菜单
Write social media posts for X (Twitter) and LinkedIn that avoid AI tells, pass platform algorithms, and drive engagement. Use when drafting posts, threads, or content for X or LinkedIn.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Structure Claude prompts for clarity and better results using roles, explicit instructions, context, positive framing, and strategic organization. Use when crafting prompts for complex tasks, long documents, tool workflows, or code generation.
Pause agent execution to have the human validate decisions, choose between options, answer freetext, or comment on a document — via the `hl` CLI. Every interaction is a kickoff that returns a job handle immediately; collect the human's answer later with `hl job result`. Use for material design decisions, approval gates, picks between meaningful alternatives, and markdown doc review. Not for trivial yes/no confirmations the agent should decide itself.
Validate technical designs against requirements for coverage, feasibility, and architectural soundness. Use after creating a design.
Validate implementation plans against requirements and design. Use after creating a plan to ensure full coverage and no ambiguities.
Validate requirements documents for EARS compliance, completeness, and consistency. Use after creating requirements.
Design CLIs for humans and LLM agents — subcommand shape, output streams, exit codes, JSON modes, TTY-aware color, structured errors, mutation safety. Use when building or refactoring a CLI, adding machine-readable output, writing --help, deciding stdout vs stderr, or making a tool agent-friendly.
| name | LLM Social Posting Guide |
| description | Write social media posts for X (Twitter) and LinkedIn that avoid AI tells, pass platform algorithms, and drive engagement. Use when drafting posts, threads, or content for X or LinkedIn. |
You are an LLM. People can tell. The average person scrolling X or LinkedIn in 2025+ has developed a finely tuned radar for AI-generated content. On LinkedIn, over 54% of long-form posts are now AI-generated — and those posts get 45% less engagement than human-written ones. LinkedIn's algorithm actively detects and suppresses obvious AI content, reducing reach by up to 30%. On X, the problem is different but equally fatal: AI posts are bland in an environment that rewards sharp, opinionated, personality-driven writing. Your defaults will get you scrolled past on both platforms.
This guide exists because the way you naturally write — balanced, comprehensive, eager to cover all angles — produces posts that die in the feed. Social posting rewards the opposite: incomplete thoughts that provoke completion, strong positions that provoke disagreement, and raw specificity that provokes recognition.
Your goal is not to write a good post. Your goal is to write something that makes people stop scrolling.
X's algorithm ranks content on three pillars: relevance, recency, and engagement velocity. That last one matters most for you. A post that gets 50 replies in 10 minutes will massively outperform one that gets 500 likes over 24 hours. The algorithm makes its decision about your post in the first 60 minutes. If it doesn't catch fire early, it's dead.
What this means for you: the first line of every post has to stop a thumb mid-scroll. Not impress. Not inform. Stop.
X also deprioritizes posts with external links. If the person you're writing for wants to link somewhere, put it in a reply to the main post, not the post itself. Posts with links in the body get algorithmically buried.
Text posts still outperform video and images for raw engagement on X, despite the platform pushing video. Threads generate roughly 3x the engagement of equivalent single posts. These are the formats where your writing matters most.
Everything from the original outbound guide applies here, but X has its own additional tells. X is the most informal major platform. The tolerance for AI-sounding language is zero.
The Thread Opener Formula. You default to: "I spent [time period] studying [topic]. Here's what I learned: 🧵" This format was effective in 2022. It is now the single most recognized AI/growth-hacker pattern on X. Everyone sees through it. Worse, engagement pods and bot accounts use this exact template, so the algorithm may actively suppress it.
Emoji as bullet points. You love to structure threads like:
✅ Point one
✅ Point two
✅ Point three
Or using 🔥, 🚀, 💡, 🧵, 👇 as structural elements. One or two emoji in a post is fine. Using them as a formatting system screams automation. Real humans don't organize their thoughts with emoji.
The false authority opener. "Most people don't know this, but..." or "The top 1% of [practitioners] do this differently..." You have no experience. You haven't observed anyone doing anything. These framings are dishonest and people sense it. Write from the perspective of the person posting, or write observations without claiming a vantage point you don't have.
Numbering everything. "7 lessons I learned..." "5 tools that changed my workflow..." Numbered lists are fine occasionally. But you default to them for every thread because they're structurally easy for you. Vary the format. Tell a story. Make an argument. Describe a process. Not everything is a listicle.
The wrap-up tweet. You always end threads with "TL;DR:" or "To summarize:" followed by a neat bow. Real threads often just... end. Or they end with a question. Or a provocation. The tidy summary is an LLM signature.
Hashtag stuffing. On X, 1-2 hashtags maximum. Often zero. This is not Instagram. A post with 5 hashtags looks like spam and the algorithm treats it accordingly.
You're too complete. X rewards incomplete thoughts. A post that says everything leaves nothing to reply to. The best-performing posts on X leave a gap — they make a claim without fully defending it, tell a story without explaining the lesson, or ask a question that doesn't have an obvious answer. You instinctively want to be thorough. Fight that instinct. Leave room.
You don't have opinions. X is an opinion platform. The posts that spread are the ones that take a position. "Hot take: most startups don't need a CTO until series B" will outperform "There are many factors to consider when determining the right time to hire a CTO" every single time. The first provokes replies (agreement AND disagreement). The second provokes nothing.
You hedge everything. "It's worth considering that..." "In many cases..." "While there are exceptions..." Stop. On X, hedging kills engagement. Say the thing. If it's wrong, people will tell you in the replies, which is actually what you want — replies drive distribution.
You sound like a teacher. Your default posture is explaining things to someone who doesn't know them. On X, the best-performing educational content sounds like someone thinking out loud, sharing what they noticed, or describing what happened to them. Not lecturing.
Observations, not lessons. "I noticed that every founder I've talked to this month is cutting their marketing team but increasing their content budget" works better than "5 reasons founders should invest in content over marketing teams." The observation invites discussion. The listicle invites skimming.
Contrarian takes. "Unpopular opinion: remote work is making junior engineers worse" will spread because people have strong feelings about it. The disagreement IS the engagement. But the take has to be genuinely held or at least defensible — not contrarian for its own sake.
Specific numbers and details. "We switched from Slack to Discord for internal comms and saved $14k/year across 50 people" is interesting. "Communication tools can significantly impact your bottom line" is nothing.
Behind-the-scenes honesty. "Our launch flopped. 200 signups when we projected 2,000. Here's what went wrong." Vulnerability with specifics massively outperforms success stories on X. People are tired of humble brags. They crave honesty about failure.
Screenshots and evidence. A screenshot of real data, a real conversation (redacted), a real dashboard, or a real result is worth more than any amount of well-crafted prose. X audiences trust evidence. If the poster has something to show, show it.
The one-liner. Sometimes the best post is one sentence. "The best sales strategy is having a product people actually want." Short, punchy, true. You tend to over-elaborate. Resist.
If writing a thread:
First tweet must stand alone. It needs to be interesting enough that someone would engage with it even if the thread didn't exist. Don't waste it on "A thread on X 🧵" — that's not a hook, it's a label.
One idea per tweet. Don't try to cram two points into one tweet to save space. White space and line breaks matter. Let each tweet breathe.
Vary tweet length within the thread. If every tweet is exactly 280 characters, it looks automated. Some should be short. Some longer. Like actual human thinking.
Don't number the tweets unless the order truly matters (like steps in a process). Numbering everything is an AI and growth-hacker tell.
Include a tweet that's just a story or example. Break up the "lesson" format with a concrete anecdote somewhere in the middle. This is where threads become human.
End with something open. A question, a provocation, an admission of uncertainty. Not a summary. "I'm still figuring this out" is a better thread closer than "In conclusion, these 7 strategies will transform your business."
Study the people who actually build audiences on X. The patterns you'll notice:
Before publishing any post for X, run through this:
The Scroll Test. Read just the first line. Would you stop scrolling for this? If the first line is setup or context, rewrite it. Lead with the most interesting thing.
The Reply Test. After reading this post, is there something obvious to reply with? Agreement, disagreement, a similar experience, a question? If the post is so complete that there's nothing to add, it will underperform. Intentionally leave a gap.
The Attribution Test. If you removed the poster's name, could this post have been written by any of 10,000 generic accounts? If yes, it's not specific enough. Add a detail, opinion, or experience that only this particular person would have.
The Word Scan. Same as the outbound guide. Ctrl+F for: leverage, optimize, comprehensive, robust, innovative, landscape, furthermore, moreover, streamline, delve, harness, elevate, foster, empower, seamless, cutting-edge, holistic, actionable, ecosystem, paradigm. Replace all of them with normal words. On X, also watch for: "game-changer," "let that sink in," "read that again," "this is huge," and "here's the thing" — all are clichés of growth-hacker Twitter.
The Emoji Scan. More than 2 emoji in a single tweet? Cut some. Emoji used as bullet points or structural markers? Restructure. 🧵 in the first tweet? Remove it — the thread format is self-evident.
The Length Check. Single post over 200 characters? Consider whether it could be shorter and punchier. Thread over 10 tweets? Cut it to the 6-7 strongest. First tweet over 100 characters? See if you can tighten the hook.
The Voice Check. Read it out loud. Does it sound like someone talking to a friend at a bar about their work? Or does it sound like a keynote presentation? X rewards the former.
LinkedIn's algorithm in 2025 operates through a four-stage process: quality filtering (immediate spam/quality classification), engagement testing (first 60-90 minutes), network ranking, and a 24-hour final push. The first 60 minutes are your "golden window" — strong early engagement means LinkedIn shows the post to second and third-degree connections. Weak early engagement means the post dies.
Key algorithm facts that should shape every post:
Dwell time is a primary ranking signal. LinkedIn measures how long someone spends looking at your post, even if they don't react. This means longer, more substantive posts can outperform short ones — but only if they're genuinely engaging enough to hold attention. A 1,500+ character post that keeps people reading will massively outperform a 200-character hot take that people glance at and scroll past. This is the opposite of X.
LinkedIn penalizes AI-generated content. Their detection systems suppress obvious AI content, resulting in roughly 30% less reach and 55% less engagement. This is not theoretical — they published this. Every AI tell in your post is actively reducing its reach.
External links kill reach. Posts with a single link perform worst. The platform wants people to stay on LinkedIn, not click away. If a link is necessary, put it in the first comment, or bundle 4+ links together (which counterintuitively signals "resource" rather than "promotion" and performs better).
Native content gets priority. Carousels get the most impressions (~1,387 average), followed by images (~703), videos (~672), and text-only (~589). If the post can include a carousel or image, it should.
Comments > likes. LinkedIn weights thoughtful comments far more heavily than reactions. A post with 15 substantive comments will outreach a post with 200 likes and 2 comments. This means your post needs to provoke discussion, not just agreement.
Don't post more than once per 24 hours. A new post can actually suppress the reach of your previous post if that one is still gaining momentum. 3-4 posts per week is the sweet spot.
Hashtags barely matter anymore. Despite conventional wisdom, recent research shows hashtags have essentially no impact on reach on LinkedIn in 2025. Use 0-3 if you want. Don't stress about them.
LinkedIn is already drowning in AI content. The bar for detection is low. People are actively looking for tells and posting about them. A Bloomberg article from early 2026 documented users scrutinizing em dashes, emoji patterns, and repetitive phrasing to call out AI-generated posts. You need to be better than the average ChatGPT output, which frankly is a low bar.
The "Agree?" closer. You end LinkedIn posts with "Agree?" or "Thoughts?" or "What do you think? 👇" This is classified as engagement bait, and LinkedIn's algorithm specifically detects and penalizes it. Ask a real question instead — one that requires actual thought to answer.
The humblebragger structure. "I was rejected from 47 jobs. I slept on my friend's couch. I ate ramen for a year. Today, I'm the CEO of a $10M company." This format was parodied into oblivion by 2023. It's now the canonical example of LinkedIn cringe. The Viral Post Generator AI trained on 100,000 LinkedIn posts produces exactly this format by default. That should tell you everything.
Single-sentence paragraphs stacked vertically. Like this.
Every sentence on its own line.
For no reason.
Other than to game dwell time.
This format was explicitly called out as engagement bait by LinkedIn in 2024. It's transparent and cringe-inducing. Use actual paragraph structure.
The emoji wall. 🎯 Point one. 💡 Point two. 🚀 Point three. ✅ Point four. Same problem as X, but worse on LinkedIn because it looks childish in a professional context.
The motivational quote opener. "Steve Jobs once said..." "As Einstein put it..." This is the single fastest way to signal AI on LinkedIn. You love quoting famous people. Stop. If the person posting has their own insight, lead with that.
The "I'm grateful/excited/humbled" post. "I'm thrilled to announce..." "I'm humbled to share..." "Grateful for the opportunity to..." These words are so overused on LinkedIn that they've become semantic noise. People's eyes glaze past them. Start with what actually happened.
"Let me know if you need any modifications." This is literally a ChatGPT artifact that people accidentally paste into their LinkedIn posts. Obviously never include this, but it symbolizes a larger problem: your transitions, closings, and structural phrases are recognizably template-driven. Vary them.
You write press releases, not posts. Your default LinkedIn voice sounds like a corporate communications team drafted it. "We are pleased to announce our strategic partnership with..." Nobody talks like this. Even on LinkedIn. Write like a human professional talking to a peer.
You moralise everything. You can't describe a business event without extracting a life lesson from it. Sometimes a thing happened and it's interesting on its own. You don't need to add "The lesson? Never give up on your dreams" to every anecdote.
You're performatively humble. "I don't have all the answers, but..." "I'm still learning, but..." These hedges don't read as humble. They read as formulaic. If you have something to say, say it with conviction. Real humility comes across in specificity and honesty about what went wrong, not in pre-emptive self-deprecation.
You sycophantically praise everyone. "I'm so inspired by [person]'s incredible journey..." "The amazing team at [company]..." This reads as hollow on LinkedIn because it usually is hollow. If you're referencing someone, say specifically what they did that mattered. Not that they're "incredible" or "amazing."
Personal career stories with real stakes. "I got promoted to VP at 29 and immediately regretted it. I wasn't ready and everyone knew it except me. Here's what the next 18 months looked like..." This works because it's specific, vulnerable, and useful to others in similar positions. The dwell time on posts like this is enormous because people are reading every word.
Framework or process breakdowns. "Here's the exact process I use to evaluate whether a startup is ready to hire a head of sales. 5 questions, and if you can't answer them all, it's too early." This works because it's immediately actionable. People save these posts for reference.
Contrarian professional takes. "Unpopular opinion in sales: discovery calls are mostly a waste of time for deals under $50k." LinkedIn's algorithm rewards posts that generate discussion. Professional disagreement is discussion.
Lessons from specific failures. Not "failure is a teacher" platitudes. Specific: "We spent $40k on a conference booth and got 3 leads. Here's every mistake we made." The specificity is what makes it credible and shareable.
Data and results with context. "We tested 4 different onboarding flows over 6 months. Version C won by 34%. Here's what it did differently..." LinkedIn's professional audience responds to evidence. Numbers with context outperform assertions without them.
Industry commentary on specific events. "The Figma/Adobe deal falling through changes the competitive landscape for design tools in a way most people aren't seeing yet..." Commentary on real events, with a specific angle, positions the poster as someone worth following.
The hook (first 2-3 lines). On mobile, LinkedIn truncates posts after about 2-3 lines with a "...see more" link. Everything depends on whether someone clicks that link. Your opening lines need to create curiosity, make a bold claim, or establish immediate relevance. Don't waste them on context or preamble.
Good hooks:
Bad hooks:
The body. Write in natural paragraphs. Vary paragraph length — some 1-2 sentences, some 3-4. Include specific details, names, numbers, and dates where possible. If telling a story, include dialogue or direct quotes. If making an argument, include evidence.
Use line breaks between paragraphs for readability on mobile, but don't do the one-sentence-per-line thing. There's a middle ground between a wall of text and a staircase of single sentences.
The close. End with one of:
Do NOT end with: "Agree?", "Thoughts? 👇", "Share if this resonated!", "Like and repost for more content like this!", or any other engagement bait.
Carousels (document/PDF posts) get the highest reach on LinkedIn. If the content lends itself to it, this format is worth using.
What makes a good carousel:
What makes a bad carousel:
The "See More" Test. Look at only the first 2-3 lines. Would you click "see more"? If the opening is setup, context, or throat-clearing, rewrite it. The hook must earn the click.
The Dwell Time Test. Is this post substantive enough that someone would spend 30+ seconds reading it? If it's thin — a vague observation with no supporting detail — the algorithm will bury it regardless of how well-written it is. LinkedIn rewards depth.
The AI Detection Test. Read the post and honestly ask: could LinkedIn's algorithm flag this as AI-generated? Check for: perfect parallel construction, em dashes (limit to one or zero), banned vocabulary (same list as outbound plus "thrilled," "humbled," "grateful," "incredible journey"), the staircase format, engagement bait closers, and unnaturally smooth transitions.
The Specificity Test. Does the post contain at least one of: a specific number, a named company or person (where appropriate), a date or timeframe, a concrete example, or a direct quote? If it's entirely abstract, add something concrete.
The Comment Test. After reading this post, is there something non-obvious to say in the comments? If the only available comment is "Great post!" or "So true!", the post isn't generating discussion-worthy content. Rewrite to include something debatable or incomplete.
The Cringe Test. Would this post end up as a screenshot on the @StateOfLinkedIn parody account or in someone's "AI posts are ruining LinkedIn" thread? If there's any chance: rewrite.
The Voice Test. Does this post sound like the specific human posting it? Or could it have been written by any of a million LinkedIn users? A good LinkedIn post reflects a recognizable voice. It has quirks. Preferences. Specific references. If it's generic, it won't build an audience.
These apply to both X and LinkedIn:
Verbs to never use: delve, explore (as an opener), leverage, utilize, harness, optimize, enhance, streamline, empower, facilitate, elevate, navigate, foster, underscore, augment, endeavor, spearhead, unlock, supercharge, transform (as in "transform your business")
Adjectives to never use: comprehensive, robust, pivotal, cutting-edge, seamless, innovative, dynamic, meticulous, invaluable, groundbreaking, holistic, bespoke, actionable, best-in-class, game-changing
Nouns/phrases to never use: landscape (as in "the AI landscape"), realm, tapestry, paradigm shift, synergy, ecosystem, framework (in casual contexts), "the intersection of X and Y," deep dive, thought leader, thought leadership, value proposition, North Star, alignment
Transitions to never use: Furthermore, Moreover, Additionally, It's worth noting that, In today's fast-paced world, As you may know, That said, Having said that, With that in mind
LinkedIn-specific never-use: thrilled, humbled, grateful, incredible journey, honored, blessed, passionate about, game-changer
X-specific never-use: "let that sink in," "read that again," "this is huge," "here's the thing," "a thread 🧵"
What to use instead: Normal words. "Use" not "leverage." "Improve" not "optimize." "Help" not "empower." "Works with" not "seamlessly integrates." Write at a 6th-grade vocabulary level.
The triple parallel list. You love grouping in threes with identical grammar. "We help teams move faster, sell smarter, and close more deals." Humans don't think in perfectly balanced triads. List two things. Or four. Or describe something without listing at all.
The "It's not about X, it's about Y" formula. Stop using this entirely. It's become an AI fingerprint.
The compliment-pivot-pitch. On LinkedIn: "I saw [person]'s post about [topic]. That's exactly why..." On X: "Everyone's talking about [thing]. But what they're missing is..." Both are transparent structures.
Perfectly smooth transitions. Real social posts don't flow like essays. They jump. They start sentences with "And" and "But." They use dashes — to interrupt themselves. They leave hanging thoughts. Your posts flow too smoothly.
The balanced take. You present both sides of everything. Social media rewards positions, not balance. Take a side. If there are caveats, save them for the comments if someone challenges you.
Em dashes. You love them. Zero preferred, one per post absolute maximum. They are the single most-discussed AI tell on both platforms now. When used, never put spaces around them (write like—this, not like — this). Spaced em dashes are an even stronger AI signal.
Semicolons. Never in social posts. Period.
Exclamation points. Extremely sparingly. One per post maximum, and only if the thing is actually exciting. "Just raised our Series A!" is fine. "Here are 5 productivity tips!" is weird.
Ellipsis. Humans use these occasionally for trailing off or building suspense. You almost never use them, which is itself a tell. An occasional "..." is more human than your perfectly punctuated everything.
Posts don't exist in isolation on either platform. The person posting needs to be an active participant in conversations, not just a broadcaster.
On X: Spend more time replying to others than posting original content. Reply to mid-size accounts (5K-100K followers) in the relevant niche. Thoughtful replies that add genuine value can drive more profile visits and follows than original posts. Quote-tweet interesting things with a specific take (not "this is so good" — say why or add a new angle).
On LinkedIn: Comment on others' posts before and after publishing your own. This "warms up" the algorithm and increases the chances that your post will appear in feeds. Reply to every comment on your own posts, especially in the first hour. A post with active author engagement in the comments gets pushed to more feeds.
On both platforms: Never be the account that only posts and never engages. That pattern looks automated and both algorithms penalize it.
Don't try to cover every topic. Pick 2-3 core themes that the poster genuinely has experience with and rotate between them. Consistency on a narrow range of topics builds an audience faster than variety.
Content mix for X (weekly):
Content mix for LinkedIn (weekly):
X: Weekday mornings perform best, specifically Tuesday-Thursday, 8-11 AM in the poster's target audience timezone. Avoid late nights and weekends.
LinkedIn: Tuesday-Thursday mornings, 8-10 AM local time. Weekend posts actually can perform well due to less competition (only 8% of LinkedIn content is posted on weekends), but engagement velocity may be slower. Never post more than once per 24 hours.
On both platforms: the first 60 minutes after posting are critical. The poster should be online and engaging with any responses during this window.