← BetterPost

Writing · July 22, 2026 · 6 min read

Best Claude Writing Skills: Which One Actually Writes Like You?

A vintage typewriter on a wooden desk, seen from above

Anyone shipping AI-assisted content in 2026 is running into the same wall: the words come fast, but they read like a machine wrote them. And now readers, platforms, and detection tools are getting better at spotting the difference. If you want copy that sounds like a person because a person still owns the hard parts, the tool you pick matters.

This is a practical look at what "best" means for Claude writing skills, why human-sounding output is now a measurable business concern, and how to evaluate your options against real-world detection.

Why "human-sounding" stopped being optional

The pressure isn't abstract anymore. Platforms are building AI detection directly into the writing experience.

This week, TechCrunch reported that Substack launched a feature integrating with Pangram, an AI writing detection tool, to estimate how much of a user's content is written by humans versus AI. Users can scan posts, comments, and replies in the Substack app, and the feature works for any post, note, reply, or comment above 100 characters.

Substack frames this as transparency rather than a ban. Writers can optionally add an "AI author's note," and publishers can run Pangram on their own drafts before publishing, then report and remove scans they believe are mistakes.

The underlying philosophy is worth sitting with. As Substack CEO Chris Best put it:

"Software should do everything else, but I think you do want the person to do the hard part."

That's the standard to write toward. The best Claude writing skills don't try to replace your judgment, voice, or argument. They handle the mechanical lift and leave the human fingerprint intact.

What actually makes AI copy sound human

Before comparing skills, it helps to know what detection tools and readers respond to. Generic AI output tends to share a few tells:

Human writing, by contrast, varies its cadence, commits to an argument, and carries specific detail. When you evaluate a Claude writing skill, you're really testing whether it preserves those qualities or flattens them.

The best Claude writing skills for copy that sounds like you, ranked

1. BetterPost — drafts that read like a person wrote them

This is the specific problem BetterPost was built to solve, so read the ranking with that in mind. It's a writing service your agent can call — a Claude skill and MCP server — whose drafts are designed to read like a person wrote them: varied structure, committed takes, and every claim grounded in real, dated sources you can check. Give it a topic and it writes the finished piece — newsletter, blog article, LinkedIn post, or an X or Bluesky thread — so the software carries the drafting load while you own the argument and the voice.

2. Dotdigital Content Agent — drafting where your emails live

Dotdigital's mid-July update, detailed on the Dotdigital blog, added a Content Agent that drafts and edits content directly in the email editor. It's scoped to email, but living inside the editor means no copy-paste round trips, and pairing it with the same update's Segment Agent keeps the writing pointed at a real audience.

3. Plain Claude with a voice brief — the free baseline

Before adopting anything, know what Claude alone can do: feed it three samples of your writing, your angle, and your sources, and ask it to match your cadence. It's the baseline every skill above should beat — and if a paid tool can't out-write a well-briefed chat session, it doesn't belong in your stack.

How to test a writing skill against your own voice

Use a consistent rubric so you're comparing tools, not vibes. Here's a repeatable process:

  1. Give the same brief to each option. Same topic, same audience, same word count. Differences in output become obvious fast.
  2. Read the first three sentences aloud. If they sound like a press release or a chatbot preamble, that's a red flag.
  3. Check sentence variety. Count openings and lengths across a paragraph. Real writers vary both.
  4. Look for a point of view. Does the draft lead with a take, or does it just summarize?
  5. Run it through a detector. Since platforms like Substack now surface Pangram scores, test your draft the way your readers' platform will.

That last step is the honest one. If a skill claims human-sounding output, put it up against the same detection your audience will encounter.

Human-sounding copy is also a personalization problem

Sounding human isn't only about sentence rhythm. It's about relevance. Copy reads as more human when it's clearly written for a specific audience, and Dotdigital's own guidance captures the reader expectation:

Customers want to feel like you know them personally.

So the best writing setup pairs a human-sounding drafting tool with genuine audience targeting. A well-written email to the wrong list still feels generic. The right message to the right segment feels personal even before you polish a word. (For the operator side of that pairing — segmentation, campaigns, workflows — see our ranking of the best Claude marketing skills.)

Putting it together: a workflow that survives detection

Here's a workflow that respects both the technology and your readers:

This is the same principle Substack is nudging its ecosystem toward: use AI for the mechanical work, keep the human on the hard part, and be honest about the mix.

Takeaways and next steps

The "best" Claude writing skill is the one that leaves your voice intact and holds up when a platform scores it. Given that detection is now built into tools like Substack, that's no longer a nice-to-have.

Start here:

Write like a person, target like you know your reader, and stay transparent. That combination outlasts any single detection update.

FAQ

How do I write better LinkedIn posts with Claude?

Give Claude your angle, your key points, and real sources instead of a bare topic — or have your agent call a writing service like BetterPost, which finds timely, dated sources and writes the LinkedIn post end to end. Either way, edit for cadence and keep the take yours; a post with a real point of view reads human.

Which Claude skill writes the most human-sounding copy?

We're biased, but human-sounding output is the exact problem BetterPost was built for: varied sentence rhythm, committed takes, and drafts grounded in real, dated sources. The honest test for any option is to run the same brief through each and score the results with the same detector your readers' platform uses.

Can Claude write a newsletter that doesn't sound like AI?

Yes — when the draft is grounded in real sources and written for a specific audience. The tells that give AI away are uniform rhythm, filler intensifiers, and hedged non-takes, so ground the draft, commit to an argument, and edit the cadence. Detection tools like Pangram respond to those same signals.

How do I make Claude write like me?

Feed it three or four samples of your actual writing, name the habits you want kept (sentence length, how you open, what you refuse to say), and give it your angle before it drafts. Or use a writing skill that commits to a point of view by default and edit from there — matching voice is much easier when the draft already argues something.