How I prompted AI app that went viral and roasted 7M people

Hey friends,

So for the past couple of months I’ve been working at Wordware. One of our projects was twitter.wordware.ai that went extremely viral. Over 7M users and $100k in 3 weeks.

And here’s how we did it:

Luck. 

But it wasn’t dumb luck. It was fortune that favors the bold. The luck created by stirring the pot and acting.

How it went down (try and copy):

As a growth team our job was pretty straightforward: we were (are) experimenting around improving underperforming areas. 

At that time it was user acquisition. We were doing unscalable things to figure out the best growth engine. Two of them had a crucial role in shaping our twitter personality project:

First tactic: attending industry events

One of the user acquisition tactics was attending AI industry events. Luckily, around that time there was the AI Engineer World's Fair happening in SF, which was organized by swyx, one of our angel investors. We were able to get permission to do guerrilla marketing of our services.

So we did.

Beside our t-shirts with catchy slogans (which found real PMF there), one of the tactics was setting up a mirror with our logo and QR code that led to an app built on Wordware. The app was called “check your rizz” where you upload your photo and AI comments your look.

To make it more fun, I prompted that app specifically to roast each person’s outfit. 

It turned out that people were having real fun while uploading it and reading their roast. They even went around the venue telling people to take a picture and get roasted themselves.

I guess the appeal was that the machine can see your photo and write very savage comments. There was no emotional connection to it, pure fun.

Second tactic: building apps

Outside of events, one of the other tactics was building cool apps that were to show people what is even possible to do with Wordware. 

First one was audioscribe.wordware.ai which was an open source text-to-note app. 

Second one, launched 3 weeks later, was twitter.wordware.ai.

Thinking process behind that was simple: how we can build something that will be fun and worth sharing. Kamil for a while had an idea for creating a personality test from twitter data, so we started working on it.

We created a scope of personality assessment that AI could generate. But still, we needed to get something to drive it through the finish line.

Couple of days before the launch, after testing AI responses, uploading our tweets and thinking about how to improve it, we added the crucial part: roast.

Initially, our roast was at the bottom of the page. Couple of hours after launch, users said that’s the most fun part of the app, so we moved it to the top. 

We made it easy for people to share each section of their personality assessment. Each time they shared it tagged us on twitter and the link they shared redirected to the personality assessment. 

Now, here’s the important part for going viral: You need to implement virality into your app. Make the most powerful feature show first. Enable easy shareability. And make it beautiful enough that no one is embarrassed to share it.

That’s when things started to really take off. 

Our app covered most of the English-speaking countries pretty quickly, and our Vercel bill was on fire.

But that didn’t matter. We needed to fuel this rocket further. 

To get more coverage, I improved the prompt to create the roast and personality assessment in a language in which the user tweets. 

We weren’t even expecting it, but it got bigger than we thought. Especially in Japan. Antropic bill was growing, API Dojo wanted to ban us because we used their service too extensively. But we didn’t bat an eye and did everything to keep the app running.

And for most part we succeed. We were even asked if a Japanese TV show could use our tool in their program.

Looking back

Did we know for sure that it’ll work? No.

Did we make conscious decisions based on previously gathered data? Not really (however maybe intuition guided us?).

So how is it that now it looks like it was a well thought out process? 

Well, we can only connect the dots looking backwards. It seems like we trusted the dots and they connected (thanks, Steve).

And that’s it. I’m glad that I found a moment to write down those thoughts. Keen to learn what other well working distribution tactics you tried for consumer & business products :) 

Lastly, here’s photo of me with the mirror that might’ve started it all.