How Our Game of Thrones Bot Got 4,000 Users in Three Days

Benjamin Duthoid, September 7, 2016

Last month we built our first chatbot using Chatfuel, as our way of testing the platform. Since we considered it an elaborate test, we wanted to build something fun and enjoyable, to attract as many people as possible. As huge fans of Game of Thrones, we made a survival bot that had you go through the 6th season, with a pinch humor. We decided to call it GOTSurvival.

We first showed this chatbot to our friends and family. They liked it but didn’t go crazy over it like we had hoped they would. While reaching out to friends for a round of user testing is nothing uncommon, bear in mind that people close to you can be the worst audience. They know you too well to be amazed by anything you do, and even if they are genuinely impressed, you never know if they’re being honest or just polite.

So, we sent the bot to a French website targeting teenagers, and they made an article about it. Right after publication, many people started chatting with the bot. In fact, there were so many users that the ‘message’ section of our Facebook page froze. For the next 48 hours, we were unable to access it.

During the first three days, more than 4,000 people talked to the bot, generating 500,000+ messages. The shortest path to winning the game required 24 clicks, which led to an average of 35 clicks per user. We also gathered 500+ email feedbacks thanks to a “user input” plugin put at the end of the news article. A total of 97% of them were positive.

Screenshot 2016-09-07 15.36.06

Here is rather unexpected metric: more than 10 percent of our users came back to the bot, despite the fact that we didn’t push any new content or sent any messages.

Let’s sum up. We got traffic, users enjoyed the bot, and some of them even came back.

Considering most chatbots’ problems with user acquisition, our experience was a pleasant surprise.

With hindsight and real-world data, we can now analyze what we got right, by chance or by design.

What We Got Right

We targeted the right audience

We shared our chatbot on a website for teenagers, on which Game of Thrones is always a trending topic. It was only natural that this audience would try our chatbot, as we were targeting their interests.

Many bot builders try to acquire users by posting their bot on Facebook groups (such as “Bots”), and by getting featured on Product Hunt. This is a good way of getting testers, but not users.

You’ll be misled by their use of your product – they’re mostly enthusiasts trying to evaluate your bot, and they’ll never come back. In other words, forget about developers for anything other than testing.

In the real world, 99.9 percent of your users won’t be developers. You should focus on these users instead.

We used their words

We talked to our audience in a way they could understand. We used everyday French slang so the bot would sound youthful and be a good fit for teenage users.

It didn’t matter if you weren’t a Game of Thrones fan. If you didn’t catch the references, you would get the jokes and still enjoy the bot. I love the topic, so I tried to tell the story in a way I would find fun and entertaining myself.

We kept it simple

With chatbots, complexity doesn’t necessarily mean quality. Our Game of Thrones bot told a simple, linear story. If you chose the wrong path, you just had to go back one or two blocks and start again. The “technology” used was dead simple: only text blocks and images.

Nobody cared about the simplicity of our bot. We got a single complaint, and it stated that the bot was too short. Luckily, that’s a valid complaint we can handle.

To be honest, I tried hundreds of bots and ended up hating most of them. They were either useless, too complex, or too spammy. However, I never hated a bot for being too simple.

We gave it life and personality

We used a lot of GIFs and images — 67 to be accurate. On average, this translates into more than 1.5 images/GIFs per block. Integrating GIFs and images works great on Chatfuel, and visual elements fit well within Messenger bots. This gave life to our story.

Screenshot 2016-09-07 15.46.44

To enhance the visuals, we used short text messages. Nobody wants to read long messages on a chatbot. In most blocks, the only message that exceeded one sentence was the last message, explaining the choices the user had.

Those are the things I think we got right. Now let’s talk about the things we messed up and should have done differently.

What We Would do Differently

Make the flow more natural

If we had to do this again, we would improve the conversational experience with a few techniques and features we are now aware of.

First, we would use quick replies to make the story flow more natural. Unfortunately, at the time our bot was designed, Chatfuel didn’t support them, so buttons had to do the job.

Second, we would use emojis. I didn’t like emojis at first and thought GIFs and images were enough. However, emojis work so well within bots that I changed my mind: I now use them in most of my chatbots ?.

Screenshot 2016-09-07 15.48.57

Customize the experience

We didn’t customize anything in our GOT chatbot. Every user had the same experience and had little influence over the outcome. If we had to do this again, we would customize the experience for every user.

First, we would retrieve user variables from Facebook and call the user by their first name from the very first block. The conversation is private, so let’s make it look like we’re friends.

We would also give users control over the story, by letting them choose a character and weapon at the beginning.

Bots offer a good opportunity to have a one-on-one conversation with the user. We should have taken advantage of this by providing each user with a unique experience.

Promote it better

The biggest mistake we made was not focusing on the virality of the bot. We added a “Share on Facebook” button one week after our launch, after most of our traffic was gone. Had we integrated the share button from the beginning, we would have easily doubled our audience. Lesson learned.

Also, we could have distributed it better. We only sent it to one website, when we could have done the same with any site catering to a similar audience. We would have inevitably ended up with a lot more users.

How you promote your bot is just as important as the bot itself. Some great bots out there struggle to get users, and Streak Trivia deserves more than 100 daily users.

To sum up, think about distribution before you start building. If you’re a brand, it’s easy: capitalize on your Facebook page and community. But if you don’t have a community yet, you’ll have to be creative. You may also have to pay to get users.

Better to ask yourself what you can build for a given channel than to look for a channel to promote what you’ve already built.

If you think there’s something we should have done differently, please let us know in the comments section.

We translated our GOTsurvival chatbot to English, so you could give it a try, and we hope the translation doesn’t suck! French speakers can talk to the original bot.

Benjamin is a French entrepreneur. He cofounded Livebotter to help companies build chatbots people use. Tweet him at: @BenjaminDuthoit