LinkedIn Post Audit Checklist
Run every draft through this before publishing. Any FAIL = fix before posting.
Hook (first 3 lines)
- Strongest insight is in line 1 (inverted pyramid structure)
- Reader would click "see more" — no throat-clearing, no "I want to share..."
- First 265 chars earn the mobile expand click (desktop gets 210 chars)
- No rhetorical question opener ("Have you ever...?")
- No "In today's fast-paced world"
Length & format
- 300-400 words (~900-1,300 chars), 20+ short sentences
- Line breaks between every 1-2 sentences
- At least one bold or listed element for scanners
- If long-form (1,500-1,900): double line-breaks between ideas
Substance
- Contains a real, specific moment (date, number, name, failure)
- Save-worthy element present: framework, template, checklist, or data point
- Not a humble-brag; if celebrating, paired with what broke
- At least 1 odd-precision number WITH a named referent (who / what / when / what it cost). A bare number does not count
- At least 1 named entity (real person, company, date, city)
- At least 1 first-person sensory detail
Close
- Ends with a genuine open question (not "Comment YES")
- No manufactured CTA bait ("tag someone who…", "drop a 🙌")
- Specific question beats generic "Thoughts?" by 20-40%
- Not "What do you think?" (dead prompt)
Algorithm safety
- Zero external links in post body (move to first comment if needed)
- Not framing LinkedIn as inferior to other platforms
- Publishing from personal profile (not company page — 6-8x reach difference)
- Publish time matches author's usual slot (audience training)
- Not the 2nd post today (360Brew cannibalizes 2+/day)
Post-publish plan (first 90 min)
- Author will engage 15-30 min before posting (warmup window)
- Author replies to every comment within the first hour
- No structural edits in first 60 min (typos only)
- Author engages with 5+ other posts in the first 30 min after publishing
- Seeding: 3-5 self-comments within 15-30 min to create thread depth
AI-tell scrub (see audit-ai-tells.md)
- Passed audit-ai-tells.md auto-fail check (forensic leakage, reveal bridges, negative parallelism)
- Em dash density at or under ~1 per 100 words (1-2 per post). Zero is not required; zero everywhere is its own tell
- No paragraph with 3+ vocabulary / grammar markers (one marker per paragraph is fine)
- No staccato stacks ("Short. Punchy. Done.", "No X. No Y. Just Z."), no one-word paragraphs, at most 2 standalone fragments in the post
- At most one natural rule-of-three; no stacked or perfectly parallel triads
- No sincerity announcements as opener or pivot ("let me be honest", "I'll be real", "honestly?", "to be direct")
- Passed humanizer scrub-rules.md (V3 density rules). Detector scores are not a check: they are noise at this length
Voice match
- Capitalizes all personal names, company names, product names
- Matches user's established voice (lowercase sentence starts OK,
..soft pauses OK, one em dash OK) - One specific, dated, uncomfortable fact stated flat, with no framing sentence around it
- Author's reactions and opinions survived the scrub (a uniformly flat tone is a humanizer fingerprint)
- Not name-dropping own product more than once
Edit-safety window
- No significant edits planned for first 3 hours (resets distribution)
- Structural restructuring >20% of text = reset the algorithm, don't do it
- Typo fixes safe after the 90-min momentum window
Scoring
- Pass: all items checked, no FAIL
- Warning: 1-3 unchecked — ship with caveats
- Fail: 4+ unchecked OR any Algorithm Safety item fails — rewrite before posting
When to override the checklist
- Time-critical newsjack: algorithm safety still mandatory, voice match can be looser
- First-ever post in a pillar: warmup / seeding steps can be skipped once, not twice
- Founder confession post: "not humble-brag" rule can stretch if the vulnerability is central to the thesis