54%
of long-form posts are AI-assisted
More than half of long-form LinkedIn posts are now AI-assisted. That is a 189% rise since ChatGPT launched. LinkedIn built a classifier specifically in response. Unhumanized AI posts are no longer a risk — they are the majority of what the filter was trained to catch.
47%
less reach for AI-pattern posts
Brixon Group analyzed 500+ B2B LinkedIn profiles in 2025. Posts with recognizable AI patterns got 47% less organic reach and lost LinkedIn's Long Dwell classification, which is the top-percentile distribution gate. Forcalabs strips every pattern from the Brixon dataset.
94%
confirmed detection accuracy
LinkedIn VP Laura Lorenzetti confirmed the platform's AI classifier reaches 94% accuracy in early tests and throttles flagged content to first-degree reach only. Generic humanizers do not know what the classifier was trained on. Forcalabs does.
LinkedIn's classifier was trained on human-annotated genuine versus generic labels. These are the six signal layers it measures. All six are built into our filter.
"delve", "leverage", "robust", "tapestry", "pivotal", "innovative", "seamless", "paradigm"
Reinhart et al. PNAS 2025: these words appear 5-40x more often in AI text than in human LinkedIn posts across an 8,795-post corpus
"It's not X, it's Y" · "Not only...but also" · Rule-of-three triplets · Trailing present-participle clauses ("...revealing key insights")
Wikipedia "Signs of AI Writing": corrective negation is among the strongest single structural detectors; instruction-tuned LLMs use trailing participials at 2-5x human rate
Em dash overuse · Bold inline bullet headers · Title-case headings · Emoji as list markers · Curly smart quotes
Wikipedia "Signs of AI Writing": em dashes used "in a formulaic, pat way, often mimicking punched-up sales-like writing." Markdown-training leakage into prose.
"I'm thrilled to announce" · "Humbled to share" · "Game-changer" · calm monotone with no opinion
LinkedIn VP Laura Lorenzetti (The Next Web, May 20, 2026): classifier is trained to flag "posts that feel repetitive, generic, and empty." Confirmed 94% accuracy in early tests.
No named specifics, numbers, or dates the author could know · Generic "leadership lesson" with no actual story · Conclusion restates the opening
Brixon Group 2025 analysis of 500+ B2B LinkedIn profiles: "posts with recognizable AI patterns achieve 47% less organic reach" and lose LinkedIn's Long Dwell classification
"Agree or disagree?" · "Thoughts?" · "Tag someone who needs this" · "Comment YES if you agree"
LinkedIn Q4 2025/Q1 2026 product comms: engagement bait phrases are actively suppressed. Hollow carousels are deprioritized.
A Founder says “burn rate, shipped, ICP.” A Recruiter says “comp band, passive candidate.” A Job Seeker says “open to opportunities, any leads.” A tool that humanizes generically produces text that reads like an AI wrote it professionally. Our engine loads the correct vocabulary cluster and storytelling pattern for your persona before generating a word.
Vocab: runway, burn, ARR, ICP, churn, PMF, GTM, shipped, hard pivot, build in public
Story: Failure then learning then next experiment. First-person, present tense, skin in the game.
Avoid: Over-polished "vision" posts; rule-of-three triplets of corporate virtues; "I'm thrilled to announce"
Vocab: rec, req, comp band, passive candidate, ATS, open to work, moved to offer, warm intro
Story: Pet-peeve then real anecdote with a specific candidate or role then tactical fix.
Avoid: "We're hiring rockstars!" or sycophantic team praise with zero specifics
Vocab: open to opportunities, gap, transition, currently exploring, any leads, referral
Story: Loss or transition then honest reframe then specific ask with a real skill.
Avoid: Over-formal humble-brag or "humbled to announce this incredible opportunity"
Vocab: quota, MEDDIC, BANT, ICP fit, discovery call, decision-maker, champion, no-show, ghosted
Story: Cold outreach fail or awkward sales moment then reframe then specific process change.
Avoid: "Game-changer," "synergy," "leverage our solution," "streamline your pipeline"
Vocab: 5 lessons, Year 1 vs Year 3, newsletter, followers, audience, post hit different, thread
Story: Hook then 5-7 numbered beats then one real reflection then a CTA that requires thought.
Avoid: Template-perfect broetry; every post has the exact same shape; repeated hook formulas
Vocab: stepping into, after 12 years, reflecting on, grateful, excited to share, milestone
Story: Tenure then one specific contestable thing learned then forward-looking statement with a real stake.
Avoid: "Honored," "humbled," AI-generated promotion posts with zero specifics or named people
210
chars
Hook constraint
First 210 characters must earn the "...see more" tap on mobile. The hook is everything.
1,242
– 2,500 chars
Post sweet spot
Long enough for Long Dwell classification. Short enough for completion. Van der Blom 2025 (1.8M posts).
3
max hashtags
Hashtag limit
More than 3 hashtags causes 70% lower reach than posts with none. Richard van der Blom Algorithm InSights Report 2025.
The same human credibility markers that fool Reddit moderators fool LinkedIn's classifier. Named specifics, body verbs, calibrated uncertainty, asymmetric paragraphs, and deliberate imperfection are the signals our engine injects.
"I was sitting at Blue Bottle on Valencia when I got the email." Specificity is not description. It is facts only the author could know.
"We hit 67% retention" beats "We saw strong retention." "$11,400 in revenue" beats "five-figure revenue."
cried, panicked, froze, laughed — not "felt challenged" or "experienced difficulty."
"I thought it was a culture problem. Actually no. It was a comp problem."
A 4-line block. Then a 1-word line. Then a sprawling parenthetical aside. Uniform paragraphs are an AI tell.
A sentence fragment. A comma splice for rhythm. One-word lines. Grammar deployed for effect, not correctness.