I Built an AI That Creates 90 Posts Per Month. Here's How
Three posts a day, across multiple platforms, every single day of the month. That's 90 posts minimum. Six months ago, that would have taken me (or a social media manager) 40+ hours of work. Now my AI system handles the first draft of all 90 in about 2 hours of batch work per week.
I'm going to show you exactly how I built this, what tools I use, and what the actual output looks like — including the stuff that doesn't work.
The Problem: Content Volume Is the Game Now
Every platform rewards consistency. TikTok's algorithm favors daily posters. Instagram pushes accounts that use every feature regularly. Facebook groups need constant content to stay active. X rewards frequent, punchy takes.
If you're running a brand (or several), you need volume. Not garbage volume — quality volume. Posts that actually get engagement, drive traffic, and build audience trust.
The math was simple for me: I manage content for my personal brand plus several client brands. At 3 posts per day per brand, that's hundreds of pieces of content per month. There was no way to do that manually without a big team. And I didn't want a big team. I wanted a system.
The Architecture: How It Actually Works
Here's the high-level flow of my content system:
- Input layer: Brand voice docs, content pillars, trending topics, past performance data
- Generation layer: AI creates first drafts of all content for the week
- Review layer: I review, edit hooks, cut weak posts, and approve
- Distribution layer: Approved content gets scheduled across all platforms
- Analytics layer: Performance data feeds back into the input layer
Each layer is mostly automated. The only part that's fully manual is my review — and I want it that way. If you've read about how agents work, this is basically an agent workflow with a human-in-the-loop.
The Tech Stack
| Component | Tool | Role |
|---|---|---|
| AI Brain | Claude + Custom GPTs | Content generation and adaptation |
| Orchestration | OpenClaw | Agent coordination and workflow automation |
| Scheduling | Buffer / Publer | Multi-platform post scheduling |
| Image Gen | Gemini Nano Banana Pro | Visual content creation |
| Voice | ElevenLabs | Voiceover for video content |
| Connectors | Zapier / Make | Glue between tools |
| Storage | Google Sheets + Notion | Content database and calendar |
Step 1: The Brand Voice Setup (Do This Once)
This is the most important step and the one most people skip. Your AI is only as good as the context you give it.
For each brand, I create a detailed voice document that includes:
- Brand personality (3-5 adjectives)
- Target audience demographics and psychographics
- Content pillars (3-5 core topics)
- Tone guidelines with examples
- Banned words and phrases (anything that sounds generic or AI-written)
- 10+ examples of posts that performed well
- 5+ examples of posts that flopped (and why)
- Platform-specific notes
This document becomes the system prompt for all content generation. I update it monthly based on what's actually performing.
Step 2: Weekly Content Generation Session
Every Monday morning, I run what I call a "content sprint." Here's the process:
Phase 1: Topic Research (15 minutes)
AI scans trending topics, competitor content, and audience questions. I use a combination of:
- Google Trends for search trends
- Reddit and Quora for real questions people are asking
- Competitor analysis (what's getting engagement in my niche)
- My own analytics (what topics performed last week)
Phase 2: Bulk Generation (30 minutes)
I feed the topics into my AI setup and generate all content for the week. For 90 posts/month, that's roughly 21-23 posts per week.
The AI generates each post with:
- Platform-specific formatting
- A hook (the most critical part)
- Body copy
- Call to action
- Hashtags (platform-appropriate)
- Image/visual direction notes
Phase 3: My Review (45-60 minutes)
This is where I earn my keep. I go through every post and:
- Rewrite weak hooks (about 40% of them need work)
- Cut posts that feel generic or forced
- Add personal anecdotes or references where appropriate
- Verify any claims or stats
- Approve or reject each post
Out of 23 AI-generated posts, I typically approve 18-20 and replace the rest with my own ideas or revised versions.
Phase 4: Visual Creation (30 minutes)
For posts that need images, I batch-generate visuals. Some use AI image generation, others use Canva templates. Video scripts go to my editors.
Step 3: The Distribution Pipeline
Approved content goes into a Google Sheet that acts as my content database. From there:
- Zapier picks up new approved rows
- Formats them for each platform
- Sends them to Buffer/Publer with scheduled times
- I get a daily morning preview of what's going out
The whole pipeline from "AI generates" to "post is scheduled" takes about 5 minutes per post with automation. Multiply by 90 and you're looking at maybe 7-8 hours per month of total work (including review time). Compare that to the 40+ hours it used to take.
What Actually Works (And What Doesn't)
Works Great:
- Educational content: Tips, how-tos, listicles — AI nails these
- Repurposed content: Turning a blog post or video into 5 social posts
- Engagement posts: Questions, polls, "hot take" prompts
- Hashtag research: AI is actually really good at this
Needs Heavy Editing:
- Personal stories: AI can structure them but can't make them authentic
- Humor: AI humor is cringe about 60% of the time
- Hooks: AI hooks tend to be generic — I rewrite most of them
- Trending topic takes: AI is often a step behind real trends
Doesn't Work:
- Controversy or hot takes: Too risky to automate
- Community responses: Engagement requires genuine human interaction
- Platform-specific trends: AI doesn't know what sounds are trending on TikTok right now
The Numbers: My Content Performance
| Metric | Manual Era | AI-Assisted Era |
|---|---|---|
| Posts per month | 30-40 | 90+ |
| Time spent per month | 40+ hours | 8-10 hours |
| Average engagement rate | 2.3% | 3.1% |
| Follower growth rate | 5-8%/month | 12-15%/month |
| Content cost | $2,500+ (time + freelancers) | ~$150 (AI tools) |
The engagement rate increase surprised me. I think it's purely because consistency and volume give the algorithms more data to work with, and more shots on goal means more viral hits.
How to Build This Yourself
You don't need to replicate my exact setup. Here's the minimum viable version:
- Week 1: Write your brand voice doc. This is non-negotiable.
- Week 2: Set up Claude or ChatGPT with your brand voice as a system prompt. Generate one week of content. Review everything manually.
- Week 3: Add a scheduling tool (Buffer free tier works). Start automating the posting part.
- Week 4: Add analytics review. Identify what's working. Update your prompts based on performance data.
From there, you can layer on more automation as you get comfortable. If you want to go the AI agent route, check out my guide on how to setup — it's the same foundational approach.
FAQ
Does posting 3 times a day annoy followers?
On most platforms, no. Social media algorithms don't show every post to every follower. You're competing for feed space. More posts = more chances to appear. The exception is X/Twitter, where timeline followers do see everything — there, 2-3 tweets plus some replies is the sweet spot.
How do you maintain quality at 90 posts per month?
The review step is everything. I'd rather publish 70 good posts than 90 mediocre ones. AI generates the volume, and I'm the quality filter. Some weeks I reject 20% of what AI produces.
What if I only manage one brand?
This system scales down perfectly. For one brand, you're looking at maybe 2-3 hours per week total. The time savings are even more dramatic because you don't need the multi-brand complexity.
Can I use this to start a social media agency?
Yes, and that's exactly what a lot of people are doing. Once you have the system working for yourself, you can replicate it for clients. I cover this more in my article about how to money.
What's the biggest risk?
The biggest risk is publishing something off-brand or factually wrong because you didn't review it carefully enough. That's why I insist on the approval gate. Never remove the human from the loop entirely.
Want to Learn More?
I teach everything I know about AI tools, automation, and building income streams inside my free community. We have 300+ members sharing wins, asking questions, and helping each other build with AI.