Scrub AI Tells from Instagram Captions with ig-humanizer

Where this fits

You've just used an LLM to draft an Instagram caption for your client's product launch. It reads like a press release: em dashes everywhere, "leverage" and "delve" in the same sentence, a perfect rule-of-three list, and a cluster of rocket emojis. It's obviously AI-generated, and your client will notice. This is exactly where ig-humanizer fits: it cleans up the artificial patterns and adds the small, human irregularities that make captions feel authentic. For developers building social media automation agents, this skill prevents the embarrassment of publishing content that screams "bot."

Why agents benefit

• Removes known AI tells: Automatically strips em dashes, AI buzzwords (leverage, delve, elevate), and rule-of-three structures—the telltale signs that an LLM wrote the text.
• Adds human fingerprints: Introduces casual phrasing, intentional grammar choices, and personal asides that make captions feel like they came from a real person.
• Pre-publish audit mode: --mode audit checks the caption against a 2026 Instagram-specific checklist—first 125 characters, length, hashtag sizing, emoji usage, CTA, and media pairing—so you catch issues before posting.
• Handles carousels too: Works on on-image text for carousel slides, not just the main caption, making it versatile for multi-slide content.
• Clears trigger words: Can be invoked with simple phrases like "humanize" or "de-AI my caption," allowing your agent to respond to natural language commands.

Practical scenarios

Post-draft cleanup for a social media scheduler

Your agent drafts a week's worth of Instagram posts using an LLM. Before scheduling, it runs each caption through ig-humanizer. The output removes the mechanical rhythm and gives each post a distinct, human voice, so you don't have to manually edit 20 captions.

Pre-publish audit for a client approval workflow

Your agent has already written a caption that the client approved. But you want a final safety check. The audit mode flags that the hook doesn't capture attention in the first 125 characters or that hashtags are inconsistently sized—issues you can fix automatically before the post goes live.

Rewriting carousel slide text for authenticity

You generate a 5-slide carousel with on-image text that's dense and formal. ig-humanizer rewrites each slide's text to be more conversational and scannable, matching the casual tone that performs better on Instagram.

Add it to your agent workflow

Install the skill with:

npx skills add sergebulaev/instagram-skills --skill ig-humanizer

Then, in your agent's configuration, you can invoke it with a trigger like:

humanize this caption: "Our new solution leverages cutting-edge AI to elevate your marketing strategy..."

The skill will return a cleaned-up version. For a pre-post check, use:

audit before posting: <caption>

Compare before adopting

Before committing, evaluate ig-humanizer against other approaches: test whether the rewritten captions still retain your brand's core message and tone. Compare the audit checklist with Instagram's latest algorithm changes—does it address the current hook length and hashtag best practices? Consider alternatives like general-purpose text rewriting prompts or other social media skills, but weigh their output against the specialized Instagram-specific patterns this skill targets. Also check the skill's maintenance cadence and how quickly it adapts to platform updates.

Why it is worth tracking

This skill directly tackles a widespread pain point: LLM-generated social media content that reads as robotic. With 17 stars on GitHub, it's early but shows community interest. The author has bundled it with complementary skills (ig-caption-writer, ig-carousel-planner), suggesting an ecosystem tailored for Instagram automation. Keep an eye on how often it updates and whether it incorporates new AI-tell patterns as they emerge. If you're building agents that need to produce authentic social media content, this skill is a strong candidate to test now and track over time. For more details and updates, see the skill page on OpenAgentSkill.