Busting election filter bubbles with a Facebook chatbot


On May 22, 17 days before the election, we launched our Red Box Filter Bubble-busting chatbot. This wasn’t a lot of time to amass users, but we pressed on. We started by thinking about how we could do something unique with a chatbot — something that plays to the strengths of the platform which we identified as: easily automating sending messages, discretely collecting data from the users, providing a personalised experience in a simple way and publishing our content in the places that people are sharing articles and talking about the election privately.

We settled on an idea of a filter bubble-busting election chatbot on the Red Box Facebook page. This is a smaller page, with 4,000 fans, but one that would have had a lot of activity on it during the election, so a good testing ground. Initially, we thought of sending our users updates at 5pm with the best of our election coverage — which is a similar idea to many other bots that we’ve seen other publishers make. I wasn’t hugely keen on many of the chatbots I’ve seen other publishers produce. Many seem to be little more than glorified search bars or stunts that have little value after the first use.

However, we quickly realised that this is a good opportunity to collect some information from our users and to send them something more bespoke. By asking who they were thinking of voting for, we could group our users by their voting intention, and send these different groups of users different sets of articles.

Running the bot

Using a Google Sheet, this was easily managed by the social media team, who filled in the link data and tagged each link with the name of the party that the article was about.

Users who said they intended to vote Labour, for example, could be sent every article except those tagged ‘LAB’ to make sure they received a broad range of content that they otherwise might not see, etc. In some cases, we felt a particular article was important enough to be sent to everyone so we tagged it ‘ALL’. A Labour voter might then receive a handful of articles tagged ‘CON’, ‘LD’, ‘ALL’ etc.

We could also collect information about the constituency that the user was in by asking them for their UK postcode. With this, we could automatically send a graphic of their constituencies 2015 general election results.

What we learnt

By June 8, we had 400 unique users. This is a pretty good effort considering the short time frame before the election. Realistically, based on the experiences of other publishers, we knew that success would look like a few thousand users at most — and in time this bot will get there, too. So to have 400 users in such a short space of time is impressive.

Post-election there has been a small drop off in active users, but nothing huge

One of the things that first appealed about building a bot was the idea of having a group of people to whom we could send articles, ask questions and get feedback who we can keep close to us. This is evident in the 93 per cent stickiness we saw with the bot up to June 7. These users came for the novelty of talking to an election chatbot but stayed for our daily updates.

The majority of our users are aged 25–34 (22 per cent), although the age demographics were fairly split; 19 per cent aged 35–44, 11 per cent aged 18–24 and 10 per cent aged 45–54. And 46 per cent of our users are male, 19 per cent female, 38 per cent unknown.

Of our users, 95 per cent are UK residents (followed by the US and France) which is not surprising. Users have to enter their postcodes fairly early on in the process, so any users outside the UK might be stuck. The fact that so few people from outside the UK ever started the process meant that it wasn’t particularly pressing to address this.

Next steps

While all of the numbers are promising and we’re optimistic that the bot will continue to grow and hold the attention of our users, it hasn’t been without hiccup. Ensuring that the links were entered into the bot’s Google Sheet in time and in the correct format isn’t a huge amount of work, but didn’t always go to plan and we did end up pushing out-of-date text on a couple of occasions. We didn’t see any sign that this affected stickiness, though.

The next phase for the bot will be to maintain the service that our users have found so useful by continuing to publish our updates at 5pm, to to stay true to the original mission of experimenting with the platform and doing creative things that other publishers are not doing. What this exactly looks like at this stage we haven’t decided on, but I’d welcome your feedback if you’ve used the bot. We’re also looking to roll bots out on other pages we manage, but more about this at a later date.

Follow me on Twitter at @williamhpark