What Makes Writing Feel Human

45%

less engagement for likely-AI LinkedIn posts vs human-written ones (Originality.ai, 2024)

53%

of long LinkedIn posts were likely AI-generated in 2025 (Originality.ai study, 99 profiles)

signals LinkedIn's algorithm uses to detect AI: sentence variance, transition density, specificity, "I" count, tone consistency, lived-experience anchors

Statistical signals

What AI detectors measure · invisible to the eye, visible to algorithms

Burstiness

Structural

Humans alternate wildly between short and long sentences. AI keeps them uniform.

"AI offers many advantages for marketing teams. These tools can automate repetitive tasks. They can also improve content quality. Additionally, they reduce production time significantly."

Human

"We cut production time by 60%. Not because the tool was magic — we spent the first two weeks breaking it in ways that forced us to actually understand what we were producing."

Perplexity spikes

Lexical

Humans pick unexpected words — chosen for rhythm, memory, humor. AI picks the most statistically probable next word.

"This approach offers significant advantages for organizations seeking to optimize their content strategy."

Human

"This fixed the thing we'd been arguing about in Slack for three months."

Vocabulary diversity

Lexical

Humans have more one-time-use words (hapax legomena) and more everyday stop words (it, but, well, just). AI overuses rare "impressive" words and underuses casual ones.

AI tell

Uses "leverage," "optimize," "holistic," "robust" repeatedly. Almost no "well," "but," "actually," "I."

Human signal

Uses plain verbs most of the time. One specific unusual word that only makes sense in context.

Structural signals

How the piece is built · detectable in the first 30 seconds of reading

Transition density

Pattern

AI uses formal connectors to stitch paragraphs. More than 3 per 200 words is a flag.

AI connectors

"Furthermore..." / "Moreover..." / "Additionally..." / "In conclusion..." / "It is worth noting..."

Human connectors

"The thing is..." / "Which meant..." / "So." / Nothing at all — just the next sentence.

Nominal inflation

Grammar

AI turns verbs into nouns. "We decided" becomes "the decision was made." It sounds authoritative. It reads as robotic.

AI (noun-heavy)

"The implementation of a strategic framework enables the facilitation of organizational alignment."

Human (verb-led)

"We built a framework. The team aligned. It took six weeks and three rewrites."

Absent agency

Voice

AI makes abstractions the subject. "The data suggests..." "The study reveals..." Humans put themselves or other people in the sentence.

AI (displaced agency)

"The analysis reveals that organizations face challenges when adopting AI-first strategies."

Human (owned agency)

"We spent four months on this. What I found surprised me."

Cognitive signals

Traces of a thinking process · what AI cannot fake

Visible thinking

Cognition

Humans contradict themselves, qualify, change direction mid-sentence. AI maintains perfect logical coherence — which reads as inhuman.

AI (no trace of thought)

"AI content performs better when personalized. Personalization drives engagement. Engagement drives conversion."

Human (thinking aloud)

"I thought personalization would fix it. It didn't, at least not the way I expected. What actually moved the number was..."

Messy examples

Specificity

AI examples are perfectly illustrative, never contradictory, never messy. Real examples have caveats and context.

AI example

"For example, a B2B SaaS company improved conversions by 40% after implementing targeted messaging."

Human example

"It worked for the client — but we had a 6-week false start because we were targeting the wrong ICP. That failure was actually the useful part."

Uneven intensity

Emotion

AI is emotionally isothermal — equal investment in every paragraph. Humans burn hotter on topics they care about and get brief on topics they don't.

AI tone

Every sentence carries the same professional, neutral register from first word to last.

Human tone

One paragraph is a controlled rant. The next is two sentences. The closing is almost quiet.

Social risk

Voice

Humans write as if they might regret it. They soften the controversial claim, add irony as armor, hedge before saying something that could be read wrong.

AI (no social risk)

"Paid media is essential for scaling reach and should be a core component of any growth strategy."

Human (social risk visible)

"Unpopular opinion, but paid before organic is usually a mistake. I know some of you will disagree."

Platform signals

What LinkedIn specifically penalizes · specificity vs generality

No concrete anchors

Platform

Posts without real numbers, real names, or real dates score low on specificity. These are the details only a person who was there would know.

AI (generic)

"Many companies have seen significant improvements after implementing AI-assisted workflows."

Human (anchored)

"Three weeks ago, a client cut their content pipeline from 14 days to 4. One change. Not a stack overhaul."

Missing "I"

Platform

First-person singular is a human marker. A professional post with fewer than 2–3 "I" statements for 200 words reads as ghostwritten or generated.

AI (absent "I")

"The key to effective LinkedIn content lies in authenticity and consistent value delivery to one's audience."

Human (present "I")

"I deleted the post, rewrote it from scratch, and it performed 3x better. I still don't fully understand why."

◎

The core principle

AI produces the most statistically probable version of a post. Human writing is full of choices that wouldn't be predicted.

Unexpected word. Rough sentence. Admitted failure. Unresolved tension. These are not imperfections — they are the proof of a mind at work.

✓

Pre-publish rubric

Run before every post · 6 pass/fail checks

Pass conditions

● At least one sentence under 6 words AND one over 25 words

● At least one concrete anchor: a real number, name, date, or place

● At least 2 first-person "I" statements (for 150–300 word post)

● At least one moment of visible thinking: a qualification, correction, or admission

● Subject of at least one sentence is a person, not an abstraction

● Fewer than 3 formal transition words per 200 words

Instant fail conditions

● Every sentence is 12–20 words (uniform length)

● No numbers, no names, no dates — pure generality

● Zero "I" statements in a personal post

● All examples are perfectly clean — no caveats, no failures, no edge cases

● 3+ formal connectors: "furthermore," "moreover," "in conclusion," "it is worth noting"

● Uniform emotional tone — equally invested in every paragraph

What makes writingfeel human

What makes writing
feel human

What makes writing feel human (and why AI posts get caught)