Don’t Fear Chatbots If You Have a Rich Content Strategy

“OK, Google, how do I integrate helpful chatbots into my existing content strategy without spending a fortune or frustrating my audience?”

Many brands fall into one of two traps:

• Indecision – taking the wait-and-see approach to AI-based bots while continuing to create content without a back-end strategy. This increasingly creates more problems down the road when the brand embraces AI conversational interfaces for customer relationships.
• Haphazard implementation – succumbing to the temptation to jump in, adopting and implementing tools without considering how they will scale or integrate with the brand’s overall content ecosystem.

To avoid these extremes, follow the advice of Noz Urbina, founder and content strategist of Urbina Consulting, who says a strong underlying content strategy can – and should – inform your chatbot strategy. If you’ve invested in structurally rich content, then you’re likely more chatbot-ready than you realize.

In his 2018 Intelligent Content Conference talk, Chatbots: How They Can Be Integrated Into Your Existing Content Strategy, Noz shows how you can repurpose many of your existing content strategy elements for use with chatbots.

First, though, Noz suggests everyone should work from the same idea of what chatbots are and what they are not.

What is a chatbot? Terms and definitions

A chatbot is simply “software that automates the task of talking with people, especially over the internet,” according to Kristina Podnar’s definition. Bots fit into the broader category of conversational interfaces or language-based user interfaces, and can be either text, voice, or a mix of both.

Here’s an example of a typical chatbot interaction.

The user types or asks, “What’s the temperature?”

If asked via Siri, Cortana, Alexa, or other voice-based option, the response is verbal.

If the interaction happens in a text-based chat window, the bot returns a text answer, which may include charts and other visuals.

In a mixed presentation, the bot vocalizes with the current temperature and displays a temperature forecast chart, with prompts the user to tap for more information.

This interaction pairs a question and an answer. It requires the bot to understand the user’s search intent and return the answer in context.

To determine the intent, the bot recognizes a command from a pre-defined grammar or uses natural language processing (NLP) to parse the input.

A bot limited to fixed grammar consists of known commands, such as “get weather.” Noz explains that this bot style is useful for chatbot devices used in cars or other cases where the goal is to get around the need for a physical interaction.

The weather example response contains fixed and variable parts. “The temperature right now in” is a fixed response. The city and temperature are variables.

NLP-based bots are the front ends to more sophisticated cognitive systems, as Val Swisher explained in a presentation on natural language processing at the Intelligent Content Conference.

An NLP parses sentences into components. Think of a sentence as a loaf of bread. Each word or phrase is a slice in the loaf.

“A natural language processor works by looking at each and every sentence and figuring out all the different parts of speech so it can understand the meaning and intent of that sentence,” Val explains.

Instead of a fixed-grammar response to a limited set of questions, NLP bots enable a more conversational flow because they enable the bot to remember the topic and context of previous questions.

In the conversation below, the bot understands that “they” in the second question refers to “elephants” because of the context of the first question. This ability to relate the pronouns to the nouns based on context is called anaphora resolution.

Conversational interfaces treat language the way programmers treat code, Noz says.

But that doesn’t mean content strategists suddenly need to become coders. Much of the work to prepare for chatbot interaction overlaps the work done by content strategists to improve customer experiences, make sure content ranks for search, and prepare for personalization.

Draw from your journey maps and task analyses

The working parts of chatbot technology shouldn’t spook content marketers who have a good handle on their audience’s tasks and real-world journeys.

Think back to the early days of your content marketing and content experience efforts. Likely, your team deliberated carefully to determine journey maps, including analyzing the questions your audience might have as they try to accomplish each task. Those questions probably determined part of your content plan, whether or not you were considering a chatbot.

“We create, as part of walking through the journey, question and answer pairs,” Noz says. “We pair the questions with answers, either from content that we do have or requirements for content that we need to make.”

If you haven’t done a task analysis, now’s the time. The quick and dirty way, as Noz puts it, is to check your search logs to see what people are looking for and ask your customer service and support department what questions people ask over and over.

To learn how to do a proper task analysis, Noz offers these resources:

• How to Conduct a Top Task Analysis
• What Really Matters: Focusing on Top Tasks

Once you’ve developed the set of user tasks, you can decide which lend themselves to chatbots and which don’t. A common pitfall is to believe that because your content has covered everything in the past, so should your new bot. But remember: A bot that successfully performs one job is more valuable than a bot that poorly attempts to do all the things.

A careful task analysis can help you decide where channel handoffs make sense. A channel handoff is the point where a user asks something your bot doesn’t understand or can’t answer. It’s an opportunity for transparency and candor users will appreciate. Instead of asking more clarifying questions, your program flags a human representative to intervene or provides a link where users can dig for more answers.

Based on your bot’s capabilities and your own objectives, determine whether your bot’s assistance will go narrow and deep (as a sort of digital subject matter expert), or shallow and wide. It’s impossible to be both deep and wide, according to Noz, because no one’s resources are unlimited.

“You have to choose. The reason we do a T and not a square is because we can’t cover everything,” he says. “It’s not going to work.”

Reuse your content chunks

Here’s where your investment in intelligent content pays off: Instead of rewriting content for your chatbot, reuse the content chunks you have in your CMS.

Wait, what is a content chunk?

Good question.

A content “chunk” is simply a unit of content – typically a smaller part of a larger work. Here’s an example drawn from a presentation Noz gave a few years ago. The revised version breaks a formerly dense paragraph into shorter labeled chunks that make the content easier to understand.

Content chunks can be fed directly to a chatbot to answer wildly different questions and even serve diverse user intents when broken up to repackage and reuse as long as they’re stored as structured content in a shared content repository.

A content chunk can provide a simple answer to a question that’s perfect for a chatbot response by text:

It’s also well suited to provide a voice response:

But that’s not all: Content chunks also happen to be SEO and featured snippet friendly.

Metadata and modeling matter, too

Another great chatbot resource is your current content model, the framework that documents the structure of your content (i.e., content types, metadata, and tags).

CMI Chief Strategy Advisor Robert Rose has said, “… a great metadata strategy is, in itself, just as important as the content that’s created.”

You may need to make your metadata and taxonomy even more specific to support chatbots. That’s because a chat interaction is like a human conversation powered by a search engine. But search engines don’t quite work the same way as the human brain.

“When you say something, I search in my brain and come out with a response,” Noz says. “The more I know about your situation the easier it will be to get that right answer first time.”

The same is true for search engines. A taxonomy makes connections among metadata tags. Connecting concepts and content through taxonomy helps your bot return the right answers.

One of Noz’s clients experienced a 70% increase in chatbot response accuracy just by fine-tuning the taxonomy, without changing the underlying content.

Adjusting your metadata and taxonomy for a chatbot takes work. But Noz offers this encouragement: “What you must have for chatbots is extremely helpful everywhere else.”

Content strategy work supports a chatbot strategy (and vice versa)

The more strategically your brand adopts chatbot technology, the less work (read: investment) you’ll have to put up later, when audiences expect to interact with brands automatically — on their terms. In other words, the more content you create today without accommodating for conversational technology, the harder it’ll be to convert that content set when you’re scrambling to catch up.

Thankfully, though, if you’ve structured and categorized your content to this point, then your team is likely more chatbot-ready than you thought.

Here’s an excerpt from Noz’s talk:

For more practical help designing the perfect chatbot strategy for your brand, plan to join fellow technologists and content marketing practitioners at the upcoming ContentTECH Summit. Register here for event updates.

 Cover image by Joseph Kalinowski/Content Marketing Institute