Social post prompts. Platform-native, every time.
Pick a platform, post goal, format, voice, length, CTA, and hashtag style; get a ChatGPT, Claude, or Gemini-ready brief that respects platform constraints and asks for three variations. Browser-only; nothing is saved.
Platform, goal, format, voice.
Your social brief.
Click build to generate.
What good looks like.
A LinkedIn post explaining why iOS attribution looks broken (and is not)
Goal: share-insight. Format: text-only, long-form. Voice: sharp-expert. Length: 6+ paragraphs. CTA: comment. Hashtag: none. The hook is a counter-consensus claim ("Your iOS attribution is not broken; the model assumed deterministic IDs that no longer exist"). Body walks through SKAdNetwork's three-tier model, what data does and does not survive ATT, and why the apparent drop in tracked conversions is the new floor. CTA: "what are you doing differently in your media plan since iOS 17 - reply with one tactic." This pattern works on LinkedIn because the platform rewards depth and the audience self-selects for ops detail. Best paired with output from ChatGPT or Claude.
An Instagram carousel caption for a launch-day skincare drop
Goal: announce-launch. Format: carousel-3. Voice: warm-personal. Length: medium. CTA: save. Hashtag: standard (5 to 10). Slide 1 (hook): a single line under the product photo - "Three years. One serum." Slide 2 (body): the formulation story, named ingredients, named formulator. Slide 3 (payoff): the price, the launch date, the link in bio direction. Caption supports without repeating; first 125 characters must hook before the "more" truncation. Best generated in Anthropic Claude because its prose is warmer than ChatGPT's defaults.
An X post on AI tooling and engineering productivity
Goal: hot-take. Format: text-only. Voice: sharp-expert. Length: cap-280. CTA: no-cta. Hashtag: none. Counter-consensus claim with stakes: "AI coding tools made the median engineer faster. Top engineers are unchanged. Bottom engineers are now invisible. The middle is the only segment whose role is structurally at risk." 240 characters, one specific stakes claim, ends without question or CTA. The format is intentionally austere - X rewards confidence, not invitations. Generate variations in OpenAI ChatGPT to test which framing lands.
A post is a first line with a body attached.
Eighty percent of a social post's outcome is the first line. The remaining twenty percent is the body and the CTA. AI tools default to writing a generic intro because the prompt did not tell them the platform-specific first-line rule. A working prompt names the platform, the cap, the first-line rule, the native style, the voice, and asks for three variations - then trusts the model to ship a post worth shipping. Skip any of those and you get a post that reads like every other AI-generated thread on the platform.
First-line is everything
Every platform has a first-line truncation point: 200 characters on LinkedIn before "see more", 125 characters on Instagram, 280 total on X. The first line decides whether anyone reads line two. The most common AI failure is starting with "I am excited to share" or "Here is a thought I had today" - both are fillers that burn the precious first 100 characters. The fix is to constrain the prompt: "the first line must be specific (named, numbered, or counter-consensus); no humble-brag patterns." After that, the model writes hooks that actually hook. See web.dev fundamentals for a deeper dive on attention economics.
Platform native is not cross-post
The fastest way to ruin a content calendar is to write one post and cross-post it everywhere. LinkedIn rewards depth and paragraph structure; X rewards punch and brevity; Instagram rewards visual-first with caption support; TikTok rewards a video hook with caption as backup. A good prompt names the platform first and writes for that platform's native rhythm. The same insight on iOS attribution becomes 6 paragraphs on LinkedIn, 280 characters on X, and a 15-second video hook on TikTok - three different posts, not one cross-posted. Tools that auto-cross-post produce content that ranks nowhere because it does not fit any platform's native style.
Voice consistency
Voice is what makes a post recognizable as belonging to a specific brand or person. A "sharp expert" voice on Tuesday and a "playful casual" voice on Wednesday from the same account is a credibility leak. Pick one voice per author and stay there. The exception is intentional voice-shifting between platforms: LinkedIn might lean sharp-expert while Instagram leans warm-personal for the same brand - that is fine because the platforms are different rooms. Within a single platform, voice should remain stable across at least 30 days of content before any shift. Voice consistency is what builds the parasocial trust that converts followers to customers; check the growth strategy service for how to map voice to funnel position.
Hashtag fatigue
The 30-hashtag-per-Instagram-post era is over. Reach algorithms now penalize keyword-stuffing patterns and reward content that earns saves, sends, and dwell time. The current rule of thumb: 1 to 3 niche-relevant hashtags on Instagram, none on LinkedIn unless the brand owns one (#OpenToWork, #BuildInPublic), none on X unless joining a real conversation. Pinterest is the only platform where keyword-dense descriptions still drive significant reach because Pinterest is functionally a search engine, not a feed. Reference: Google's SEO starter guide on the underlying signal-vs-noise principle.
CTA clarity
The most common CTA failure is a generic "what do you think?" - it produces no engagement because it asks for nothing specific. Better patterns: "Reply with one tactic you have changed since iOS 17"; "DM me the word PILOT if you want the spreadsheet"; "Save this for the next time someone asks". Specificity drives engagement; vagueness produces silence. The "no CTA" option is also valid - let a strong post breathe; not every post needs to ask for something. Posts that ask too often become asks rather than substance, and the audience tunes out. Use the conversion rate calculator to model expected lift from CTA changes if you run paid social.
Test with the bystander question
Before publishing, read the post out loud and ask: "Would someone who does not follow this account stop scrolling for this?" If the honest answer is no, the post is for the existing audience only - which means it will not grow the account. Posts that work in the algorithm are posts a stranger would stop for. The bystander question filters out inside-baseball posts, low-effort thoughts, and self-congratulation in a way no other heuristic does. Apply it to every post before publishing, especially after using AI generation - LLMs default toward inside-baseball because their training data is full of it.
Related work: Growth strategy for full content systems. Headline analyzer for scoring the first line. Hero prompt for the matching landing page hero. UTM builder for tracking link clicks from social. Open Graph preview for the link card preview.
Five answers.
Should I use ChatGPT, Claude, or Gemini for social posts?
Each handles a different voice better. ChatGPT is the most consistent at platform constraints (it remembers character caps and first-line rules well) and is strongest for LinkedIn long-form. Claude tends to write warmer, more personal prose - it is the right choice for Instagram captions, customer stories, and any "warm personal" voice setting. Gemini is strongest when the prompt requires real-time context (current events, trending topics) because of its search integration. The prompt this tool produces is generator-agnostic; it works with all three. Generate 3 variations in your preferred tool, then run the same prompt in a second tool to compare.
Why does the prompt always ask for 3 variations?
A single output anchors the model to one angle of attack. Three variations force the model to generate three distinct hooks, three different opening moves on the same topic. This gives you something to react to and combine - the strongest shipped post often takes the hook of variation 1, the body of variation 2, and the CTA of variation 3. Asking for one post and then asking for "another version" produces near-duplicates because the model lost context.
How many hashtags should I use?
Platform-specific. LinkedIn: 0 to 3 niche hashtags; the platform deprioritized hashtags significantly in 2023. X: 0 to 1 hashtag, only when joining a real conversation. Instagram: 1 to 5 niche-relevant; the era of 30-tag spam is over and reach is now penalized for keyword stuffing. TikTok: 3 to 5, including one trending tag if relevant; TikTok's discovery still uses hashtags as a primary signal. Pinterest: 5 to 10 keyword-dense; Pinterest is functionally a search engine. Threads: 0 to 1; the platform is too new for hashtag culture to have settled. Aggressive (15+) is reserved for explicit reach experiments only.
Can I use the same prompt across platforms?
No - that is the most common cross-post mistake. The topic stays constant, but the format, length, voice, and CTA should all shift per platform. The same insight becomes a long-form 6-paragraph post on LinkedIn, a 280-character punch on X, and a 15-second video hook on TikTok. Run this generator three times for a single topic - once per platform - and you get three native posts rather than one cross-posted thread. Cross-posting is the easiest way to look low-effort to anyone who follows you on more than one platform.
Does this tool save my prompts?
No. Every value you enter and every prompt that gets built lives in your browser tab only. Nothing is transmitted to a server, stored in a database, or synced across devices. Close the tab and the data is gone. The Copy prompt button puts the assembled brief on your clipboard; that is the only output path.
Posts are first lines.
Our content engagements pair a prompt library with platform-specific calendars, voice docs, and a four-week ramp on the channels that drive your funnel. Built for founders and operators who do not want to write every post themselves.