Final Major Project: PollyBot.

The Brief

To develop a way to make freethinking 16-30 social media users aware of the curation of content occurring around them and encourage or provide them with a way to view opposing viewpoints to broaden their thinking and pop their digital filter bubble.
View the brief, here.


Every project in the past which has attempted to make a user aware or pop their digital filter bubble has been done so via a stand-alone app, website or desktop browser plugin. As such, there have been no attempts at popping one’s filter bubble in a way that is integrated into a social platform and readily available for younger users.

With the likelihood of a user downloading a standalone app or desktop browser plugin decreasing, I believed the best way to attract the attention of social media users was to approach them directly on social media in a way that felt organic and authentic. And since 42% of social media users said they use it to keep in touch and communicate with others, approaching the target market through the art of conversations seemed only fitting for this project.

Initial ideas included the creation of a chatbot which sent you news on both sides of the story or one of which scanned your engagement history and recommended you articles outside your comfort zone, however, these both felt a little stale and too unengaging for the younger audience, not to mention un-innovative, simply repackaging a website in the form of a chatbot.

I then stumbled across the article “Future chatbots will be able to argue with you enough to help change your mind”, and I started thinking about how I could create a meaningful interaction between bot and human. And the answer was simple: Create a bot in which the user can argue with, and it always gives the opposite opinion, and from there, the concept was born.

The “Halt! Bot or Not” was a useful introduction into the world of bots, and how they can be created in such a way that promotes a natural, organically flowing conversation. It highlighted that the small things, such as fallbacks when the bot doesn’t understand, can mean a lot when creating a bot.

Bot Development

Creating the bot was extremely daunting at first. Equipped with very little technical knowledge, I scanned Google and tested out a variety of bot-making platforms. Using machine learning was an obvious solution to create a conversational bot, however, this is a development & learning process which takes place over months, which was not possible with the FMP deadline. Instead, I found, one with the best interface for someone with little coding knowledge immediately stuck out, and I started testing the platform to see how easy it would be to create a conversational bot.

Screen Shot 2018-05-09 at 14.36.31.png

The platform turned out to be a huge mess, and with so many different entities and keywords needed to make the bot fully conversational, the platform became extremely laggy and almost unusable. Like most other bot-building tools, it was good for creating a linear-style bot, but not a conversational one.

From here I moved back to one of the first bot-making platforms, Chatfuel. Upon first glance it didn’t appear to be very user-friendly for someone without technical knowledge, so I joined the Chatfuel community page to see how I could create my conversational bot.
Screen Shot 2018-05-09 at 14.32.50.png

I was informed general-bot builder DialogFlow would be better: a platform which focuses on NPL (natural processing language) to create agents which can be exported into Facebook Messenger chatbots – a more complicated approach, but the only one I found to create my conversational bot.

I joined DialogFlow community page and used their expert knowledge, along with a variety of YouTube and web tutorials to develop my bot. I contacted a YouTuber who specialised in DialogFlow tutorials, who was also able to help me 1-on-1 with problems I was having.
Screen Shot 2018-05-09 at 14.34.23

Throughout development, I was continually testing & improving the bot. When available, I also got users to test the bot. Unfortunately, test users identified that hierarchy of the conversation meant that the user was led down a general path and sometimes received comments that didn’t match their intents. Based upon this feedback, I had to rework the introduction to make it more stable to ensure the conversation flowed smoothly.

With the deadline quickly approaching, I had to limit the argument topics of the bot to Donald Trump, Brexit, Conservatives & Labour Party, to ensure I felt enough time to submit the bot to Facebook for approval to make it available to the public.


However, due to the Cambridge Analytica scandal, Facebook created an embargo on reviewing all bots for over a month, meaning the public testing and public use of the bot was extremely delayed.

Screen Shot 2018-05-10 at 11.32.49

Outcome: PollyBot

PollyBot Beta

Not only did I have to create a working chatbot, but also create an identity around it. With arguing with people being such a bold move, I wanted the personality of the chatbot to match this. The name itself also Polly derived from the main issue at hand – political polarisation due to the filter bubble.

Polly’s tone of voice was built to be bold, authoritative and sassy. The sort of empowered women who knew what she was talking about and was not afraid to be sarcastic to get her point across. From this, I implemented slang, emojis, and personal phrases into her vocabulary. Phrases such as ‘boo’ and ‘poppin’ were constantly used to provide an informal narrative which linked back to the issue at hand.


The font & illustrative style was also very bold in nature. Inspired by BlinkTheBee, Polly was created with bold, black strokes and block colours, twinned with a heavy yet curved font in mind, all aimed to create an authoritative yet playful tone of voice for the brand and Polly as a whole. Background iconography was also used to give context to Polly’s mood, whether that be angry or confused.


Case Study Video

A short case study video was created to give users an explanation of the project. Labour was used as an example throughout to increase relevance, as these would be the main supported party from the target audience.

PollyBot Website

Acting as a home to all things Polly, the website gives a description of the project and the problem at hand, as well as housing the case study video. The website has also been used as a way of promoting the project, with it being a handy tool to send to editorial outlets, combined with the email.


Twitter conversation with the author of the original article “Future chatbots will be able to argue with you enough to help change your mind”

Social Media Presence

To supplement the chatbot, I also created and populated Polly’s Facebook & Instagram pages. A variety of social posts were posted to update viewers on the status of the project, as well as create hype.

As an experiment, I also ran 3 social ads on Facebook and Instagram. The first, used to give an overall description of the project, was displayed on Facebook, however, surprisingly received little response. I then turned my attention to advertising on Instagram, where I received a much higher level of engagement on both the video and static posts.


In reflection, I feel that I challenged myself well. I experimented with different mediums I had previously very little knowledge in, and have now widened my breadth of technical knowledge.

Whilst I encountered many different problems, ranging from developing the bot to getting it approved, I feel I overcame these issues well by allocating myself enough time for each task, providing myself contingency days to sort out any issues.

I feel that by reaching out to community pages, forums and YouTube mentors, I was able to build my chatbot successfully. Whilst reading tutorials are somewhat helpful, having bespoke help for issues definitely helped me overcome many technical issues I was facing.

I also feel as if I managed my time well. By mapping out which tasks should be completed by their retrospective dates, I was able to complete the project before the deadline. Whilst the embargo set by Facebook hindered the project by limiting the amount of public testing I could carry out, I feel as if this could not have been predicted.

If I was to carry out the project again, I would have liked to learn and implement deep machine learning, to make Polly more stable, independent & conversational, however ,this would only have been possible if we had more time allocated to the project.

Overall I am very pleased with what I have developed. Starting with no technical knowledge, I was able to build a fully functioning chatbot and develop its’ iidentityand market it in such a way which attracted the attention of consumers and news outlets alike.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s