THE RESULT
Instead of starting with a long introduction, I want to show you the result first. Then I’ll break down exactly how I went from finding a format that was already working to creating my own version, generating the assets, building the slideshow, organizing the content, and scheduling it.
1. FIND A FORMAT THAT IS ALREADY WORKING
The first thing I do is look for content that is already working. I don’t start by trying to come up with a completely original idea from scratch.
Once I know which niche I want to target, I go directly to TikTok’s search bar. Let’s use fitness as an example. I would search for terms like “fitness apps,” “gym apps,” or “workout apps.”
Then I use TikTok’s filters to sort the results by the most liked posts from the past month. This gives me a much better idea of what is actually working right now.
One of the biggest mistakes I see is people replicating content that went viral a year ago. It may have worked at the time, but there’s a good chance that format is already saturated.
I’m looking for formats that work repeatedly, not just one viral video. A video can go viral because of timing, luck, or a very specific situation. But when I see the same format working multiple times, that’s when I know there is a repeatable pattern worth testing.
Once I find something interesting, I save it and move on to the next step.
In this case, I found a slideshow using the notification format, which is everywhere on TikTok right now and has been extremely viral. The one that caught my attention the most was this one:
This format is definitely worth testing because, as I mentioned earlier, it meets three key requirements.
First, it has gone viral repeatedly. I’m not looking at a format that worked once by chance. I’ve seen this type of slideshow go viral multiple times, which is a strong signal that the format itself is working.
Second, it works across different niches. This is another important factor for me because it shows that the format isn’t dependent on one specific topic or audience. If the same structure can generate strong results across completely different niches, it becomes much more interesting to replicate.
Third, the CTA is introduced on the second slide through the app notification. I particularly like this because the app is introduced naturally within the story rather than making the first slide feel like an advertisement.
For those reasons, this is exactly the type of format I would save and test with my own app.
2. ANALYZING THE VIRAL POST
Finding a viral post is only the beginning. The important part is understanding why it worked.
When I find a slideshow or video I want to replicate, I break it down piece by piece.
I look at the hook, the first image, the avatar or character, the storyline, the number of slides, the pacing, the captions, the CTA, and especially the comments.
The comments are one of the most useful parts of the research. I look at the comments with the most likes because they can tell me what people actually connected with.
I also pay particular attention to the first image and the hook text. The goal isn’t to blindly copy the post. I want to understand the mechanism behind it so I can rebuild the same type of content around my own concept and app y añadirle mi propio estilo.
The music matters too. If the story is emotional or sad, for example, I’ll use music that reinforces that feeling. I like using sounds that are already performing well, but I would never choose a trending sound if it completely contradicts the story.
3. DECONSTRUCTING AND REPLICATING THE POST
Once I understand the format, I adapt it to my own app.
A lot of people think going viral requires coming up with an idea nobody has ever seen before. In my experience, it’s almost the opposite.
You need to learn how to identify what is already working, understand why it works, and create new variations around the same underlying mechanism.
Once I have the content I want to replicate, I take it to a custom GPT specifically designed for prompt engineering with GPT Image 2.5, which is the image generation model I use throughout this process.
I ask it to reverse-engineer the prompt behind each individual image.
The goal is to get a detailed prompt that I can then take directly into GPT Image 2.5 and use to generate my own version of the image.
Once I have the prompts, I go to GPT Image 2.5 and paste the prompt for each image I want to recreate.
I highly recommend also adding the original image from the competitor as a reference. This gives GPT Image 2.5 a visual reference for the image I’m trying to replicate and helps me recreate the same type of visual while making the changes I need.
On top of that, if I want to have a completely personalized character, I can also add my own AI avatar as a reference.
The custom GPT I use for this process is:
The results I’ve been getting from this process are honestly amazing.
Here’s an example:
Here you have a screen recording to prove none of my screenshots are fake:
4. I use GPT-6 Astra as the brain behind the content
Once I have a few examples of content that are already working, I don't just start creating random new videos.
This is probably one of the most important parts of my workflow.
I take the patterns that are already working and use GPT-6 Astra to turn them into new concepts.
The goal isn't to ask ChatGPT:
“Give me 20 viral TikTok ideas.”
That usually gives you 20 generic ideas that sound like they were generated for everyone else.
Instead, I want GPT-6 Astra to understand what is actually working in my niche, identify the underlying patterns, and then create new ideas that follow those patterns without simply copying the original content.
I start by feeding it real examples
For example, let's say I've found 5–10 TikToks that are performing extremely well in the niche I'm targeting.
I give GPT-6 Astra the examples and explain what I'm trying to achieve.
I want it to look at things like:
- The first 1–2 seconds
- The hook
- The type of story being told
- How curiosity is created
- The progression from slide to slide
- The emotional payoff
- The CTA
- The type of images being used
- The pacing
- The amount of text per slide
- What makes someone want to keep swiping
I don't want it to simply tell me what the videos are about.
I want it to figure out why someone would stop scrolling and continue watching.
Then I make GPT-6 Astra reverse-engineer the pattern
This is the prompt structure I use:
“I'm going to give you several examples of TikTok slideshows that performed extremely well. Analyze them as a content strategist. Don't focus only on the topic. Identify the underlying patterns that make them work: hook structure, curiosity loops, pacing, storytelling, emotional triggers, visual progression, payoff and CTA.
After analyzing them, create a repeatable content framework that I can use to generate new concepts without copying the original videos.”
Then I give it the actual examples.
This is where having real examples makes a huge difference.
The model isn't inventing a strategy from scratch.
It's looking at content that has already demonstrated that people are willing to watch it.
5. From one winning format to dozens of new ideas
Once GPT-6 Astra understands the pattern, I can take it one step further.
Instead of asking:
“Give me more ideas.”
I ask:
“Using the patterns you identified, generate 20 new slideshow concepts. Each concept should use the same underlying structure but explore a completely different scenario. Avoid repeating the original topics or wording. Prioritize ideas that create an immediate curiosity gap and can be explained through 6–10 visual slides.”
Now I have a list of concepts that are derived from a proven format, rather than generic AI ideas.
For example, imagine one of the winning videos is built around:
“I thought my boyfriend was cheating on me. Then I found this on his phone.”
I don't want GPT-6 Astra to give me 20 variations of someone cheating.
Instead, I want it to understand the mechanism behind the idea.
Then it can apply that mechanism to completely different stories:
“I thought my best friend was hiding something from me. Then I saw this photo.”
or:
“My boss called me into his office after work. I had no idea why until he showed me this.”
or:
“My girlfriend told me she was going to sleep early. At 2:14 AM, I received this message.”
Same psychological structure.
Completely different content.
That's the difference between copying a viral video and reusing a viral format.
6. I batch-create the slideshows with GPT Image 2.5
Once I have a list of slideshow ideas that fit the viral format I've identified, I repeat the same process for each one.
At this point, I already know what I want each slideshow to be about, how the story is going to progress, and what each slide needs to communicate.
Now I just need to turn those ideas into the actual visual content.
For this, I use GPT Image 2.5 to generate the images for each slideshow.
The key for me here is batching the production instead of creating one slideshow at a time.
If I have 10–15 strong ideas, I'll usually generate all of them in one batch.
That gives me roughly 10–15 slideshows ready to go, which is enough content to cover an entire week.
I repeat the same process for every concept:
For example, if GPT-6 gives me 15 concepts based on the patterns I found earlier, I'll go through those concepts one by one and generate all the visual assets I need.
I also try to keep the visual style consistent across the slides within each slideshow, while adapting the scenes to the story.
One more thing that I highly recommend: use Pinterest images whenever possible, especially images without visible human faces. They tend to work extremely well for this type of content because they feel more organic and native to the format. In fact, around 80% of the images I use for app marketing come from Pinterest. I’ve found that these kinds of images often perform better than overly polished AI-generated visuals because they look more like something a real person would naturally come across and share.
The goal isn't to spend my entire day making one perfect slideshow.
It's to build a repeatable production system where I can go from a list of proven content ideas to a full week's worth of content in one session.
Once I have all 10–15 slideshows finished, I have my entire content library ready.
And that's when I move on to the next step: organizing and scheduling everything with an agent.
7. Let my agent organize everything
Once I have all the content ready, I don't want to manually manage every single post for the rest of the week.
If I've generated 10–15 slideshows, I already have enough content to cover several days. At that point, I can give everything to my ChatGPT agent and let it take care of the organization.
I can provide the agent with the slideshows I've created and the context it needs, and have it decide how to distribute the content throughout the week.
This is especially useful when you're producing content in batches.
Instead of creating something, posting it, and then having to come back the next day to repeat the entire process, I prefer to sit down once, create a large amount of content, and then let the system handle the repetitive part.
And this is where I connect the agent to Postiz.
Postiz allows me to manage and schedule the content across my social accounts, so the agent can essentially prepare everything and then push the content into my publishing workflow.
The important part here is that I'm not asking AI to magically create a successful content strategy for me.
I've already done the difficult part.
I've researched the formats that are working, analyzed the patterns behind them, created my own variations, and produced the actual content.
The agent is there to take all of that work and make it easier to manage.
For example, once I've finished creating my 10–15 slideshows for the week, I can give them to the agent and have it organize when each one should go out. From there, I can use Postiz to schedule the posts rather than having to manually upload and schedule every piece of content myself.
That might sound like a small improvement, but when you're publishing consistently, these little repetitive tasks add up very quickly.
And that's ultimately what I'm trying to eliminate.
I don't want AI to replace the creative part of the process. I want it to remove as much of the boring, repetitive work as possible.
Once the system is set up properly, I can spend a few hours creating the week's content and then leave the distribution side organized in advance.
That gives me something much more valuable than simply generating more AI images:
a repeatable content system that I can run every single week.
If you're already creating AI UGC but you're still uploading and scheduling every post manually, this is probably one of the first things I'd automate.
The content creation is only half of the problem.
You also need a system that makes it easy to consistently get that content in front of people.
And that's exactly where connecting your ChatGPT agent with Postizbecomes useful.
8. DOUBLE DOWN ON WHAT WORKS
When a format starts working, I don’t immediately move on to something completely different.
I create variations around the winning structure.
Same hook, same story. Same character, different details. Same structure, different angle. Same emotion, different context.
That’s the key to going viral again and again without relying on “luck.”
This is also one of the main reasons I’m able to generate millions of views over and over again. I’m not constantly trying to reinvent everything from scratch. When I find something that works, I keep pushing it and testing different variations of it.
The biggest fear people usually have is that their content might become repetitive or “not creative enough.”
But if you’ve been doing this for a while and actually understand how this works, you know that it’s exactly what you should be doing.
For me, any format that gets more than 10–15k views is worth replicating with small variations.
If a format stays below 3k views, I kill it immediately and don’t try it again.
Here I’m leaving two examples of almost identical images that both went viral, from different slideshows, so you can see that this works over and over again.
The winners become templates. The losers become data.
**
9. THE COMPLETE SYSTEM / FROM ONE VIRAL FORMAT TO A CONTENT MACHINE
The biggest takeaway from this entire process is that I don't rely on a single viral video.
I rely on a system.
I find formats that are already working, break them down, adapt them to my own app, use AI to produce the content at scale, batch everything together, and then let my agent and Postiz handle the organization and distribution.
Then I look at the results and use what I learn to create the next batch.
And the process starts again.
That's what allows me to generate millions of views repeatedly instead of treating every viral post as a one-off.
You don't need to reinvent content every day. You need to find what works, understand why it works, and build a system that allows you to reproduce it consistently.
Once you have that system, the amount of content you can test becomes the real advantage.
Want to go beyond this article?
I share the full process, experiments, prompts and strategies I’m using to grow apps organically inside my Skool community, which I run in Spanish together with @AlbertoAIcode
→ Join the community: Spanish Skool Community
If you’re not Spanish-speaking but want to work with me directly, I also offer 1:1 consulting for founders looking for personalized guidance.
→ 1:1 consulting: Book a 1:1 Consulting Call
If you're already creating a lot of content, the next bottleneck isn't always content creation. It's managing and distributing everything you produce.
Once I have a full batch ready, I can connect it to my AI workflow and let the agent organize everything for me, from deciding how to distribute the content throughout the week to preparing the posts for publishing. This means I don't have to manually manage every single post or spend time uploading and scheduling everything one by one. I can create the content in batches, have my AI handle the organization, and then use Postiz to schedule everything.
