Overview
What Is In This Guide
The 3 things I was having AI do for me that kept me in 300 view jail
The exact 2-tool stack that changed it (Blotato + Sandcastles), setup included
All 4 copy-paste prompts, given away in full, nothing held back
The 7 free Claude skills that turn the whole loop into single commands
The honest part: what actually took time, and what did not
Before Anything Else
30% off Blotato, on me
Blotato partnered with me on this guide. Use the code below when you subscribe and get 30% off, valid for 3 months from the day you apply it.
ZACH30 Blotato requires a paid subscription to connect via MCP. This code is the cheapest way in.
Section 1
Why I Was Stuck for 2.5 Months
I want to be clear about something, because it is the part most people get wrong when they hear this story. I was not stuck because I was not using AI. I was using it every single day.
I was stuck because I was pointing it at the wrong three things.
I had AI write my scripts
It wrote things that were correct and completely forgettable. AI is great at structure and terrible at taste. The script is the one part that has to sound like a person.
I had AI guess what was trending
It was guessing from training data that was months old. It had no idea what was actually working in my niche that week, so I was chasing topics that already peaked.
I had AI do the creative work
This is the big one. I gave away the part I should have owned and kept the part I should have automated. Flop after flop after flop for 2.5 months.
The lesson
I had it exactly backwards. I automated the creative and did the grunt work by hand. Flipping those two things is the entire difference between the first 2.5 months and the next 2.5 months.
Section 2
The Flip
Here is the rule I run everything on now. AI takes the two ends. I keep the middle.
AI does the research
What already worked for my audience, and what is heating up in my niche right now. Real data, this week, not a guess.
AI does the distribution
Repurposing one video into the right format for every platform, then scheduling all of it. The boring part that used to eat my afternoons.
I do the creative
The angle, the story, the delivery. The part that is actually me. AI hands me the data and takes the finished thing away. I stay in the middle.
Two tools do the two ends. Sandcastles and my own analytics handle the research. Blotato handles the distribution. That is the whole stack. Everything below is how to set both up and the exact prompts that run between them.
Blotato
Claude MCPBlotato connects Claude to every social account you run. This is the tool Sabrina Ramonov used to go from 0 to 3 million followers across platforms in under a year.
Three things I use it for every week: repurposing one video into the right format, captions, and cuts for every platform, so I stopped manually redoing everything for Instagram, TikTok, and YouTube. Auto-scheduling and posting everything at once, so I stay consistent without sitting in front of a screen for hours. And reading and replying to comments plus running automated DM flows, which is how people get from a comment to my newsletter without me touching it.
Set it up
- 1
Go to Customize, then Connectors, then Add custom connector.
- 2
Name it Blotato.
- 3
Paste this URL: https://mcp.blotato.com/mcp
- 4
Press Connect and approve access. You need to be logged into Blotato in the same browser for this to work.
Prefer the terminal? In Claude Code, run:
Verify it worked
Ask Claude: "What social media accounts do I have connected?" If it lists your platforms, you are live. If not, make sure you are logged into Blotato in the same browser, then restart whichever Claude app you are using. Full troubleshooting: help.blotato.com/api/mcp/setup
Sandcastles
Claude MCPThis is the one that fixed my "AI is guessing at trends" problem. Sandcastles shows Claude what topics and formats are working in my niche right now, this week, with real outlier posts instead of a guess from old training data.
I run it before I sit down to write anything, so I am never starting from a blank page. Every outlier it surfaces becomes a starting point instead of a scroll session.
Set it up
- 1
In Sandcastles: Settings, then Connectors, then the MCP section, then Download Plugin.
- 2
In the Claude Desktop app: switch to Cowork, then Customize, then the "+" under Personal Plugins, then Upload Plugin, and drag the Sandcastles zip in.
- 3
Press Connectors under the new Sandcastles plugin, then Install (the URL https://mcp.sandcastles.ai/ comes pre-filled), then Add.
- 4
Switch to Chat, then Customize, then Connectors, then Sandcastles (Custom), then Connect, and log in to your Sandcastles account in the browser tab.
Full official walkthrough with video: help.sandcastles.ai/mcp
Test it
"Give me a list of the top 10 performing videos from my watchlist in the last 7 days. Tell me the common topics and formats that are outperforming, and suggest my next 3 videos based on what is working."
The Exact Prompts
4 prompts, one chat, start to finish
Both tools connected? Open one chat and run these in order. Do not skip prompt 1. It is the one that makes the other three work, and it is the one everybody wants to skip.
Every prompt is here in full. Nothing held back, nothing blurred out.
Find out what is already working
This is the one people skip, and it is the one that changed everything. Do not write anything new until Claude has told you what your own audience already responded to. It pulls your real numbers from Blotato, finds your outliers against your own median (not against big accounts), then uses Sandcastles to see what is heating up in your niche this week. It saves everything to a file so you never run it cold again.
Copy-paste prompt
You are my content growth analyst. I want you to find out what is ALREADY working, before we write a single new thing. Use my connected Blotato account to pull my own posts, and my connected Sandcastles account to pull what is working in my niche right now. MY NICHE: [describe it in one sentence, e.g. "teaching non-technical people how to use AI"] MY PLATFORMS: [e.g. Instagram, TikTok, YouTube Shorts] Do this in four passes and show me each one before moving to the next. PASS 1 — MY OWN DATA Pull every post I have published in the last 90 days with its view count, likes, comments, shares, saves, and follower gain if available. Calculate my median view count. Then flag every post that did 3x my median or better. Those are my outliers, not my "best posts" by vibes. For each outlier, write down: - the exact first line / hook, word for word - the format (talking head, screen recording, listicle, story, tutorial) - the topic - the length - the CTA - the outlier multiple (views divided by my median) PASS 2 — MY OWN PATTERNS Across those outliers only, tell me: - which 2 or 3 hook structures repeat - which topics repeat - which format repeats - what my outliers have that my median posts do not Be specific and quote real lines from my posts. Do not give me generic advice like "be authentic" or "use strong hooks." If you cannot find a pattern, say so instead of inventing one. PASS 3 — THE NICHE Now use Sandcastles. Pull the top performing posts in my niche from the last 14 days. Find the ones that are outliers relative to the account that posted them, not just big accounts getting big numbers. Tell me which topics and formats are heating up right now that I have NOT covered yet. PASS 4 — THE OVERLAP Give me a short list of topics that sit in the overlap between "what already works for MY audience" and "what is heating up in the niche right now." Rank them by how confident you are, and say why for each one. Finally, save all of this to a file called content-audit.md so we can reference it in every future session instead of re-running this from scratch.
Remix the winners into new posts
Now it writes, but only from the patterns it just proved. Every concept has to trace back to a real outlier, and Claude has to say what it changed. That constraint is the whole trick. It stops the generic-AI-slop problem before it starts.
Copy-paste prompt
Read content-audit.md. Using ONLY the patterns you found in there, write me 10 brand new post concepts. These are remixes of what already worked for me, not random new ideas. Rules: - Every concept must map back to a specific outlier from my audit. Name which one it came from and what you changed. - Do not repeat a topic I have already covered in the last 90 days unless you are deliberately re-cutting it with a new hook and a new format. If you do that, say so. - Mix the sources: about 6 from my own proven patterns, about 4 from the niche topics that are heating up but I have not touched. For each of the 10, give me: 1. The working title 2. The hook, written as the actual first spoken line, not a description of a hook 3. The format (talking head, screen recording, listicle, story, tutorial) 4. A 3 to 6 bullet outline of the body 5. The single CTA at the end 6. Which outlier it is remixed from, and the one variable you changed (topic, format, hook structure, or angle) Write the hooks the way people actually talk. Short sentences. No corporate voice, no "in today's video," no "let's dive in." Then rank all 10 from most likely to outperform to least, and tell me which 3 you would shoot first if I only had time for three this week. Save the output to content-queue.md.
Sharpen every hook until it carries the video
Five hook variants per concept, each built on a different structure, each graded, anything under a 7 gets killed. If you have the viral-hooks skill installed it pulls from a 100-framework library. If you do not, the prompt runs the same process manually so you are not blocked.
Copy-paste prompt
Read content-queue.md. For each of the top concepts, I want the hook sharpened until it can carry the whole video. If you have the viral-hooks skill installed, run /viral-hooks on each one and pull from the 100-framework library. If you do not, use this process manually. For every concept, give me 5 hook variants, each built on a different structure: 1. NEWS / FOMO — lead with the most shocking number, dollar amount, or result 2. CONTRARIAN — state the thing everyone believes, then kill it 3. PERSONAL PROOF — the specific result I got, with the real number and the real timeframe 4. OPEN LOOP — name a thing the viewer does not know and promise it by the end 5. DIRECT CALLOUT — name the exact person this is for in the first four words For each variant also give me the on-screen text version, which must be 8 words or fewer and must not just repeat the spoken line word for word. Then grade all 5 out of 10 on: - Does it stop the scroll in the first 1.5 seconds - Is it specific (a real number beats a vague claim every time) - Does it create a reason to keep watching past second 3 - Would a complete beginner understand every word Kill any variant that scores under 7. Tell me your pick and why, in one sentence. Rules: no questions as hooks unless the question is genuinely shocking. No "have you ever wondered." No word a third grader would not know. Update content-queue.md with the winning hook for each concept.
Spread it everywhere and schedule it
The same idea, reshaped for every platform instead of the same caption pasted five times. Then Blotato schedules the whole week. It shows you the plan and waits for your approval before anything goes live, and it sets the reminder to re-run prompt 1 in seven days on fresh data.
Copy-paste prompt
Read content-queue.md. Take the concepts that are already shot and ready. STEP 1 — PLATFORM NATIVE VERSIONS For each one, write the platform-native version for every platform I am on. Do NOT paste the same caption everywhere. Each one gets: - INSTAGRAM: caption that opens with the comment-keyword CTA, then the value body, then the closing CTA, then hashtags - TIKTOK: shorter caption, on-screen-text-first thinking, trending-audio note if relevant - YOUTUBE SHORTS: a real searchable title (this is the one platform where SEO actually matters), plus a description - LINKEDIN: rewrite it for a professional audience, no hashtag spam, lead with the result - X: the thread version, first post carries the whole hook Same idea, different shape for each. If a concept genuinely does not fit a platform, say so and skip it instead of forcing it. STEP 2 — SCHEDULE IT Use Blotato to schedule the finished posts to my connected accounts. Before you schedule anything, show me the full posting plan as a table: platform, date, time, and the first line of the caption. Wait for me to approve it. Space the posts out so I am publishing consistently across the week rather than dumping everything on one day. Use my own analytics from content-audit.md to pick the times, not generic best-time-to-post advice. STEP 3 — CLOSE THE LOOP Set a reminder for 7 days out to re-run the audit prompt so the next batch is built on fresh data, including how these ones performed. That is the whole loop. Research, remix, sharpen, publish, measure, repeat.
Level It Up
7 free Claude skills built for this
The 4 prompts get you the whole loop by hand. These 7 skills turn the same loop into single commands, so Claude remembers your brand voice, grades every post before it ships, and hands the finished thing to Blotato to schedule.
Install them from blotato.com/skills. Run brand-brief first, everything else reads from it.
/brand-brief
Captures your business info (what you sell, who buys, your voice) once, so every other skill uses it automatically. Run this one first.
/content-coach
The beginner’s front door. Say "help me post something" and it walks you through the whole process end to end, no other skill names required.
/post-writer
Writes a complete post (hook, body, CTA) for any platform from just a topic. Sized right for wherever it is going.
/post-grader
Scores any post’s virality out of 10 and gives you the top 3 fixes, ranked by impact. Harsh but fair.
/viral-hooks
A library of 100 proven hook frameworks. Picks the right one for your topic and platform instead of you staring at a blank first line.
/repurpose
Turns one long piece of content into a week of platform-native posts, not the same paragraph copy-pasted five times.
/post-scheduler
Schedules the finished post straight to your connected accounts through Blotato. The last step, once everything above is graded and approved.
The Honest Part
What This Does Not Do
I hit my first 100K view video in the first week of using both tools. That is true, and it is also not the normal case. What is normal is that the research stops you from wasting a shoot on a topic nobody wants, which compounds fast when you are posting daily.
The part that still takes real time is filming and delivery. No tool fixes that. If your delivery is flat, better research just means you flop on better topics.
And you still have to post consistently. The stack removes the excuse of not knowing what to make and not having time to distribute it. It does not remove the work.
Research the outlier, remix it, sharpen the hook, publish everywhere, measure, repeat. That is the whole loop. Run it weekly.
Try It Yourself
Try Blotato free for 7 days
Everything in this guide runs on the free trial, so you can connect it, run all 4 prompts, and see it work before you pay anything. When you are ready to subscribe, use the code below.
ZACH30 30% off, valid 3 months