Social Media Content Automation

The Problem

Creating social media content takes time. You need catchy captions, relevant hashtags, and fresh ideas, often for multiple platforms. An AI agent can generate all of this in seconds.

What we’re building: An agent that takes a topic or business type and produces ready-to-use social media content including post ideas, captions, and hashtags.

The Approach

We’ll create a system prompt that turns any AI platform into a social media content agent. The agent will:

• Ask for the topic, platform (Instagram, Twitter/X, LinkedIn, Facebook), and tone
• Generate multiple post ideas with captions
• Include relevant hashtags
• Format everything ready to copy and paste

Build It

Step 1: Set up your system prompt

Copy this system prompt into your platform:

Claude

Go to claude.ai and start a new conversation. At the beginning of your first message, paste the system prompt below, followed by your actual request.

Alternatively, create a Project in Claude and set this as the project instructions.

ChatGPT

Use Custom Instructions (Settings > Personalization > Custom Instructions) or paste the system prompt at the start of a new conversation at chatgpt.com.

If you have a paid plan (Plus, Team, or Enterprise), you can also create a Custom GPT with these instructions for reuse.

Gemini

Go to gemini.google.com and paste the system prompt at the start of a new conversation.

If you have Gemini Advanced, you can also create a Gem with these instructions for reuse.

Ollama

Create a Modelfile with the system prompt:

FROM llama3.2
SYSTEM """<paste system prompt below>"""

Then run: ollama create social-media -f Modelfile

LM Studio

Open LM Studio, load your preferred model, and paste the system prompt into the System Prompt field in the chat settings panel.

The system prompt to use:

You are a social media content specialist. When given a topic, business type, or theme, generate engaging social media content.

For each request, provide:
1. Three post ideas with full captions (ready to copy and paste)
2. A set of 3-5 highly relevant hashtags per platform
3. Best time to post suggestion
4. A content variation for each major platform (Instagram, Twitter/X, LinkedIn)

Guidelines:
- Keep Instagram captions engaging with a hook in the first line
- Keep Twitter/X posts under 280 characters
- Make LinkedIn posts professional but conversational
- Include a call-to-action in each post
- Suggest emoji usage where appropriate
- Adapt tone based on the business type (professional, casual, fun, etc.)

Always ask the user for: topic/business type, target audience, and preferred tone if not provided.

Step 2: Test with a sample request

Try this prompt to test your agent:

Create social media content for a small coffee shop called "Morning Brew"
that specializes in locally sourced beans. Target audience is young
professionals aged 25-35. Tone should be warm and inviting.

Your agent should respond with multiple post ideas, hashtags, and platform-specific variations.

Step 3: Iterate and refine

Try different follow-up requests:

• “Make the tone more playful”
• “Add a post about a new seasonal drink”
• “Create a week’s worth of content”
• “Focus on Instagram Reels ideas”

Notice how the agent remembers the context of your coffee shop and builds on it.

Test It

Try your agent with these different scenarios:

• A fitness trainer promoting online classes
• A local bookstore announcing a reading event
• A freelance photographer showcasing their portfolio
• Your own business or project idea

Make It Better

Once your basic agent works, try these improvements:

• Add brand voice rules to the system prompt (e.g., “Never use slang,” “Always mention sustainability”)
• Include content calendar logic: ask the agent to plan a full week or month
• Add competitor awareness: tell the agent about competitors to differentiate the content
• Save winning prompts: when you get great output, save the exact prompt for reuse

Self-Modifying Improvement

After each session, note what worked and what didn’t. Add rules to your system prompt like “Always start Instagram captions with a question” or “Never use more than 3 emojis per post.” This is the self-modifying prompt pattern in action.

Sample Output

Here is an example of what your agent should produce when given the coffee shop prompt:

**Post Idea 1 — Instagram**

☕ What if your morning coffee was brewed from beans grown just 50 miles away?

At Morning Brew, every cup tells a local story. Our single-origin beans are roasted fresh each week, and you can taste the difference.

Come discover your new favorite brew. Link in bio for this week's featured roast.

#MorningBrew #LocalCoffee #FreshRoasted #CoffeeLovers #SupportLocal

**Post Idea 1 — Twitter/X**

Your morning coffee was grown just 50 miles away. ☕ At Morning Brew, we source locally and roast fresh weekly. Come taste the difference. #LocalCoffee #FreshRoasted

**Post Idea 1 — LinkedIn**

At Morning Brew, we made a commitment to source our beans from local producers within 100 miles of our shop. The result? Fresher coffee, a smaller carbon footprint, and stronger relationships with the farmers who grow our beans.

If you're in the area, stop by and ask about our featured roast this week.

📅 Best time to post: Tuesday or Wednesday morning (7-9 AM), when your audience is starting their workday and thinking about coffee.

If your output doesn’t look like this, with separate platform versions and specific hashtags, refine your system prompt.

Troubleshooting

The output is too generic or sounds like every other brand. Add more specific details to your request: your brand personality, what makes you different, your location, and specific products. The more context you give, the more unique the output.

The hashtags don’t seem relevant or real. Verify hashtags by searching for them on the actual platform. If the agent suggests hashtags that don’t exist or have very few posts, add this to your system prompt: “Only suggest hashtags that are commonly used and have active communities.”

The tone doesn’t match what I want. Be very specific about tone. Instead of “warm and inviting,” try “friendly and casual, like texting a friend who loves good coffee. Use short sentences, questions, and 1-2 emojis per post.”

All three post ideas are too similar. Add this to your system prompt: “Make each post idea distinctly different: one should be educational, one should be promotional, and one should be community-focused or behind-the-scenes.”

Free Tools to Enhance This Project

• Buffer: Free tier lets you schedule posts on up to 3 social media channels. Use it to schedule the content your agent creates.
• Canva: Free tier for creating social media graphics. Pair your AI-generated captions with eye-catching visuals.
• Google Trends: Free tool for validating whether a topic is trending before you create content around it.
• Hashtagify: Free tier for researching hashtag popularity and finding related hashtags.

Making It Better Over Time

This is where the self-modifying prompt pattern becomes powerful. After each batch of content:

1. Note what performed well: which posts got the most engagement?
2. Add rules: “Posts with questions in the first line get 2x engagement, so always start with a question for Instagram”
3. Remove what doesn’t work: “Stop suggesting hashtags with fewer than 10,000 posts”
4. Build a brand voice guide: over time, your system prompt becomes a detailed style guide unique to your brand