells (The Final Pass Before You Publish)
Even a well-researched draft in your voice will pick up AI patterns. They creep in around the edges, in transitions and summary sentences and the third bullet of every list. The last step is a deliberate audit specifically for those tells, not a vague "make it better" pass. This is the step most people skip, and it's the one that decides whether the piece reads as human.
An AI pattern removal skill audits a draft for the exact patterns detectors and readers catch: overused punctuation, promotional language, formulaic structures, vague attributions, "it's not just X, it's Y" constructions. It flags each instance with an explanation, and rewrites to remove them while preserving your meaning. Crucially, it gives you a summary of what changed and why, so you learn the patterns instead of depending on the tool forever.
"Audit this draft for AI writing patterns. Flag vocabulary tells, structural uniformity (paragraph lengths, rule-of-three), formatting habits, and any vague attributions. Rewrite to remove them while keeping my argument and voice intact, and give me a short summary of what you changed and why."
Run this as the required last step before anything publishes. Over time, the "what changed and why" summaries retrain your own instincts. You start writing cleaner first drafts because you've internalized the patterns, which is the real goal. The skill that helps you write like a human is also the skill that teaches you to.
Best for: content marketers, bloggers using AI drafts, technical writers, students, comms teams.
The Human-Writing Workflow: Research, Voice, Audit
The three steps form a sequence, and the order matters. You can't audit your way out of a draft that had no substance or voice to begin with:
- Research: ground the piece in real, cited specifics. Substance is the first human tell.
- Voice: draft in your configured voice, with your samples, tone, and banned words as inputs.
- Audit: final pass that strips the tells and teaches you the patterns as it goes.
You don't need all three for every piece. A quick post might only need the draft and the audit. A flagship article earns the full research-first treatment. But the principle holds across all of them: you supply the specifics (sources, voice, opinions) and the skill handles the labor. That's the difference between AI as a content blaster and AI as a writing partner that knows how you write.
The Honest Caveat
"Passes AI detection" is a useful side effect, not the goal. Detectors are unreliable enough that chasing a green checkmark is a losing game. The durable goal is writing that's genuinely worth reading: specific, sourced, and written in a voice a real person would recognize as theirs. Do that, and the detection question mostly takes care of itself, because the reasons content gets flagged as AI are the same reasons it's boring. Fix the boring and you've fixed both.
Getting Started
The trap isn't using AI to write. It's using it generically. Give it your sources, your voice, and a real opinion, then strip the tells on the way out. That's how you write like a human with AI in the loop instead of in the driver's seat.
I publish free skills for every step of this workflow at claudecodehq.com: research-first writing, voice-configured drafting, and AI pattern auditing. Each one is a single file you drop into a folder. No coding, no subscription. Start with the pattern remover on your next draft and see what it catches.
Originally published on claudecodehq.com
Top comments (0)