2016 was the year of the bot in journalism. In this edited extract from the forthcoming second edition of the Online Journalism Handbook, I outline what bots are, how bots have been used by media organisations from early Twitter bots to the recent wave of ‘chatbots’, and some tips and tools for getting started with journalistic bots.

‘Bots’ are ‘robots’ – only on the internet. Without the mechanical body of their physical counterparts, all that leaves is a disembodied computer script, normally created to perform repetitive tasks.

This broad description takes in a whole range of activities, and so the term ‘bot’ is used to talk about very different things in different contexts:

• In search you might talk about bots used to index webpages, such as the ‘Googlebot’.
• In finance and commerce you might talk about bots used to monitor information online and respond to it by buying or selling things.
• And in advertising and politics you might talk about bots being used for nefarious purposes: for example, to make it look like more people are viewing webpages, clicking on adverts, or arguing for a particular candidate.

This article isn’t about any of those.

In the context of journalism and publishing, the term ‘bot’ is normally used to refer to something which users can interact with. Examples include:

• A bot which automatically publishes updates on a particular social media account when it receives new information from a feed (such as new articles)
• A bot which can supply article suggestions in response to a query from a user
• A bot which attempts to provide answers to questions given by users

The Twitter bots: alerting, aggregating and monitoring

One of the earliest uses of bots in journalism was the ‘Twitter bot’. These have tended to automate three particular aspects of journalism: alerting, aggregating, and monitoring. The result is a level of coverage which would be unlikely to be provided by human reporters alone.

Alerting is relatively simple: many website Twitter and Facebook accounts are actually updated by ‘bots’ which automatically publish every time a new story appears on the publication’s RSS feed.

Aggregating can take a number of forms. The @North_GA news bot, for example (now suspended), aggregated RSS feeds from a number of news sources in the northern part of the state of Georgia.

Others automatically retweet material matching a particular search (for example updates with a particular hashtag or phrase, or stories matching a particular search term on Google News).

Monitoring often involves a script which tracks what public figures or bodies are doing or saying: @stopandfrisk, for example, fetches and tweets data from the New York Civil Liberties Union on police use of stop and search; while the Facebook page Sir Keir Starmer in Parliament posts an update every time the politician does something in Parliament (as recorded by TheyWorkForYou).

Some include a threshold: WikipediaLiveMonitor (@wikilivemon) monitors edits on Wikipedia and tweets when the frequency of edits on a page is unusually high (indicating something newsworthy), while other Twitter news bots tweet when an edit is made from an IP address associated with the Houses of Parliament or other public body.

Your Reps On Guns (@yourrepsonguns) retweets politicians when they mention firearms or related terms.

Useful for amplifying, revealing and highlighting

Lainna Fader, Engagement Editor at New York Magazine, notes that the obsessive repetition of bots makes them useful:

“For making value systems apparent, revealing obfuscated information, and amplifying the visibility of marginalized topics or communities.”

An analysis of 238 Twitter ‘news bots’ identifies further uses of news bots, including critique and opinion (@NYTanon, for example, critiques the use of anonymous sources in the New York Times, and @cybercyber highlights the overuse of the term ‘cyber’ in news reporting) and entertainment (some bots, such as @DrunkBuzzfeed, ‘remix’ content with amusing results).

The first wave of Twitter bots was largely based on automated technology: tools like Twitterfeed (now closed) made it simple to connect an RSS feed to a Twitter account.

Later, services including IFTTT and Zapier made it easy for people to connect all sorts of services (for example, to update a Facebook page whenever a Google spreadsheet was added to).

A brief history of chatbots

A second wave of ‘chatbots’ came as messaging platforms opened up their platforms to developers. The chat app Telegram was one of the first, launching a bot creation API in June 2015. “[Bots] can do anything,” the company announced.

“Teach, play, search, broadcast, remind, connect, integrate with other services, or even pass commands to the Internet of Things.”

TechCrunch and Forbes Telegram bots would invite users to choose particular topics, authors and sections to subscribe to, but could also ask questions “like “Who is Jack Dorsey?” or “What is Disrupt?””. The BBC realised it was an opportunity to reach audiences in places where the BBC Uzbek language website was blocked.

Messaging app Slack added a bot directory six months later. The New York Times used it to create a bot for election alerts while Digg’s Slack bot provided story recommendations based on keywords. Bots could also be used for internal purposes: nyt-campfinbot was created to notify New York Times reporters about filings to the Federal Election Commission’s website.

In 2016 the trickle became a flood: Kik, Line, Skype, Facebook Messenger, and Viber all introduced bots, while Twitter added bot functionality to direct messages and Amazon Lex was launched to help people create bots for its Alexa platform.

Facebook’s launch of bots on its Messenger platform included a demonstration of CNN’s app. The Guardian used the platform to create a Sous-Chef Messenger bot that provided recipes to users when they suggested an ingredient or keyword.
Bots, it was said, were the new apps.

Why news organisations rushed to build chatbots

Why the big rush? For publishers a big driver was the realisation that people were no longer adding new apps to their mobile phones, and the ‘new frontier’ on mobile was within those apps which people already spent most of their time using.

For the chat apps it was the rise of ‘messaging as platform’: in China people were already able to pay for goods, services and content within chat apps like WeChat, and western chat services saw an opportunity to expand.

But a third trend was related to technical capability: artificial intelligence (AI) was now at a stage where it could be used to power bot technology, opening up all sorts of new possibilities. Even some Twitter bots have become more sophisticated and able to respond to queries in different ways.

Chatbot creation tips and tools

Tetyana Lokot and Nicholas Diakopoulos suggest four elements to consider when designing a bot:

1. The input;
2. The output;
3. The algorithms that turn inputs into outputs;
4. And the function or intent.

A number of services have popped up to make it easier for people and companies to create bots, including Chatfuel, Botsify, Api.ai, and Gupshup.

These tended to offer basic automation, as well as ergodic functionality (users choose from a limited number of options). But the real challenge is designing bots which respond to the wide range of language used by users: the Slack scheduling bot MeeKan reportedly needed over 2,000 sentences just to deal with one meeting request.

Simple bots can be relatively easy to create with these tools, and it is also easy to create automated bots using IFTTT or Zapier to connect an RSS feed or Google spreadsheet to Slack, Twitter or other services. The key is to decide what you want to do this for.

Have you created a bot for journalistic purposes? I’d love to hear about it.