Before we start, I want to show you why I think this workflow is worth paying attention to. The organic content system I’m about to break down has helped me get results like these: more than 20M views across just six videos.
One of those posts alone reached 7.5M views and brought in 3.3K new followers. What I’m going to show you now isn’t a theoretical AI setup. It’s the exact kind of process I use to find ideas, turn them into content, create the assets and remove as much repetitive work as possible.
I’m going to build the whole setup from scratch, show you exactly what I do at every step, and give you enough detail to replicate the same workflow yourself.
1. Building my AI marketing team
I’d start by opening Grok Bot and creating the different Bots I’m going to use throughout the experiment.
Rather than giving one Bot a huge prompt and asking it to do everything, I want to create several specialized Bots that gradually learn how I work: Researcher, Creative Director, UGC Creator and Publisher.
Prompt for the coordinator: Master Prompt
I do it this way because I want each Bot to have a very specific job and only the context it actually needs. Giving an LLM more information doesn’t necessarily make it better at a task. Research on long-context models has shown that performance can decrease as context gets longer, even when the model can still retrieve the relevant information correctly.
So instead of teaching one Bot everything about my apps, avatars, content, analytics and publishing strategy, I separate those responsibilities.
Before creating the team, I also create a Coordinator Bot. This is basically the Bot I talk to when I want something done. It knows what every member of the team does and can delegate the task to the right one.
For example, if I ask:
“Find me three viral videos in the quit-drinking niche that I could potentially adapt.”
The Coordinator knows that task belongs to the Researcher.
If I later ask it to turn one of those references into an original concept for my app, that task goes to the Creative Director.
Another reason I like Grok Bot for this is that the work happens on its cloud computer. That means I can give a Bot a task, close my laptop and let it continue working in the background.
For example, I can leave my Researcher analyzing competitor accounts before going to sleep and check the results the next morning.
For me, that’s where agents become genuinely useful: not replacing the creative decisions, but removing repetitive work that normally takes hours.
2. Creating my Researcher
The first member of the team I create is the Researcher, because almost every piece of organic content I make starts by finding something that is already working.
This is also one of the most time-consuming parts of organic marketing.
The difficult part usually isn’t creating the video. It’s finding ideas that are actually worth testing.
So I go back to my Coordinator and ask it to help me create the Researcher step by step.
Once it’s created, I explain what my app does, the audience I want to reach and the niche I’m targeting.
In this case, I’m working with an app in the quit-drinking niche.
I also give the Bot competitor accounts so it has real examples to analyze instead of asking it to invent ideas from scratch.
Then I give it a task like:
“Find content formats that are currently performing well in this niche. I don’t want generic content ideas. Find actual examples, explain why people are stopping to watch them and show me how the underlying concept could be adapted to my app.”
I specifically ask it to look at things like the hook, the first visual, the curiosity gap, the way the story develops and how the product is introduced.
After a few minutes, it gives me several real examples, explains why they may have performed well and highlights the parts I should pay attention to.
This is where I make the decision myself.
The Bot can find opportunities, but I still decide which one I think is worth testing.
I’ve always said that one of the hardest parts of organic marketing for apps is simply finding enough good content. Being able to delegate that research to a Bot working in the background is incredibly valuable.
Once I choose one of the references, I send it to the next member of the team.
3. Creating my Creative Director
Once the Researcher gives me a concept I think is worth testing, I move to the Creative Director.
The first thing I do is copy the brief generated by the Researcher and paste it directly into this Bot.
At this point, I’m not asking for new ideas. I already have the reference and an explanation of why it worked. What I need now is to turn it into something I can actually produce for my app.
So I paste the Researcher brief and ask the Creative Director to create a complete production-ready concept for my quit-drinking app.
The Bot then gives me the full structure: the hook, character, scenario, number of slides, story, what each visual should show and how my app should be introduced.
In this case, it creates a simple two-slide concept.
The first slide is a natural phone-camera style photo with the hook:
“what was the text that made you stop drinking?”
The second slide is the payoff: an iPhone-style lock screen showing a personal message from a family member, with a small notification from my app underneath.
One thing I found interesting is that the Bot specifically tells me not to make the app feel like an ad.
It describes the app as a soft proof element rather than a hard pitch.
So instead of adding a huge logo or asking people to download it, the app simply appears naturally inside the lock screen with something like:
12 days and $160 you haven't drunk. Tonight you don't have to quit forever, you just have to make it morning.
The reader understands what the product does without interrupting the story.
It also gives me a few believability checks, like keeping the visuals natural, avoiding overly polished AI imagery and not stretching a simple concept into too many slides.
Finally, the Creative Director creates a clean handoff containing only the information the next Bot needs to produce the visuals.
Now I have a complete concept ready to send to the next member of the team: the UGC Creator.
4. Creating the AI avatar from scratch
Creating a hyper-realistic AI avatar is actually much easier than most people think, as long as you have the right prompts and tools.
To make this easy to replicate, I’ve put all the prompts I use into a separate document. You can use the same system to create as many different avatars as you want just by changing the initial characteristics.
Prompt pack: Get all the AI avatar prompts here: Avatar Prompt
I start inside Grok Images, select the Quality 2.0 image setting and paste Prompt 0.
That first prompt is already prepared to generate a 23-year-old Mediterranean woman, but you can change the age, ethnicity, hair, style or any other characteristic to create a completely different avatar.
Once I get a result I like, I use that image as the reference for the next generations.
Then I simply run Prompt 1, Prompt 2, Prompt 3... all the way to Prompt 6, always keeping the same reference image.
Each prompt is designed to generate the avatar from a different perspective: close-up, side profile, full body and other useful angles.
By the end, I have enough reference images to use the same avatar consistently across all the content I want to create.
You end up with outputs like these:
Four views from the same AI avatar reference pack.
5. Teaching the UGC Bot how I want my content to look
Once I approve the concept, I send the final brief from the Creative Director directly to the UGC Creator.
The first thing I do is connect Higgsfield, so the Bot can actually generate the assets instead of only giving me prompts. It creates the first versions for both slides, and from there I start correcting everything that doesn’t feel realistic.
For example, I tell it that notification content needs to look like a real recent iPhone, not a fake phone mockup on a random background. I also correct things like notification placement, icon size, wallpaper, spacing and the overall composition.
I also upload the real icon of my app and tell the Bot to use it correctly inside the notification. After a few iterations, I define some permanent rules: use Nano Banana Pro 2K, default to 9:16, keep notifications natural and high on the screen, use personal-looking wallpapers and avoid anything that feels too polished or AI-generated.
Finally, I give it one of my existing AI avatars and tell it to use that identity for the hook image instead of generating a random person.
At this point, the Bot is starting to learn my actual visual style, which means future generations should require much less correction.
6. Turning one concept into multiple pieces of content
Once I have one version that I like, I don’t stop there. If the structure is good, I want to get as much content as possible from the same idea.
So I go back to the Creative Director and ask it to keep the core concept but create several new variations around it.
For example:
“Keep the same underlying idea, but give me five different opening scenarios that create the same type of curiosity.”
The important part is that I’m not asking it to copy the original post. I want it to preserve the mechanism that made the concept interesting while changing the scenario, hook or context.
Once it gives me the variations, I choose the ones I like and send those briefs to the UGC Creator, which generates the extra assets using the same avatar and the visual rules it has already learned.
This is how I can turn one good idea into several pieces of content without having to restart the entire process every time.
7. Creating my Publisher Bot and connecting Postiz
Once the content is finished, I move to the Publisher Bot.
This Bot has a much simpler job: it doesn’t create ideas or change the content. Its job is to take everything I have already approved and organize it for publishing.
Before the Publisher can use Postiz, I connect it directly from inside Grok Bot. I open the Marketplace, search for Postiz, open the plugin and click Add. There is no API key or manual credential setup involved.
After adding it, Postiz appears inside Grok Bot and asks me to authorize the connection. I click Authorize, which opens a Postiz permission screen showing exactly what the connection will be allowed to do, including accessing my connected channels, creating and scheduling posts, and reading post analytics.
I approve the connection inside Postiz, and my browser then asks if I want to reopen Grok Bot. I confirm it and go straight back to the app.
When I return to Grok Bot, Postiz is already connected to my account. That’s the entire setup.
I don’t need to copy an API key, paste credentials anywhere or configure anything manually. The whole connection is handled through the authorization flow.
Once it’s connected, the Publisher can work with the channels I already have inside Postiz.
From there, I can move straight into the actual publishing workflow.
Then I give it the finished assets together with some basic instructions. For example:
“Here are the finished posts for this week. Schedule two per day, leave enough time between each post and prepare everything for my approval before publishing.”
From there, the Publisher prepares everything and places the posts into the calendar so I can review the final schedule before anything goes live.
What I like about this is that I can batch-create the content first and then delegate one of the most repetitive parts of the process: uploading, organizing and scheduling everything one by one.
I still approve the final calendar, but I don’t have to manually prepare every single post myself.
8. What I still do myself
Even with all of this automated, there are still a few things I don’t want AI deciding for me.
The Bots can research, organize ideas, generate assets, create variations and schedule content. But I still decide which ideas are actually worth pursuing, which creatives feel natural enough to post and when a format deserves another iteration.
I also still review the final content before anything goes live. AI can save me hours, but it still doesn’t fully understand context, taste or what feels native to a platform the same way I do after testing hundreds of pieces of content.
That’s basically how I see AI in my workflow: not as something that replaces marketing, but as something that removes the repetitive parts so I can spend more time on the decisions that actually move the needle.
If you want to go deeper into how I grow apps with organic content, AI UGC and the systems I use behind the scenes, I share the full process inside my Spanish Skool community with @AlbertoAIcode:
Join the community here: Skool Community
And if you’re not Spanish-speaking or you want help with your specific app, I also do 1:1 consulting calls where we can go through your content, app, current results and growth strategy together.
Book a 1:1 call with me: Book a call
9. Showing the complete setup at the end
At the end, I like to step back and look at what I’ve actually built.
I now have a Researcher finding opportunities, a Creative Director turning them into production-ready concepts, a UGC Creator generating the assets and a Publisher organizing the finished content.
The important part is that none of these Bots were created just for the sake of having “AI agents.” Each one has a very specific job, and throughout the article I’ve shown exactly how I use them: what I give them, what they return, what I correct and how the work moves from one Bot to the next.
That’s also why I wanted to use one real example from beginning to end instead of explaining everything in theory.
You’ve seen how I find the original idea, turn it into something that fits my app, create the avatar and visual assets, prepare variations and organize the content for publishing.
The system isn’t perfect, and I still make the final decisions, but it removes a huge amount of repetitive work from my organic marketing process.
And once all the creative work is finished, I need one place where everything can actually be organized, scheduled and distributed without going back to doing it manually.
That final layer of the workflow is Postiz
