How to develop content for useful, not-scary chatbots and other talking interfaces

Lately, I’ve been thinking a lot about how to do content strategy for chatbots and other conversational interfaces. These days, everyone seems to want their interface to talk. Here’s an unscientific graph that shows how many more enquiries I get about bot content now than I did about a year ago:

I grew up in the 1980s and my feelings about robots are mostly informed by pop culture. My first and most enduring memory of robots are these wondrous and helpful guys:

PS Hoth Set by Gordon Tarpley

I really wanted one of these:

Or one of these:

But then there was also this scary thing:

The Terminator

Is he even a robot? Are the machines really coming to get us? Also, where are all the lady robots? (Oh right, they’re busy being personal assistants …)

Bots are cool. Machines that act like sentient beings are cool. What business wouldn’t want their own chatty, quirky, friendly interface robot? Yay!

But before we get too excited, let me put on my buzzkill hat and remind you that in a business context, talking interfaces exist to provide a service. To people. To your customers. Remember them?

As strategic business people with limited resources, we need to shelve how neat all of this is for a minute and think about how — no, IF — a conversational UI will advance our business goals. More importantly, how will it improve the experience we provide to the humans who are good enough to give us their time and money?

So my first tip is to ask yourself and smart people you trust whether your business needs a bot.

Does your business need a bot?

Developing a bot is technically pretty easy these days and there are lots of tools and frameworks to get you started. The tricky part is designing a talking interface that will:

• Sound authentically like your business
• Enhance your customer’s experience with your business
• Provide a valuable service to your customers
• Advance your business goals
• Avoid being a creepy uncanny valley-ish clown

Although there are some exceptions, chatbots and most talking interfaces are not intelligent. Although there are lots of smart people working on artificial intelligence, the bot you’re probably going to build is a piece of software that is programmed to converse with people in a natural way. It’s basically a glorified phone tree designed to respond to requests posed in natural language. As such, bots carry the same assumptions, blind spots and blunders as their creators.

Bots carry the same assumptions, blind spots and blunders as their creators.

5 tips to help you think about bot design

1. Write like a human, but also like a robot
I’m a big fan of writing like a human. This means being thoughtful about the real people who will use your product and writing in a way that’s clear, welcoming, and helps your users complete their most important tasks.

It’s important to write in a way that considers your audience, avoids jargon and aims for clarity whether you’re writing a product page for your website, text on an interactive button, or the script for a bot. Bot content is unique because users interact with it as though the interface is responding intelligently in real time.

The balance I’ve found is to write conversationally while building in responses that acknowledge the limitations of the interface. For example, instead of responding to an unanswerable question with a funny (awkward) response, consider something that embraces and clarifies the limitations of your bot:

“I can’t answer that question because I’m just a bot. Should I put you in touch with a human?”

Another approach that seems to work is to remind users that they’re talking to a bot with all of its quirks and limitations early on in the conversation:

“I’m Finley. I’m a bot that was made to help you track your order.”

Remember: we love characters like R2D2 because of their robot-ness, not because they fool us into thinking they’re human.

2. Build a character based around your brand values
There’s a lot of great advice about how to develop the voice and tone of your business. Spending some time to think through how your company should sound when it talks to customers is especially important if you’re thinking about developing a bot.

Your bot will be more engaging, more bot-like, if you consider the characteristics of its personality. You may even want to consider writing a backstory about your bot to help inform its character and the language choices in your script. Your bot should also share the same values as your company and those values should be reflected in how it responds. For example, if you’re company tends to communicate more formally and you provide a service that requires a lot of trust and access to personal information, it would probably be inappropriate to create a bot that responds to queries as if it’s a stand-up comedian.

If you’re not sure where to start, consider mapping out where your company sits within the four dimensions of tone of voice. Here’s an example from Nielsen Normal Group:

3. Beware of sexism, stereotypes and bias
Bots are the children of humans with all our limitations, assumptions and biases. They inherit our deficits. Consider:

• Does your bot need a specific gender? If so, why?
• If your bot has a gender, why have you made the assignment?
• Does your bot respond to customers using gender specific pronouns? How do you decide which pronouns to use? Has the user explicitly asked to be referred to using she, he or they? Can they ask the bot to call them something else?
• If your bot speaks in accents or colloquialisms, what is their origin? Is it appropriate or appropriation for you to use them?
• What is the demographic makeup of your customers? Have you sought feedback on your content from a diverse group of people?
• How will your bot handle queries from people who may struggle with spelling or literacy? What about people who don’t speak English as their first language? Literacy rates are lower in the US than most people realize so these are not hypothetical concerns.

4. Design a holistic experience
When you’re building a bot, content is the interface design. It’s the part of the experience that your customers interact and engage with. It’s important to approach the creation of bot content not as a static writing project, but from a product-design perspective.

I’m relatively new to this (as we all are), but here are some of the ways I approach content design for a bot or conversational interface:

• Opportunities: List all of the product opportunities you’re trying to solve by developing a bot alongside the ways your interface will be able to tactically respond to them. This list should incorporate the information your bot has access to that will allow it to to customize and enrich the experience. For example: If your app tracks deliveries, your bot needs to be able to access near real-time information about the location of a package. If this is beyond your technical limitations, then your bot is essentially useless.
• Limitations: List all of the opportunities that you know exist that can’t be resolved through a bot. This is important, because it allows you to plan thoughtful, helpful responses when your customers query for these things (as they inevitably will).
• Off-script moments: Plan for all the inappropriate and ugly things that people will inevitable say to your bot and think about how it will respond. It’s natural for people to push against the edges of a conversational interface to see what they get back. The important thing is to decide how you’re going to handle it. Violent, sexist language is never OK and in my opinion, it’s important not to imply that it is by having your bot play or joke along. On the other hand, you don’t want to preemptively cut off an exchange if someone is just messing around. In my experience, the best thing to do when a conversation goes off-the-rails is to first provide a suggestion to get things back on track and if that doesn’t work, offer to refer your customer to a human being.
• Privacy: When people are talking to a chatbot, they often feel like they’re having a private conversation because we design the experience to feel intimate. Conversing with a bot is a new thing for most people and it’s reasonable to expect that they might not understand how it works. How are you disclosing privacy (or lack thereof) to your customer? What checks and balances do you have in place to protect personal or sensitive information that might be disclosed in a chat? The more sensitive the information a customer is likely to disclose, the more important it is for you to think through the implications of privacy and be transparent.
• Context: Think about the context or situation people will be in when they interact with your bot. What will people be doing when they interact with your product and where will they be? Are they likely to be sitting at a desk with few distractions, or will they be walking, driving or doing something that requires more of their attention? Thinking about the situation your users are in can impact how you design the experience. For example, if your bot is accessed via voice commands and is likely to be used while your customer is driving, it’s probably best to get to the point and avoid cuteness and humor.
• Design for before, during and after: Think about an interaction with a chatbot as one part of your customer’s journey with your company. Understanding and considering where people are in their journey when they encounter your bot will help you better respond to their needs and provide them what they’re looking for.
• Think through entrances and exits: Consider all the ways that a customer moves into and out of interaction with your bot. Invisible UI is a buzzy, cool thing right now, but remember that it’s new and weird to lots of people. It shifts all the standard ways we’ve taught people to move through our interfaces. Many of your customers may never have interacted with a bot before yours so ask yourself how you can clarify what’s happening. Consider methods that give people an explicit choice to interact with your bot rather than simply dropping them into the experience. Make sure to build clear sign posts that allow people to navigate to more conventional forms or support, such as talking to a real person. At the beginning of an interaction, consider providing prompts to help people understand how the experience works and to nudge them towards the kinds of things they can ask.
• Be strategic: Like anything else your business invests time and energy in, it’s important to think through how you will measure and monitor the success of your bot. How is the experience making people feel? Consider explicitly asking for feedback and using what you learn to make adjustments.

5. Write conversations, not independent strings

Stephanie Hay articulates a great process for developing content prototypes in her article Content-First Design. Her premise is that focusing on your users needs and developing a story that answers those needs is the best way to build a useful website or product. Hay refers to this as “design for discovery” and uses the video game industry as an example of an group that does this well.

Developing chatbot content is all about developing a story around a conversation. Each piece of content that you write will feel like it’s spoken to the human being on the receiving end. It will feel personal, so write with care.

A great way to start developing your bot content is to interview real customers about the opportunities and pain points you hope to solve. Listen to their questions and concerns, provide answers, and note their follow-up questions.

You’re doing this right if you end up with a big, messy source document with a stream of questions, answers, follow-up questions, and more follow-up answers. Let the conversation flow naturally and keep track of how it progresses and branches. Make sure to note the language your customers use so you can plan for it and reflect it back to them in how your bot responds.

Once you’ve got a big pile of source material, begin to categorize and organize it to create relationships between different questions and answers. Rewrite your answers in the voice and tone you’ve established for your bot. Read your work out loud to see if it sounds natural, consistent, and kind.

Rearrange questions and answers into patterns so you can map out the beginning, middle, and end of conversations to see how the experience flows. Are you able to resolve your customer’s concerns?

Go back to your customers to see how what you’ve created works and feels to them. How do they feel about the voice and tone in the content? Are you missing important points? Does it feel like your bot is making this experience easier, or more frustrating? If it takes people longer to get to their answer from your bot than through conventional methods, a bot is probably not the right product solution for you.

Final thoughts

People are busy and distracted. Bots are cool, but not cool enough for people to suffer though a confusing, complicated, or offensive experience. Chatbots and conversational interfaces are becoming common enough that the days of getting attention just because you build a talking interface will soon come to an end. Like any product experience, a bot will only be successful if it adds value to your customers.

We’re living in the future and there isn’t yet a blueprint for how to do this work. And the ground is always shifting as new technologies and approaches are developed.

I’d love to hear how the rest of you are approaching this work (post your thoughts in the comments).