Overview

Remove the signs of AI-generated writing from emails, briefs and reports before anyone reads them. Claude runs a 13-point checklist: inflated symbolism, promotional language, empty -ing analyses, vague attributions, em dashes, AI vocabulary, passive voice, 'not just X' constructions, filler, missing contractions, sentence length, greeting-first openers, and dates or facts that drift from what was given.

Use cases

Cold emails that don't read as machine-written
Every draft gets the checklist before delivery: 80 to 120 words, no sentence over 30 words, contractions, no filler openers and no em dashes. The email opens on something specific to the recipient, built on data that actually supports it.
Briefs your team is comfortable sending externally
Account briefs, QBR pre-reads and executive summaries lose the inflated symbolism and promotional tone while keeping the source and date of every claim, so they read like an analyst's work.
Dates and facts kept exactly as given
Claude keeps 'Q2 2026' as 'Q2 2026' instead of turning it into 'recently', and doesn't add budgets, plans or years the input never stated.

View full skill

Humanizer

When to use

Drafting or reviewing any text that will be read by humans (emails, briefs, reports, summaries)
Final pass before delivering output to the user
Any workflow step that produces prose (especially email drafting)

The checklist

Apply every rule below to all prose output. If a draft violates any rule, rewrite the offending passage before presenting it.

1. Inflated symbolism

Pattern: Treating ordinary business events as profound turning points.
Fix: State facts plainly. Write "They adopted a new data warehouse in Q3", not "This pivotal shift signals a transformative reimagining of their data strategy."

2. Promotional language

Pattern: Reading like marketing copy or a press release.
Fix: Remove superlatives and hype words. Say what happened, not how amazing it is.

3. Superficial -ing analyses

Pattern: Gerund-heavy sentences that sound analytical but say nothing. "Leveraging their existing infrastructure while optimizing for cloud-native architectures..."
Fix: Use concrete verbs. Who did what, when, with what result.

4. Vague attributions

Pattern: "Many experts believe..." / "Studies show..." / "It is widely recognized..."
Fix: Name the source or drop the claim. If you can't cite it, don't assert it.

5. Em dashes

Pattern: Em dashes used as dramatic pauses or to bolt clauses together.
Fix: None at all, in any output. Use commas, colons, periods, or parentheses. En dashes and double hyphens used as dashes count too. For ranges, write "80 to 120" or "80-120".

6. AI vocabulary words

Banned words and replacements:

Don't Use	Use Instead
leverage	use
synergy	overlap, shared benefit
game-changing	significant, important
revolutionary	new, improved
unlock potential	improve, enable
holistic	complete, full
robust	strong, reliable
cutting-edge	new, advanced
paradigm shift	change
deep dive	detailed look
landscape	market, space, field
navigate	handle, manage
ecosystem	system, network
seamless	smooth
empower	help, enable

7. Passive voice

Pattern: "The platform was adopted by the team." "It was determined that..."
Fix: Use active voice. "The team adopted the platform." "We determined..."

8. Negative parallelisms

Pattern: "This is not just X, it's Y." "Not merely a tool, but a transformation."
Fix: State what it IS. Drop the "not just" construction entirely.

9. Filler phrases

Remove entirely:

"It's worth noting that..."
"It's important to remember that..."
"I hope this email finds you well"
"I wanted to reach out because..."
"At the end of the day..."
"In today's fast-paced world..."
"Moving forward..."
"Going forward..."
"As we move into..."
"On a broader note..."
"That said..."
"That being said..."

10. Contractions

Use them. "Don't" not "do not." "It's" not "it is." "Can't" not "cannot." Formal writing without contractions reads robotic.

11. Sentence length

Target: 80 to 120 words per email body. Mix short and medium sentences. No sentence over 30 words. No paragraph over 3 sentences.

12. Opening lines

Never start with a greeting + filler. Start mid-thought with something specific.

BAD: "Hi Sarah, I hope this finds you well. I wanted to reach out because..."
GOOD: "Sarah, your team is actively researching data observability tools, and HG still shows your current monitoring platform as a core install."

The good opener is built on fields the data actually returns (an intent topic on resolved topics, an install verified in the last 90 days). Never open on a figure the data can't support, such as a renewal probability.

13. Dates and facts

Dates and facts as given. Keep every date as it was given ("Q2 2026", "since 22 Sep"), with its year; don't turn it into relative time ("this month", "recently"). Don't add years, dates, budgets, plans or heuristics the input doesn't state.

Common pitfalls

Fixing vocabulary but leaving the structure. Swapping "leverage" for "use" doesn't help if the sentence still says nothing concrete.
Over-correcting into terseness. Short is good; cryptic isn't. Keep the subject, the verb and the number.
Humanizing away the evidence. Plain language must keep the source and the date of each claim.

Reference

Full docs for LLMs: https://phoenix.hginsights.com/docs/llms-full.txt
Related skills: executive-prose-conventions, opening-line-discipline, pvp-composition, hg-citation-discipline

Skill: Humanizer