Skip to content

How and Why (or Why Not) to Build a Chatbot

how-why-build-chatbotLet’s say your organization has the best content in your industry. Prospective customers go to your site, enter their questions in your search box, navigate through a few clicks, and – voilà – they get instant answers to their questions.

Excellent. For today. But are you ready for tomorrow, when your competitors lure those customers away with a superior Q-and-A experience? They won’t do it by hiring thousands of people to take phone calls. In 2011, Gartner predicted “by 2020, customers will manage 85% of their relationship with the enterprise without interacting with a human.”

Customers will manage 85% of business relationships w/o humans by 2020, says @Gartner_Inc. Share on X

How will your competitors lure those curious customers away from your superior content if not with human beings?

With chatbots. Friendly, helpful chatbots.

Unless you beat them to it.

So says Cruce Saunders, founder and principal content engineer at [A], in his Intelligent Content Conference talk Engineering Content for Chatbots, AI, and Marketing Automation. In this article, I sum up some of Cruce’s advice. Unless noted, all images and quotations in this post come from his talk.


Chances are, you’ve interacted with a chatbot, even if you didn’t know it. A chatbot (also called a bot, a virtual assistant, or an intelligent personal assistant) is “software that automates the task of talking with people, especially over the internet,” says Kristina Podnar in this article from which I borrowed the animated example below. This example shows Taco Bell’s chatbot – “tacobot” – sounding downright personable (“Sounds good,” and so on.)



Some chatbots use artificial intelligence (AI) and some don’t. A simple, scripted chatbot, like tacobot, uses programmed-response technology based on rules or decision trees. “Its paths are limited, and users select from defined options,” according to a recent UX Booth article. On the other hand, the article explains, an AI-powered chatbot – like Google, Siri, or Alexa – responds based on machine-learning or natural-language-processing systems. It deciphers people’s input, responds based on what it knows so far, and then “turns the user’s input into more data,” continuously updating its algorithms.

AI-powered or not, chatbots aim to respond to basic requests in real time, “freeing up humans to do more creative problem solving,” says Cruce. He describes chatbots as an “increasingly interactive and vital way to get at content.”

#Chatbots are an increasingly interactive & vital way to get at content, says @MrCruce. #intelcontent Share on X

Why you might want to build a chatbot

Not every company needs a chatbot. You may want to build one if certain queries could be handled in an automated way. A successful implementation, according to the UX Booth article, can have the following benefits:

Chatbot examples

Cruce cites three progressively human-like chatbots:

  • Mastercard chatbot (via the Facebook Messenger app)
  • Alexa (via Amazon Echo)
  • Nadia (created by Soul Machines and powered by IBM’s Watson software)


The Mastercard chatbot, which communicates via text messages in Facebook Messenger, answers questions that don’t need to be handled by a person: How much did I spend in restaurants in September? What are my offers? What are the benefits of my card? How do I reset my password?

Amazon Echo’s intelligent personal assistant, Alexa, moves the conversation from text to voice. You talk to Alexa and Alexa talks back. This assistant goes beyond answering your questions; it can play music, create to-do lists, set alarms, stream podcasts, play audiobooks, and provide updates on weather, traffic, and news. This type of assistant can even be programmed to have a little fun. (Siri is a similar example. Try telling Siri, “I see a little silhouetto of a man”; the response especially tickles me when delivered in one of the Aussie voices.)

The Soul Machine’s intelligent personal assistant, Nadia, is an experimental avatar designed by a team in New Zealand and Australia. Cate Blanchett created the voice recordings. If you talk with Nadia, “she” sees and hears you, adapting her answers according to your tone and facial expression to fit your presumed emotional state. Here’s what Nadia looks and sounds like:

Chatbot content behind the scenes

Most chatbot platforms depend on authors to develop an independent repository of questions and answers. Often, the authors duplicate this content from other systems. This redundant effort is expensive. As Cruce says,

Subject-matter experts need to be able to maintain content in a single source. The more overlapping content repositories we introduce, the more effort, cost, and risk we introduce. Chatbot content should exist in the core CMS.

#Chatbot content should exist in the core CMS, says @MrCruce. #intelcontent Share on X

Within your CMS, chatbot content could look something like this:


Within the CMS, the chatbot content can live alongside related articles and documentation instead of living in a separate chatbot platform. “The chatbot ideally calls that content in real time,” Cruce says.

Chatbots and humans

When visitors interact with a chatbot, they don’t necessarily know it’s not a person. It’s up to the chatbot owner to clarify – by the image and the chatbot’s name, for example – that visitors are interacting with a machine. Here, for example, you can tell (or can you?) that Cruce is talking with a chatbot.


Eric Savitz, a Forbes writer, describes chatbots as giving people a self-service experience that combines “the conversational attributes of live chat or a phone call” with “the ultimate in automation – zero human contact.”

Granted, robot-powered customer experiences can be annoying or weird in these early generations of the technology. Today’s chatbots often miss the mark, sometimes embarrassingly so (for the company anyhow – entertainingly so for the rest of the world). Many high-performing customer service and sales chatbots have an option to hand off questions they can’t answer to human representatives. In this way, robots and humans work together to serve customers.

Even though chatbot technology is far from perfect, it holds undeniable potential – in a way that scales – for answering the most common questions asked by your most promising audiences. There’s no fighting the use of chatbots. Someday, they will be as common as automated phone systems. How about we make them better?

Chatbots hold undeniable potential to scale answers to common questions of audience. @MrCruce #intelcontent Share on X

Why chatbots need input from content strategists and content engineers

When creating chatbot content, companies often make the same mistake as when creating any new type of content: They copy and paste from existing sources instead of a single source. Ideally, you set up your system so that all chatbot content – primarily compact answers to common questions – flows directly from the same CMS that supplies your other customer-facing content.

Cruce is talking about classic content reuse, aka COPE content: create once, publish everywhere.

On the theme of reuse, Cruce announced, midway through his talk, that, at that moment, history was being made. SpaceX, Elon Musk’s private space-flight company, had just launched Falcon 9, “the world’s first re-flight of an orbital-class rocket” according to this video.

A part of the rocket had been refurbished after an earlier flight, something that had never been done before. The savings were estimated in many millions of dollars.

Cruce drew the parallel: Our content assets are like those rocket assets. We make significant investments in our content assets. Why in the world wouldn’t we reuse them if we can?

Setting up your content for reuse is easier to say than to do. You need to work with a content strategist (or at least think like one) to develop appropriate content models and metadata, among other things. You may need to work with a front-end and a back-end content strategist.

And you’ll need to work with a content engineer to connect the chatbot with your CMS, among other things.

In case you’re thinking of going DIY and tackling the strategy and engineering of chatbots yourself, I can only wish you luck. A blog post like this can’t give you the guidance you need. I include this diagram below (with chatbot functionality represented by robot heads) not so much for you to study as for you to appreciate the need to collaborate with people with backgrounds in content strategy and content engineering. Few of us could create this sort of diagram on our own:

Going DIY for your #chatbot? Good luck, says @MarciaRJohnston. #intelcontent Share on X

The new multichannel content stack:

Click to enlarge
Click to enlarge

Image source

For a deeper dive into the techier aspects of this topic, see Cruce’s Resource Guide: Engineering Content for Bots, AI, and Marketing Automation.

New Tech Friends on the Marketing Block

5 steps to develop a chatbot

When you’re ready to develop a chatbot, follow these steps: journey, research, model, engineer, and deploy, as shared by Cruce.


Step 1: Map the customer journey.

Journey maps “show us the customer experience in context.” Work on your journey maps with a content strategist and business stakeholders “to understand what your content needs to be doing in its context.”

Step 2: Research what your audience wants to know.

Figure out your audience’s burning questions and the terms they use to phrase those questions. You can do this research in various ways: gather SEO data, review session data – even get radical and talk with people. Whatever it takes to get inside prospective customers’ heads.

Step 3: Build the content model.

Create a content model that specifies the structure of each of your team’s commonly created content types. Also, specify the ways those content types relate to each other. Build your content model with a content engineer and content strategist. Without an accurate model of your organization’s content, you can’t know what kind of technology you need or how it needs to be set up.

Step 4: Engineer the technology to support your content model.

The content engineer maps your content model to technology. Cruce had a lot to say about content engineering that I couldn’t squish into this post even if I had understood it all. (Code snippets, anyone? Microdata, schemas, taxonomy, and clean-content APIs?) Here’s what you need to know: Find yourself a good content engineer.

Step 5: Deploy your chatbot – when it’s ready.

Before you unleash your bot on the world, test it in development, staging, and production environments. Work out as many kinks as you can before your prospective customers make it say silly things. Work on the voice, tone, and message targeting and interactions with a content strategist who understands interactive content.


Increasingly, when we humans get curious about something, we expect answers to materialize instantly. Chatbots – done well – provide a scalable way for companies to fulfill that expectation.

Done poorly, of course, chatbots frustrate people and damage brands. Don’t bother making a chatbot unless your company is committed to dedicating the resources needed for creating a positive user experience.

As Cruce says, “Consumers are increasingly talking with our content, asking it questions. We need to make sure our content can talk back.”

Is your content team creating chatbots? Thinking about them? Let us know in a comment.

Here’s an excerpt from Cruce’s talk:

Sign up for our weekly Content Strategy for Marketers e-newsletter, which features exclusive stories and insights from CMI Chief Content Adviser Robert Rose. If you’re like many other marketers we meet, you’ll come to look forward to reading his thoughts every Saturday.

Cover image by Joseph Kalinowski/Content Marketing Institute