The Humanization Playbook

Part 2, Module by Module

---

Module 1: Personal Voice (25%)

The rule. Use first-person pronouns 8 to 12 times per 1000 words.

Why it works. Detection models train on human writing. Humans naturally talk about themselves. AI defaults to a disembodied, encyclopedic voice. Drop your I-count below 5 per 1000 words and you read as AI to both the algorithm and the careful reader. Push it above 15 and you read as narcissistic, which also flags.

The validator logic. Strip code blocks and markdown formatting. Count every instance of I, me, my, mine, myself, I'm, I've, I'll, I'd. Divide by total word count. Multiply by 1000.

Author bio in the introduction. Validators check the first 500 characters for the author's name. If you're publishing under your byline, mention yourself in the first paragraph. One line is enough: "I'm Piyush, and I've spent the last three years building content engines for service businesses."

What to avoid.

• Stripping out all I's because a style guide told you to. The style guide is wrong for AI-detection content.
• Stuffing I's into every sentence. Density should be steady, not bunched.
• Writing in the third person about yourself ("the author has found that..."). It triples the AI signal.

---

Module 2: Specific Examples (25%)

The rule. Name real people. Use exact numbers. Never use a phrase from the blacklist.

Why it works. AI cannot invent specific facts safely, so it defaults to vague abstractions. A real human writer reaches for the specific because the specific is what they actually remember. Specificity is the single hardest thing for AI to fake convincingly.

The validator logic. Score starts at 100. Deduct points for each generic phrase. Add bonuses for real proper-noun names and exact numerical values.

The 13 generic phrases (and their penalties)

High severity (deduct 15 each):

Medium severity (deduct 10 each):

Low severity (deduct 5 each):

The bonuses

• +5 per real person (proper-noun first name + last name pattern), capped at +20.
• +3 per specific number (dollar amount, percentage, or 3+ digit number), capped at +15.

So a piece with 0 real people, 0 specific numbers, and three "studies show" phrases starts at 100, loses 45, and lands at 55. A piece with the same body but four named people and five exact dollar figures starts at 100, gains 20+15, and lands at 100.

If you genuinely don't have a specific source or name, swap to first-person: "In my own experience..." reads as legitimate uncertainty. "Studies show..." reads as fabricated authority.

What to avoid.

• Inventing fake names or fake percentages to pass the check. Detectors don't care about the truth, but humans do, and reputation matters more than score.
• "Around $50K" when you mean "I don't remember." Just say "I don't remember the exact figure."
• Generic "many users have reported" when you can say "three of my clients last quarter."

---

Module 3: Credibility (20%)

The rule. Show experience with timelines, results with numbers, and proof with names.

Why it works. AI-generated content reads as opinion without provenance. Human writing carries earned-it markers: "I've done this for five years," "I've worked with 200 clients," "We hit 4x ROAS on this campaign." Those markers signal a real history that AI can't fake without inventing.

The validator logic. The scorer looks for three pattern families:

1. Experience patterns (15 points each, up to 45)

• "In my X years of..." or "I've been doing X for Y years..."
• "Over the past X years..." or "Over the last X..."
• "I've helped X..." or "I've worked with X..." or "I've built X..." or "I've created X..."

2. Results patterns (5 points per match, up to 20)

• Dollar amounts: $5,000, $73,660, $1.2M
• Multipliers: 4x ROAS, 3x ROI, 10x growth
• Percentages tied to outcomes: 20% increase, 64% improvement, 120% growth

3. Social proof patterns (15 points)

• Client mentions in the form [Name], from/at [Company]
• Capitalized first-name patterns followed by a comma or "from" or "at"

A piece with no experience claims, no dollar/percentage outcomes, and no named clients scores 0 on this module.

What to avoid.

• Round-number claims that read as fake: "I've helped 1000 clients." Use the precise figure or a range you can defend: "228 service businesses since 2022."
• Citing "a client" instead of naming them. If you can't use a real name (NDA, privacy), at least say "a Boston-based dev agency I worked with last spring."
• Stacking dollar figures with no context. "$73,660" lands when it answers a specific question; otherwise it reads as filler.

---

Module 4: Parentheticals (12%)

The rule. Drop 3 to 5 conversational asides per 1000 words. Place them in safe zones only.

Why it works. Parentheticals are the hardest pattern for AI to use naturally. They're the verbal equivalent of an aside in a conversation. AI either skips them entirely or sprays them everywhere. Human writers use them at irregular intervals, in specific emotional contexts, with specific tones.

The validator logic. Count every parenthesized phrase between 5 and 50 characters long. Divide by word count. Multiply by 1000. Target density: 3-5 per 1000 words.

The 14 templates (by type)

Conversational (after surprising statements):

• (yup, really)
• (I know, I know)
• (trust me on this)
• (seriously though)
• (hear me out)

Shared observation (after common frustrations or surprising facts):

• (we've all been there)
• (shocking, right?)
• (who knew?)
• (plot twist)

Personal (before vulnerable thoughts or admissions):

• (honestly?)
• (admittedly)
• (in my experience)

Humor (after unrealistic expectations or bad options):

• (yeah, right)
• (good luck with that)
• (no thanks)

Safe zones

Drop parentheticals only in these places:

• Mid-article body paragraphs. Not in the first 200 words.
• After the point is made. Not in the middle of explaining a mechanism.
• In personal anecdotes. Not in technical or data-heavy passages.

Forbidden zones

Never put parentheticals in:

• Headings (H1, H2, H3, none of them)
• Direct quotes or testimonials
• Statistics or technical specifications
• Credentials, citations, or attributions

• Surprising claim just landed? "(yup, really)" or "(shocking, right?)"
• You're about to admit something inconvenient? "(honestly?)" or "(admittedly)"
• Reader is probably skeptical right here? "(I know, I know)"
• You're about to mock the obvious bad choice? "(yeah, right)" or "(good luck with that)"

What to avoid.

• Repeating the same parenthetical twice in one piece. It reads as a tic.
• Stacking parentheticals in consecutive sentences. They lose their function.
• Putting them in headings. The validator catches it instantly and so do readers.
• Forcing them where they don't fit. If you can't pick a template that lands, leave the spot alone.

---

Module 5: Objections (10%)

The rule. Address 2 to 5 reader objections inline, using a 4-step pattern.

Why it works. AI never anticipates pushback because it has no opponent. Real writers addressing real readers know which sentences will trigger an eye-roll and answer the objection before the reader closes the tab. That dynamic (claim, anticipate-the-doubt, defuse) is structurally invisible in AI output and unmistakable in human writing.

The validator logic. Scan for trigger phrases that signal an objection-handling moment:

• "I know what you're thinking..."
• "You might be thinking..." or "You might be wondering..." or "You might be skeptical..."
• "The concern I hear most..." or "The pushback I get..." or "The fear I hear..."
• "I get it..."

Count instances. Target: 2-5 per article. Below 2 = not enough reader empathy. Above 5 = the article reads as defensive.

The 4-step pattern

Every objection-handling block follows the same structure:

1. Acknowledge. Name the objection out loud. "I know what you're thinking..."
2. Validate. Tell them their concern is reasonable. "I get it. [Concern] is legitimate."
3. Counter with experience. Open with "Here's what I've found..." or "Here's the thing..."
4. Provide evidence. A specific example, number, or case from your work.

A worked example

That paragraph hits all four steps in 76 words. The reader's likely objection is named, validated, countered, and proven.

Where to place them

Place objections in sections where reader skepticism is highest:

• Right after a bold claim
• Right before a sales-feeling section
• After the third or fourth main point, when attention starts to flag
• In the conclusion, before the call to action

What to avoid.

• More than 5 in one piece. The article starts to feel like a defense brief instead of a confident point of view.
• Generic objections that don't match your reader. "You might be thinking this is too good to be true" is a tell. "You might be thinking this only works for B2B" lands.
• Acknowledge-then-dismiss without the experience step. The 4-step pattern works because the experience is the proof. Skip it and the objection-handling reads as performative.

---

Module 6: Imperfections (8%)

The rule. Allow strategic human texture in safe zones. Never ship critical errors.

Why it works. AI-generated text is too clean. Real human writing has texture: the occasional informal contraction, the colloquial repetition for emphasis, the sentence fragment for drama. Detectors have learned to flag the sterile uniformity of polished AI prose. A small, controlled amount of texture reads as human.

This is the lowest-weight module on purpose. It's a tiebreaker, not a foundation. Get the other five right first.

The validator logic. Two checks:

1. Critical-error scan. Look for unreplaced placeholders ([YOUR_NAME], [YOUR_COMPANY]), incomplete markers, double periods, missing spaces after punctuation. Any critical error caps this module's score at 0.
2. Density check. Validate that texture appears in safe zones at the right rate, without forming a detectable pattern.

What counts as strategic texture

Colloquial repetitions (in parentheticals only):

• "really really effective" (in an aside)
• "many many options" (in an aside)

Sentence fragments (sparingly, for emphasis):

• "And honestly? It works."
• "The result? Mind-blowing."

Informal contractions (in personal stories only):

• "gonna" instead of "going to"
• "wanna" instead of "want to"

Where you can use them

• Inside parenthetical asides
• In personal anecdotes
• In the casual middle sections of an article

Where you cannot

• In headings
• In the first or last 200 words
• In statistics, citations, or data
• In credentials or claims

What to avoid.

• Trying to inject errors deliberately to fool detectors. If a typo doesn't sound like one a real person would make in flow, it reads as sabotage.
• Overusing fragments. More than two per article and the writing reads as twitchy.
• Any of this in the first 200 words. The opening sets the contract with the reader. Keep it clean.
• Critical errors. Placeholder text in published copy is the single fastest way to fail this module and lose reader trust at the same time.

---

Part 3, The Self-Audit Checklist

Run this on any draft before publishing. 32 yes/no questions across the 6 modules.

Module 1: Personal Voice

• Have I counted my I-count? (Aim for 8-12 per 1000 words.)
• Does my introduction mention me by name in the first 500 characters?
• Have I converted at least three "Studies show..." or "Research indicates..." patterns to first-person?
• Are I-pronouns spread throughout the piece, not bunched in one section?

Module 2: Specific Examples

• Zero high-severity generic phrases in the final draft? (studies show, research indicates, experts say, many experts, "a [role] I know")
• At least 2 medium-severity phrases removed or replaced? (around $X, approximately, roughly, "it is recommended")
• At least 2 named real people in the piece, with full names?
• At least 5 specific numbers (dollars, percentages, or 3+ digit figures)?
• Every claim that could carry a number has one?

Module 3: Credibility

• At least 1 "in my X years" or "over the past X years" experience marker?
• At least 1 specific dollar amount or multiplier tied to a result?
• At least 1 named client or case, even if anonymized by city or industry?
• No round-number claims that read as fake?

Module 4: Parentheticals

• 3-5 parentheticals per 1000 words?
• None in headings?
• None in the first 200 words?
• None in technical specifications or statistics?
• No template repeated more than once?

Module 5: Objections

• 2-5 objection-handling blocks?
• Each block follows the 4-step pattern (acknowledge, validate, counter, evidence)?
• Objections placed where reader skepticism is highest?
• No more than 5 total?
• None of them are generic ("this sounds too good to be true")?

Module 6: Imperfections

• Zero unreplaced placeholders ([YOUR_NAME], [YOUR_COMPANY], etc.)?
• Zero incomplete markers?
• Zero double periods or missing spaces after punctuation?
• At most 1-2 strategic fragments or contractions, all in safe zones?
• First and last 200 words are clean of any informal texture?

Final pass

• Read the whole piece aloud. Does it sound like you?
• Hand it to one person who knows your voice. Do they recognize it?
• Run it through one detector tool (GPTZero, Originality, or Copyleaks). Score 75 or higher?
• If under 75, identify the lowest-scoring module and run that section again?

---

Part 4, Three Worked Rewrites

The fastest way to internalize the system is to watch text move from a failing score to a passing one. Each rewrite takes a real-feeling AI draft and runs it through the rubric.

---

Rewrite 1: From 42 to 87

Topic. A short blog opening on email list building.

Before (score: 42 / 100)

Module-by-module diagnosis

Weighted overall: ~42.

After (score: 87 / 100)

Why it scores

Weighted overall: ~87.

---

Rewrite 2: From 38 to 81

Topic. A LinkedIn post about productivity.

Before (score: 38 / 100)

After (score: 81 / 100)

Module deltas

• Personal voice went from 0 to 11 I-pronouns in 169 words.
• Specific examples: removed all 4 high-severity phrases. Added one named real person (the 22-person agency owner), exact numbers (22, 9, three, 8, 24, 30-40), and a quoted line.
• Credibility: experience marker ("In my last 8 months running this system"), dollar/time outcomes implied, named case study.
• Parentheticals: 1 well-placed aside.
• Objections: 1 full 4-step pattern.

---

Rewrite 3: From 51 to 89

Topic. A landing-page section selling a paid course.

Before (score: 51 / 100)

After (score: 89 / 100)

Module deltas

• Personal voice: 11 I-pronouns in 200 words.
• Specific examples: zero generic phrases. Two named people (Sarah, Marcus) with city/industry. Eight specific numbers. Two named outcomes.
• Credibility: experience marker ("In my 3 years running this format"), comparison number (81% vs 8%), named clients.
• Parentheticals: 1 conversational aside.
• Objections: 1 full 4-step pattern, dropped at the moment of highest skepticism (the price-objection point).
• Imperfections: 1 deliberate fragment ("Not learned about them. Shipped them.") for emphasis at the moment of impact.

---

Closing

The 6-module system isn't magic. It's six small disciplines that, applied together, change how the text reads to a model and to a reader at the same time.

If you only remember one thing: AI fails detection because it talks like nobody, names nobody, has no track record, never thinks aloud, never anticipates pushback, and reads too clean. Fix those six failures and you fix the score.

Run the self-audit checklist on every draft. Track your score. Watch what improves when you fix the lowest-weighted module first.

Built more like this at the Massive Impact resource library.

---