How to Build a Chatbot-Powered Newsroom
Bot & ☕️ with Rebecca Harris
In the consumer space, three big areas of interest seem to be emerging among chatbot developers: travel, recommendations, and personal finance.
Products in these areas are mostly “pull-products”.
You would want a travel bot to engage with you only when you intend on traveling for business or leisure.
You would want chatbots to send recommendations for shows, movies, books, and restaurants mostly when you have an intent and a pre-existing demand for consumption.
You would want the personal finance bot to tell you how much you spent on groceries last week mostly when you are curious.
These products have some push elements but they are mostly on-demand or pull experiences.
And Then, There’s News
But news is a category in consumer space where a lot of people expect to get content pushed to them. News products are fundamentally “push products”. It’s quite common for people to consume news via push notifications from their news apps. So, in our “post-app” world, such expectations and behaviors will most likely continue.
I recommend reading Sandi MacPherson’s piece on her personal bot. She talks insightfully about the push vs pull world in that piece.
Several publishers — from The New York Times, The Washington Post, Forbes, and CNN to Quartz, Mic, TechCrunch, and Fusion — have already experimented with varying forms of chatbots on various messaging platforms.
We have also seen narrow and niche implementations of chatbots like WTF is brexit, Minitextbot, Rumour Filter, and others.
Check out Johannes Klingebie’s piece on the state of journalism bots for details on these “weird” bots.
Most of these news bots do not fascinate me at all. They send you articles that fit a specific keyword. They let you subscribe to certain keywords and topics which will prompt the bot to pro-actively send you related articles. Some of them send us a daily digest of most popular articles at fixed times.
Boring. None of the publishers’ chatbot feel quite sticky to me.
I would rather consume news in the form of articles on Pocket, Nuzzel, news apps, and the browser or social feed.
To be fair, a lot of the product development and user experience really depends on what is and isn't allowed by the underlying messaging platforms.
For instance, the implementation of CNN’s chatbot is so different on Kik and Facebook. This is mainly because of how Kik and Facebook have designed the usage of buttons in their chatbot experiences.
But… there’s always an exception to the rule. For news, it’s Purple. And it’s sticky.
I remember Chris Messina talking about it at a bot brunch that was hosted by him and Esther Crawford at betaworks a couple months ago. I tried it out there and then. I have been a user ever since.
The mindset with which I started using Purple was that of an average person who knew nothing about chatbots. Over the months, my bar for getting impressed by a chatbot has become considerably higher. Yet I’m still continually impressed by Purple.
It pushes all the political news at the right time in the right amount. It looks so simple — yet, it is so effective. Most bot enthusiasts I know love its implementation. I have always admired how Purple’s founder Rebecca Harris has developed the experience.
I thought it would be very helpful to learn from her how she built the news experience using chatbots as a medium and how she thinks about chatbot-powered journalism. So I asked her a few questions.
Sar: I have been a big fan of Purple. It has been my go-to source for political news for the most part of past couple months. Tell us what Purple is and what your vision for it is.
Rebecca: Thank you so much, I’m so glad you’re part of the Purple community! Our vision is to build something that makes it really easy to be informed and actually learn about the issues that matter.
The media industry has totally failed to inform my generation. Content either gives you headlines but no context, or dumbs things down way too much, or it’s biased so you can’t trust it. That’s scary to us for the future of democracy, so we started Purple to make it easy to understand important, often wonky issues.
The easiest/best way to learn about something is to talk to someone who is really knowledgeable about it and super interested in it. So we thought ‘how do we recreate that experience with technology?’, and that’s how we came up with using messaging.
From that emerged a new form of content: bite-sized, choose-your-own adventure stories written in a way that feels like your nerdy friend is texting you.
Sar: Your product has built quite a name for itself. It gets brought up a lot in tech communities and twitter. Did you put in targeted efforts in marketing and user acquisition early on? Or is it just word of mouth?
Rebecca: So far, the vast majority of our growth has been via word-of-mouth. It started as a Google Voice experiment that my cofounder David Heimann and I tried with 50 people.
We covered an early Republican primary debate live via text message, and saw an amazing response in terms of engagement. It grew very quickly from 50 to 100 people, and from there we built a custom system to create and distribute messages as our user base grew.
We’ve found immense power in the Purple community. What unites every Purple user is a desire to learn. They are intellectually curious. And because we’re messaging with them we’ve built up really personal relationships with our userbase.
Sar: Most bot builders are trying to figure out what the right cadence for pushing content and sending re-engagement notifications, and what the right length of a message is. As a user, I think Purple has been doing a great job on all these fronts. How did you figure it all out? Do you think that being a news channel makes it relatively easy for you to keep your product top of users’ mind due to the core pushing experience?
Rebecca: The frequency of messages is something that we think about a lot, especially when we started on SMS. Users were inviting us into a really intimate space that is typically solely reserved for family and friends. We didn’t want to blow up people’s phones or spam them. We found that one message a day is the right cadence for our users, with the exception of breaking news alerts or live event coverage. That one message becomes something users look forward to, and we’re not overwhelming them with information.
That’s how we developed the use of keywords (words in all caps that users can text to dig deeper into that topic). That way, it’s up to the user to determine how many messages they want from Purple. We try to limit our push messages to one per day, unless there is something breaking or there is a live event that we’re covering.
I think sending re-engagement notifications is a good idea, but really difficult to find the right frequency/cadence. Length of message is something that we think about a lot, but with most messaging platforms there’s a character limit anyway. That might seem like a big constraint, but I think it’s a great thing because it forces you to focus on the content itself.
I think that focusing on current events keeps us on top of users minds because it means we have a push mechanism that actually adds value to their lives. We don’t push notifications for the sake of pushing notifications.
Sar: I believe that conversational design and the voice of the chatbot would become core differentiators as more layers of the tech get commoditized, . This is especially true for news chatbots: the voice and the story-telling skills are what differentiates one journalist or publisher from another, while the factual content is almost always the same. How did you think about building a voice for Purple?
Rebecca: I 100% agree, I think voice is the most important thing when it comes to the messaging experience. Think about it: we use messaging organically to have conversations with family and friends. So naturally, we think about our voice very much from a position of how we would text this information to our friends.
It developed literally as an extension of my own voice, but it’s not that it was me specifically that made it successful. What makes our voice successful is that it’s genuine. Our goal is for Purple to feel like your nerdy best friend is messaging you, that’s our north-star when it comes to writing content.
Sar: Do you have any favorite chatbots in news space?
Rebecca: I really like the Verge’s messenger bot, and I like the Quartz app, but honestly I’m really disappointed by most news bots because they’re essentially RSS feeds in Messenger.
Media companies are looking at messaging platforms as a giant new distribution platform, which they are, but I think taking to that new platform with the mindset of the old platforms is a mistake. Just like when the internet was created and the New York Times’ first website was basically a PDF of their front page.
Sar: You recently switched your product from SMS to Facebook Messenger. Any specific reasons behind that decision? Did you notice any interesting changes in how users engage between the two platforms?
Rebecca: There were a few reasons we made the switch (cost, deliverability issues, etc.) but the main reason was that we wanted to focus on one platform first.
We want to be where users are, so we definitely plan on being multi-platform again soon.
Facebook’s UI allows us to use quick reply buttons as a way for users to navigate through stories rather than having to message a keyword every time. So we’ve noticed a major uptick in how far into a Purple story users go because the ease of navigating that story has increased.
Sar: Have you thought about monetization strategies for Purple? If not, how do you think news bots might make money based off of these new messaging channels?
Rebecca: We’ve thought about monetization, but the main thing we’re focused on right now is engagement and focusing on the product and the content itself.
I think that there is definitely an opportunity for bots to make money in creative ways. Reply Yes (the SMS bot I mentioned before) sends you album recommendations, and if you decide to buy one it’s as easy as texting “buy” (once you’ve input your info into Stripe). They’ve sold millions in vinyl sales.
The opportunity to monetize bots increases as mobile payments become more frictionless on messaging platforms.
It’s still very early, but I think we could see a resurgence in the subscription model with news bots. It is completely dependent, however, on what news bots are serving up. If it’s just links to digital content, then it’ll be almost impossible to monetize via subscription.
Sar: How do you track user engagement? Analytics has still bot been figured out in the bot space. We are still waiting for the “MixPanel of bots”.
Rebecca: We built our own internal system for tracking user engagement. Unfortunately, there is no analytics platform that tracks data on “conversations”. This form of content is so new that an analytics platform for it doesn’t exist, but a Mixpanel of bots would be dope!
We look at daily user engagement including how far into one of our interactive stories users go through.
Sar: There are so many frameworks and tools available. It is very easy to ship a simple chatbot with the frameworks and tools available — but, the best chatbot products are deceptively difficult to build, ship and grow. What is your take on the flood of chatbots we are seeing today?
Rebecca: Here’s my take on it from a media perspective. When the internet was created, media companies saw it as a huge distribution platform and the early news sites were basically PDFs of the newspaper. When social media platforms were created, media companies saw them as a huge distribution opportunity, and they basically took traditional content and posted links to it.
The problem with these early approaches was that we weren’t looking at how the platforms were being used organically by human beings. Today’s news sites look very different than the newspaper PDF days because now, the media industry understands how the internet fundamentally works. We know the opportunity it gives content creators to go way, way beyond just posting a PDF of an article.
Well, I see the same thing happening right now with messaging platforms. Media companies are building bots that are essentially RSS feeds. That to me is the equivalent of putting a PDF of a newspaper online.Think about it: These are natural communication interfaces. In order to truly take advantage of the power of messaging, we need to think conversation first, not transaction first.
Think about it: These are natural communication interfaces. In order to truly take advantage of the power of messaging. We need to think about how they’re organically used and adapt to that, rather than trying to force media made for different platforms onto messaging just because it’s a new way to distribute it.
Purple is really one chatbot that I find myself going back to almost daily. I love it.
I wanted to catch up on all the news cycle on Clinton and her health yesterday. I pulled up Purple. I don’t like reading long pieces when it comes to conspiracy theories and speculations in politics. Since I was watching the tennis final with friends, I rushed through typing my query. It was vague. Purple handled it very nicely and contextually.
Most people do not care whether its a human or a machine behind the hood. If the experience matches expectations and is always reliable, the product is fantastic.
Give it a shot.
