PARTLY

Partly true: building a reusable 'humanizer' skill with Claude's skill-creator is a real workflow, but the '100% human' detector badge proves nothing, there is no 'Claude watermark' to strip, and the prompts are behind a comment hook

Facebook · Oct 2026

The claim'Your Claude responses still sound like AI because you're missing this one setup.' Tell Claude to humanize everything, have it research the patterns AI detectors and readers use to spot AI writing, tell it never to use them, then 'type slash and open skill creator' and say 'create a reusable humanizer skill using everything we've just discussed'. Install once and 'type slash human before your prompt'. On screen: a Claude chat on Sonnet 5, a '/skill-creator' command, a 'Humanizer skill' card, and an 'AI Detection Results: 100% of this text appears to be human-written' badge. 'Comment human and I'll send you the full guide.'

The mechanics are real. Anthropic publishes a skill-creator skill in its public skills repo, described as 'Create new skills, modify and improve existing skills... use when users want to create a skill from scratch', and it packages the result for install. Claude's help center says Skills are available on Free, Pro, Max, Team and Enterprise plans, and the screen recording shows the real thing: a claude.ai chat on Sonnet 5, Claude searching for detector signals (burstiness, stock transitions, hedging, tidy bullet structure), then a 'Humanizer' skill card with 'Save skill'. So yes, you can turn a style brief into a reusable skill in one chat. What is oversold is the payoff. The '100% human-written' badge is one third-party detector's score on one sample; detectors are built on the very surface signals the skill removes, so passing one proves the text is smoother, not that it is good or that a different detector agrees. 'Never use those obvious AI giveaways' makes writing less generic, but readers detect AI by missing specifics, not by em dashes; the skill adds no facts, stories or opinions, which is the part that actually reads as human. Two details did not check out: Claude's help article describes skills as auto-invoked and a ZIP upload, not a slash picker in chat (the picker appears on screen, so it may be newer than the article), and no Anthropic documentation of a Claude text 'watermark' exists. The sell is the comment-for-guide hook; every prompt you need is spoken in the reel.

What holds up

Reel read frame by frame: claude.ai chat titled 'Making writing sound more natural' on 'Sonnet 5 Medium'; Claude lists detector signals (low perplexity, low burstiness, stock transition phrases, hedge-y evenhandedness, bullet overuse, watermarking 'though this isn't universal'); a slash menu lists 'skill-creator'; a 'Humanizer skill' card appears; a separate screenshot shows 'AI Detection Results 100% of this text appears to be human-written'.
github.com/anthropics/skills/skills/skill-creator/SKILL.md (fetched Oct 2, 2026): name 'skill-creator', description 'Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch...'; it drafts, tests and packages a skill for install.
support.claude.com 'Using skills in Claude': 'Skills are available for users on Free, Pro, Max, Team, and Enterprise plans'; custom skills are added via Customize > Skills and 'Claude will automatically use these tools when relevant'. The slash-picker in chat shown on screen is not described in that article; typing '/' is documented for the Microsoft 365 add-ins.
No Anthropic documentation of a hidden text watermark in Claude output was found; the on-screen research note itself calls watermarking 'not universal'. The '100%' badge is a single detector's output on a single sample and is not evidence that other detectors or human readers agree. x-ref #411 for the same creator's 'five detectors' claim.

What doesn’t

The '100% human-written' badge is one detector, one sample; detectors are unreliable in both directions and passing one is not a result.
The skill removes surface tells (em dashes, 'delve', tidy bullets) but adds nothing specific; generic content still reads as AI to a human.
'Claude's watermark' is asserted with no documented watermark to remove.
Comment-for-guide hook for prompts that are read out loud in the reel; the funnel is the product.

The catch

Yes, make the skill; it is a ten-minute job and it will stop Claude from writing like a press release. Just do not mistake a green detector badge for human writing: the fastest way to sound like a person is to give Claude your own numbers, names and opinions, and no humanizer can supply those.

How to actually do it

In a Claude chat, paste three samples of your own writing and tell Claude the rules you want (short sentences, your vocabulary, no stock transitions, contractions, concrete numbers), then ask skill-creator to turn that into a reusable skill and save it.
Every time you use it, feed the prompt one real detail a reader could not get elsewhere (a figure, a story, a decision you made); edit the first two sentences by hand.
Ignore detector scores unless a specific client or school requires one; if they do, ask which detector and threshold they use and test against that one, not a random badge.

Claude's skill-creator and reusable custom skills are real on every plan, and the workflow shown would produce a working style skill. The reel oversells the result with a single detector badge and a 'watermark' that is not documented, and withholds nothing of value behind the comment hook since the prompts are spoken on camera.

Confidence: Medium
Posted by: an AI-tips creator posting Claude tutorials (about 110k views on this reel); reel sent to Buddy Wed Sep 30, 2026, ~7:24 AM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.