To use AI to go viral on social media, don't ask it to write the script. Ask it to research the idea. 99% of what makes a video go viral is the idea, not the editing or the design. Use GPT-6 Astra or Fable to find timely and timeless ideas that move millions of people, then add a strong hook.
This post expands a 47-second video that goes live on September 23, 2026. The embed will be added here once it's published.
Key takeaways
- The idea is 99% of it. Editing, pacing, delivery and design decide how well a good idea travels. They don't rescue a boring one.
- AI is a research tool first, a writer second. Its best use is scanning for ideas that are trending right now, or timeless and under-told, before a single line is written.
- Two kinds of ideas go viral. Very timely ones (news, launches, drama) and very timeless ones (principles, famous people, frameworks). Both need to speak to millions of people.
- Emotion is the filter. Research on what spreads online found content that evokes high-arousal emotions like awe, anger or anxiety is more viral than content that evokes low-arousal ones like sadness.
- Then the hook. A relevant idea plus a strong hook or headline is the whole formula. That's it.
Contrary to what most people think, the secret to using AI for viral content isn't asking it to "write me a viral video script." I've watched people do that, and the scripts all sound the same: competent, forgettable, and dead on arrival. The key is to use the AI, whether that's GPT-6 Astra, Fable or whatever model you have, to help you research ideas that will go viral.
How do you use AI to go viral on social media?
You use it to find the idea, not to write the video. Give it the job of researching what people are already paying attention to, and what timeless concepts get a reaction every time, then you pick the winner and write the hook.
The reason is simple. A model asked to "write something viral" averages over everything it has read, and the average of the internet is the definition of not viral. A model asked to "find me 10 stories from this week that people in my niche are arguing about, with sources" is doing something it's genuinely great at: reading a lot, fast, and sorting.
So the order of operations is research first, then selection, then a hook, then the script. Most people start at step four.
Why is the idea 99% of what makes a video go viral?
Because a relevant idea gets watched, shared and argued about no matter how it's packaged, and an irrelevant idea gets scrolled past no matter how well it's cut.
It's not the editing. It's not the pacing. It's not how you talk. It's not the colors or the design. If the idea is relevant enough, the video will pop. I say this as someone who runs a company that edits and packages content for a living. Editing decides how far a good idea travels. It very rarely makes anyone care about a bad one.
One example from our own work: a thread about David Ogilvy's advertising principles that we wrote for a client's account didn't do much on the day it went out. A few days later it caught fire and reached almost 250,000 views. Nothing about the packaging changed between day one and day four. The idea was timeless, it spoke to a huge audience, and it eventually found them.
What kinds of ideas go viral?
Two kinds: very timely ideas and very timeless ideas. Both need to be interesting to millions of people, and ideally able to generate an emotional response in them.
| Type | What it looks like | Shelf life |
|---|---|---|
| Timely | A news event, a product launch, a public falling-out, a study that just dropped, something a famous person said this week | Days to weeks. We mark timely angles as stale after 60 days. |
| Timeless | A principle from a famous operator, a counter-intuitive framework, a story everyone half-knows told properly, a universal human problem | Years. Can be reposted and reframed repeatedly. |
The emotional part isn't decoration. Jonah Berger and Katherine Milkman studied every New York Times article over a three-month period to see what got shared, and found that "content that evokes high-arousal positive (awe) or negative (anger or anxiety) emotions is more viral. Content that evokes low-arousal, or deactivating, emotions (e.g., sadness) is less viral." When you're sorting AI research, that's the question to ask of each idea: does this make people feel something with energy in it?
How do you actually prompt the AI for viral ideas?
Give it your niche, your audience, and a specific hunting brief, and ask for sources. Here's the process we run for clients, which lives in a reusable instruction file so any agent can execute it:
- Feed it the themes. The client's core topics, their expertise, and who they're talking to. Without this, you get generic trends.
- Ask for 10 timely angles. Things from the last few weeks that intersect with those themes. Each one needs a source link and a date, so you can verify it and know when it goes stale.
- Ask for 10 timeless angles. Famous people, frameworks, studies and stories that connect to the themes and that a broad audience would recognize or find fascinating.
- Check for duplicates. Keep a running file of used angles. The model sees the list and skips what's already been covered.
- Score for reach and emotion. For each idea: how many people care, and what do they feel? Kill anything that's niche and flat.
- Only then write. Take the top idea and write the hook first. If you can't write a hook that makes someone stop, the idea isn't as strong as it looked.
One structure that works well with external ideas: make 95% of the video about the interesting external thing (the news, the person, the study), and use the last 5% to bridge to what you do. The idea carries the viewer. Your pitch rides along almost hidden inside it.
Where does the hook fit?
After the idea, and it's the only other thing that matters. A relevant idea combined with a strong hook or a strong headline is the whole secret to going viral on social media.
The hook's job is to prove, in the first line, that the idea is as interesting as it is. It doesn't create interest from nothing. That's why writing hooks for a weak idea is so painful: you're trying to promise something the video can't pay off. Pick a better idea and the hook usually writes itself.
Who this is for (and who it isn't for)
This is for creators, founders and marketers who post regularly and have been using AI mostly as a writing assistant. If your output is consistent but nothing pops, the problem is almost certainly upstream of the writing, and this is the fix.
It isn't for anyone who wants to go viral for its own sake. Viral reach on a topic unrelated to your business gets you followers who will never buy anything. The research step should always start from your themes, which is why our process loads the client's profile before it goes looking.
Common mistakes when using AI for viral content
- Asking for a viral script instead of viral ideas. You get the average of the internet, which is the opposite of viral.
- Polishing a weak idea. Better editing, better lighting, a new intro. None of it moves a video whose idea nobody cares about.
- Only chasing timely ideas. They burn out in weeks. Timeless angles are the ones you can reuse for years.
- Skipping the source check. Models will confidently hand you a "trend" that doesn't exist. Ask for links and dates, and open them.
- Ignoring emotion. "Interesting" isn't enough. If the idea doesn't produce awe, anger, anxiety, delight or surprise, it will inform a few people and travel nowhere.
- Forgetting the hook. A great idea still needs its first line to stop the scroll.
That's the whole secret. Use the AI to find an idea that's relevant enough, timely or timeless, that millions of people can feel something about. Then give it a strong hook. That's it.
Frequently asked questions
Can AI write a viral video script?
It can write a competent script, but the script is rarely why a video goes viral. The idea is. Use AI to research and rank ideas first, then have it help with the script once the idea and the hook are set.
Which AI is best for researching viral ideas?
Any capable model with web access works: GPT-6 Astra, Fable, or whatever you already use. The prompt and the process matter more than the model. Ask for sources, dates and a list, not a single answer.
What's the difference between a timely and a timeless idea?
A timely idea is tied to something happening now, like a news event or a launch, and goes stale within weeks. A timeless idea is a principle, story or framework that stays interesting for years and can be reframed many times.
Does editing matter at all for going viral?
Yes, as a multiplier. Good editing helps a strong idea reach more people and hold them longer. It doesn't rescue an idea nobody cares about.