LinkedIn rewards engagement on the first line. If a reader doesn’t expand the post within 2 seconds, the algorithm buries it. Most AI-generated posts fail at this — they open with “I’m excited to share…” or “Today, I want to discuss…” and the rest doesn’t matter.

Our LinkedIn Post Generator tries to fix this with prompt engineering, not generation tricks. Here’s what’s actually happening behind the button.

The hook problem

A LinkedIn hook is the first line and a half of your post — what shows in the feed before “see more.” It has roughly 220 characters. Out of those, 80 characters are above the fold on mobile.

Three things make a hook work:

  1. Specificity. “I shipped 12 features in Q1” beats “I had a productive quarter.” The number forces the reader to update their priors.
  2. Tension or reversal. “I deleted half our engineering documentation last month — and our team’s velocity went up 30%.” The reader has to expand to find out why.
  3. A claim that opens a loop. “There’s one mistake I see senior engineers make at offer-negotiation time, and it costs them ₹3-5 LPA.” Now the reader needs to know what.

What kills hooks:

  • Vague openers (“I want to share…”, “Excited to announce…”)
  • Cliches (“Hard work pays off!”, “Mindset is everything”)
  • Self-deprecating fluff (“Just my two cents but…”)
  • Hashtags or emojis at the start (cuts immediately into the 220 chars)

What our prompt actually asks

When you click “Generate” on the LinkedIn Post Generator, we send roughly this to the LLM:

You are a LinkedIn ghostwriter who writes for senior engineers and 
product leaders in India. Audience reads on mobile during commutes.

Topic from user: [USER INPUT]
Tone: [professional / casual / contrarian]

Write a LinkedIn post with these constraints:

1. Hook (first line, max 220 chars):
   - Open with a specific number, a reversal, or a claim that 
     creates a curiosity loop.
   - DO NOT start with: "I", "Excited", "Today", "Just", "Recently"
   - DO start with: a number ("12 ways…"), a contrarian statement 
     ("Documentation is overrated."), a confession ("I burned 6 months 
     on this before realising…"), or a reader-direct address 
     ("Stop optimising for engineering velocity.")

2. Body (3-5 short paragraphs):
   - Each paragraph max 2 sentences.
   - Use line breaks generously for mobile readability.
   - Concrete examples > abstractions. If you say "we saved time", 
     specify how much (₹X, Y hours, Z%).
   - Don't use the words: "synergy", "leverage" (as a verb), 
     "ecosystem", "value-add", "stakeholder", "mindset"
   - Indian-context references where natural (rupee figures, 
     Indian companies, cultural references).

3. Close (1-2 lines):
   - End with a question OR a contrarian statement that invites 
     replies.
   - Avoid: "Thoughts?" — too lazy.
   - Try: "What's the most expensive lesson you learned from this?" 
     or "Disagree with anything here?"

4. Format: 4-7 lines total length. NO bullet points (LinkedIn 
   strips formatting). NO emojis at the start. Hashtags at the 
   bottom only, max 3, all lowercase.

Output the post directly, no preamble.

The system prompt is opinionated on purpose. Most LLM-generated LinkedIn posts feel generic because the prompt is generic (“write a professional post about X”). Constraints — banned words, hook patterns, line-break rules — produce variance.

What works (real examples)

Three hooks the prompt has produced for various inputs:

Input: “Talk about how I migrated 50 microservices from EKS to Lambda.”

“We migrated 50 microservices from EKS to Lambda. The infra bill dropped 60%. Here’s what nobody tells you about that decision.”

Input: “Sharing why I quit my Big Tech job to join a startup.”

“Quit Google last month for a Series A startup. The pay cut is ₹40 LPA. The decision took 4 hours, not 4 months. Here’s why.”

Input: “Lessons learned from building an AI coding tool.”

“We shipped a coding tool that wraps Claude. 80% of users churned in week 2. Here’s what we got wrong.”

Notice the pattern: number, specificity, reversal, curiosity loop. None of them start with “I am” or “Today” or “Excited.”

What flops (and why we filter against it)

If you’ve used Bard / Gemini / GPT-3.5-style models for LinkedIn posts, you’ve seen these bad hooks:

“In the rapidly evolving landscape of artificial intelligence, organisations must embrace innovative strategies to remain competitive…”

This fails because:

  • “Rapidly evolving landscape” is a corporate cliche
  • It’s third-person and abstract
  • 220 chars of nothing happening

Our prompt explicitly bans “rapidly evolving”, “landscape”, and the third-person framing. The LLM still occasionally drifts there; that’s why the system prompt repeats the bans rather than just listing them once.

The CTA problem

The closing line of a LinkedIn post matters almost as much as the hook. Most generic AI ends with “Thoughts?” or “What do you think?” — these get 0 engagement because they’re too lazy to invite a real reply.

Our prompt forces the LLM to pick from this menu:

  • A question that requires a specific opinion (“What’s the most expensive lesson you learned…”)
  • A contrarian statement (“Disagree? Tell me why I’m wrong.”)
  • A reverse poll (“Vote: should we ship this Friday or wait two weeks?”)
  • A pattern interrupt (“If you read this far, you’re either bored or you’ve made the same mistake.”)

These see ~3-5x the comment rates of “Thoughts?” in our internal A/B testing — measured across roughly 200 posts the tool has generated and tracked when users opted in.

When the AI gets it wrong

The generator is good. It’s not perfect. Common failure modes:

  1. Sounds like AI when the topic is too generic. “Tips for working from home” → expect cliches. The more specific your topic, the better the output.
  2. Fabricates statistics. If you don’t supply a number, sometimes the LLM invents one (“Studies show that 73% of employees…”). Always read the post before publishing and remove invented statistics.
  3. Repeats Indian context where it’s awkward. If your topic is purely technical (AWS infrastructure, Kafka), “rupee figures” don’t apply. Edit those out.
  4. Three-paragraph posts that should be one. Sometimes the LLM expands a 30-word idea into 200 words. Trim ruthlessly. Short LinkedIn posts outperform long ones in the feed algorithm.

How to actually use it well

  1. Start with a specific input. Not “leadership” — “the time I had to fire my best engineer for repeatedly missing deadlines.”
  2. Pick the right tone. Contrarian for opinion posts, casual for lessons-learned, professional for company milestones.
  3. Generate 3 versions. Pick the best hook from version A and the best body from version B; combine.
  4. Edit the close. The CTA is where AI most often ghosts you. Replace generic with personal.
  5. Read it out loud before posting. If it sounds like a press release, rewrite.

The tool is a draft accelerator, not a substitute. It compresses the 30-minute “stare at blank page” phase into 30 seconds. The 5-minute polish at the end is still yours.

Try the LinkedIn Post Generator on a topic you’ve been wanting to write about. If you hate the output, paste the rough draft into another AI tool and ask it to “rewrite the hook in the style of [author you respect].” That two-step usually beats either tool alone.