Every HooklyAI score is 1–100, built from seven equally-weighted dimensions with hard quality gates and anti-inflation calibration. This page documents the entire system — because a score without a published method is just a number.
Last updated: September 2026 · Applies to every generator and the analyzer
Each dimension maps to a behavior LinkedIn's feed can actually observe. The final score is the equal-weight average of all seven.
Dwell time and the "see more" click. The feed truncates near the first 210 characters; whether a reader expands decides whether the rest of the post exists.
Skimmability on mobile. Single-sentence lines, white space, and scannable structure keep readers moving through the post instead of bouncing off a wall of text.
The pause. Curiosity, urgency, surprise, and empathy are what stop a scroll — the mechanical precursor to every other engagement signal.
Comments and replies. A genuine question or ask that invites discussion outperforms generic "thoughts?" bait, and discussion depth feeds distribution.
Saves, reposts, and tag-a-colleague behavior. Content that professionals find useful enough to attach their name to travels beyond the first-degree network.
Penalty avoidance. External links in the body, engagement bait, wrong length, and hashtag stuffing are all known distribution suppressors — the dimension checks for them explicitly.
The read test. Readers — and increasingly the feed — discount prose that sounds machine-generated: generic openers, buzzword padding, over-symmetric constructions, hollow superlatives. This dimension checks for those AI-tells and rewards concrete specifics, first-person experience, and varied sentence rhythm.
Included verbatim in every scoring prompt — ours and the model's. The bands exist to prevent the inflation every other AI score suffers from.
Two rules in the calibration do most of the anti-inflation work. First: “an average LinkedIn post scores around 45–55” — anchoring the middle of the scale where it belongs, instead of letting every decent draft drift toward 90. Second: the model is explicitly instructed not to award scores above 70 unless the content genuinely earns it. The result is a scale with consequences: a 62 means something, because most posts — including competent ones — live in the 40s and 50s.
Temperature is controlled, too. Generation runs warmer (creative variation); scoring runs cooler for consistency, so the same text graded twice lands in the same band.
Some flaws aren't stylistic — they're known distribution penalties. Two of them override everything else.
LinkedIn's distribution system is known to suppress posts that send readers off-platform. A post with external links in its body cannot score above 60, regardless of how strong the writing is. The fix is free and instant: move the link to the first comment (that's what the first-comment tool is for).
“Comment YES if you agree!”, tag-baiting, and recycle-emoji prompts exploit the feed instead of earning it. Bait cannot score above 60 either — the rubric treats manipulation of the ranking system as a defect, not a tactic.
The gates are deliberately blunt. A weighted average could let excellent writing “buy back” a distribution penalty — but the penalty applies on LinkedIn regardless of how good the prose is, so the score should reflect that reality rather than flatter the writer. Gates make the score honest in the direction that matters for a publishing decision.
The rubric is not just described here — it is audited against its own output. The State of the LinkedIn Hook, 2026 data study publishes how real scores distribute across these bands (mean 66.3, no hook below 45, 22% at 75+ as of the October 2026 snapshot), which is what a calibration claim looks like when it is checked in public.
The rubric's first dimension has a hard, mechanical reference point — the same one our preview tool uses.
The feed truncates long posts at roughly the first 210 characters on desktop — about three lines — behind a “see more” link, and mobile truncates earlier. Everything below the fold is invisible until a reader opts in, and most never do for a weak opening. That visible slice is the post's actual first impression, so the hook dimension explicitly asks: does the opening grab attention within 210 characters?
Using the same number in the scorer and in the post preview tool is deliberate: when the analyzer grades your hook and the preview shows you the fold, both are pointing at the same 210 characters. Tools that share a reference point can't quietly disagree with each other.
The generator's self-scores and the analyzer's independent review read from the same rubric file.
A common failure in AI writing tools is that the generator praises its own output at 95 while an independent review would say 60. HooklyAI closes that gap structurally: every generator (hooks, posts, comments, ideas) and the standalone analyzer all score from the same rubric and the same calibration text — a single shared definition of what “good” means. A hook that earns 78 from the generator would land near 78 in the analyzer.
This is also why before/after deltas on our pages mean something: the “before” and the “after” are graded by the same scale, so the delta measures the change — not a change of judge.
Toggle between a typical draft and its strengthened edit. Every number below is computed in your browser by the same deterministic scorer the analyzer uses — not a screenshot, not a stock graphic.
What the feed shows before …see more (210 chars)
I am happy to share some exciting news about my professional journey. In today's world, networking is very important for everyone, and I have learned a lot during my time working on various projects. I want to tell you all about my experience with professional development and the growth opportunities that I have encountered along the way. Hard work always pays off, and I am grateful for everything I have learned from my team and my managers over the past few years.…see more
Live output from the same deterministic scorer that powers the analyzer — not a screenshot.
Six published dimensions, hard quality gates, methodology in the open. How scoring works
The limits, stated plainly — because a methodology page that hides its limits isn't a methodology.
The score is not a LinkedIn metric. It is produced by an AI model applying a fixed rubric — it does not have access to your account, your audience, or LinkedIn's internals, and it cannot predict reach for a specific post on a specific day. Timing, audience fit, and the platform's own experiments all live outside any pre-publish score.
It is also not a plagiarism or originality detector, and it does not measure truth. A fabricated statistic inside a beautiful story will score on its merits as prose — fact-checking remains a human job, and we won't pretend otherwise.
What the score does reliably: tell you, before you spend the publish, whether the writing gives the post a fair chance — measured against calibration that is published, identical across every tool we ship, and open to scrutiny on this page.
Score your draft against these seven dimensions, see the reasoning, and fix what the rubric catches — before the feed judges it for free. Free tier: 30 scored generation requests a month.
Tools powered by this rubric