“Hello.” “Hi.” “Hey.” “What’s up?” “Howdy!”

When writing for a conversation bot, simply starting the conversation presents a host of questions. How formal or casual should the conversation bot sound? Should the bot use greetings that are common in certain regions, or ones that are accepted by everyone who speaks that language?

Voice, tone, and register strongly affect how humans respond to one another, even when we’re not fully conscious of their influence. They also affect how customers react to conversation bots. Choosing the right approach for each audience is crucial to building that relationship.

Fortunately, conversation bots are up to the task.

How We Speak: An Introduction to Voice, Tone and Register

Humans typically process spoken and written conversation without thinking about the vast quantities of information we use to do so. Only when we need to translate that information — for instance, in order to teach a conversation bot to process conversations as humans  do — does the scope of the task become apparent.

One of the first steps is to understand the difference between voice and tone in a given situation, says Karen Taylor at Kuno Creative. Brands that lack a strong voice often produce conversation bots that act “staunch, corporate, or robotic,” even if they’re well-designed to handle a customer’s needs in the moment.

In addition to voice and tone, conversational language also displays register. “Language register is the level of formality with which you speak,” says educational researcher Sarah Elaine Eaton. Most people move between registers frequently. For instance, “it is not uncommon for Canadian teachers to address one another casually in the staff room, and then adopt a more consultative register when speaking with a parent or school board trustee,” Eaton says.

“Register,” as Eaton uses the term, refers to the level of formality in speech. Voice, tone, and register all play a role in a bot’s communications with human conversation partners. Establishing all three — and adapting them when necessary — can help conversation bots achieve goals more easily.

Does Your Conversation Bot Need to Sound Human?

With so many variables at play in determining tone, some designers are tempted to skip the question entirely. Doing so puts your bot’s effectiveness at risk for two reasons.

First, a more human-sounding conversation bot engages customers more easily. “Humanizing chatbots make it feel as if the bot is actually listening and responding to us,” says Tuba Tezer, head of growth at Botanalytics. Interacting becomes easier for customers, making them more likely to follow the process to its conclusion.

Second, eliminating tone from word-based interactions is nearly impossible. As Tezer notes, a lack of emotional inflection doesn’t create a sense of non-feeling or neutrality. Instead, it makes the interaction feel impersonal or even cold.

chatbot voice train.png

Finding the Right Voice and Tone for Your Bot

Lacie Webb at Roboboogie distinguishes between voice and tone as follows: “Voice is at the core of your brand’s personality and character. It can be human, familiar, formal, casual, or any number of combinations of these attributes. Tone is how you express that character in any given situation.”

In other words, your conversation bot’s voice is its general personality and approach. The voice should consistently sound like your brand across interactions. Your conversation bot’s tone, by contrast, is how it adapts to specific human visitors or customers based on their inputs and their location.

Tone affects how customers interact with the conversation bot in the moment. As an example, imagine a conversation bot for a fast food chain whose voice is well-developed: The bot’s general persona is courteous, friendly, and diligent.

But if the bot greets customers with a “Hey, what do you want to eat?” that’s going to create an awful customer experience, Oscar Ibars writes in Chatbots Magazine. “The customer experience is now broken,” Ibars says, “and many customers will likely place smaller orders just to ‘get through’ the transaction at hand.”

That hypothetical illustrates how tone sets customers at ease and engages them in the right frame of mind to complete their goals. Conversation bots that can read a customer’s own tones and adapt can make for even better conversation partners. For instance, Tricia Surber at Botsociety describes a bot named Rose who “understands conversational slang like ‘haha’ and ‘lol,’” and who can keep conversations moving by responding appropriately to these interjections.


Writing Voice and Tone Into a Conversation Bot’s Flow

Establishing a conversation bot’s voice is easy when the bot has a strong persona, Nivedita Chandra writes in Chatbots Magazine. When a bot’s design team conceives of the bot as having personhood, it becomes easier to predict how this person will generally behave, as well as how this person would likely respond in specific situations.

Next, it’s important to remember that conversation bots communicate in a particular genre: that of the text message. Each genre, including text messages, has its own conventions that influence tone, fiction editor Beth Hill notes.

Since a conversation bot’s genre is clearly defined, it’s possible to use the features of this genre to boost the bot’s overall tone. For instance, the bot can offer responses that use certain interjections found in spoken language. The key is to choose interjections effectively, says Kai Rikhye at Inside Design.

For instance, interjections like “ah,” as in “ah, I see. Let me find someone who can help,” can make the conversation flow more easily. Pauses like “huh” or “um,” however, can make the conversation bot seem stilted, even though these words regularly punctuate spoken conversation.

It’s important to text the way humans tend to text, as well. For instance, periods at the ends of sentences are dropping out of text interactions, as Dan Bilefsky noted in a 2016 New York Times article. Periods are still a mainstay of the consultative register, found in articles, books and business correspondence. In the casual or intimate registers, however, the appearance of a period can take on a passive-aggressive tone.

Some conventions of text interaction that affect tone aren’t text-based. For instance, while testing a conversation bot developed by a friend, GoodFirms content consultant Kim Smith felt a sense that something wasn’t quite right.

“I realized the replies were following the input precisely by 1.5 seconds,” said Smith. “Real conversations don’t have a precision of 1.5 seconds.” By setting the reply time to random intervals between 1.5 and 6 seconds, the developer was able to create the sense of anticipation that appears naturally in human text conversations.

chat bot regional slang.png

Localizing Conversations: How to Change Your Conversation Bot’s Tone and Register for Its Audience

While a conversation bot’s voice should align with your brand and stay relatively static, its tone and register can adapt to suit the needs of localized customers.

One way to do this is through the use of colloquialisms, informal expressions that appear frequently in casual conversation. These expressions are usually region-specific rather than universally understood, linguist Richard Nordquist writes at ThoughtCo.

The right colloquialisms can make a conversation bot’s responses feel more personalized for their human chat partners, Gillian Armstrong notes at Chatbots Magazine. Having timely, relevant colloquialisms at its disposal helps the conversation bot establish and express conversational context.

Some companies are already exploring localized conversation bots, which can adapt the language they use to help customers in a variety of linguistic contexts, says Mitul Tiwari, CTO and cofounder of Passage AI. While a bot that can switch easily among languages like English, Somali, or Chinese expands its reach, it often does so at the expense of personalization.

By incorporating colloquialisms in one or more languages, a conversation bot can make its users feel as if they’re talking to someone who speaks not only the same language, but the same regional dialect they do. For instance, a conversation bot that adapts its language for English-language users in Texas, Liverpool, and Mumbai delivers a more personalized experience for each.

With the help of natural language processing, conversation bots can even teach themselves slang, Andrew Myers-Stanford notes at Futurity. As this feature develops, it will help conversation bot language evolve with human language, reducing the appearance of outdated colloquialisms that can make the bot sound antiquated.  

Into the Future: Automated Conversations That Feel Personal

“Despite its complexity, the human language is a highly unstructured system of synonyms, homonyms, terms, and slang, not to mention possible misspellings, contracted forms, abbreviations, and omitting punctuation rules,” says Serhii Znakhur, an AI developer, data science expert, and project manager at Sloboda Studio.

Historically, computers have lacked the capacity to understand natural language at the level we do, but natural language processing is allowing them to do so. Combined with a deep understanding of your customers and their local or colloquial language conventions, conversation bots can have more personalized conversations in the moment, striking the right tone for the situation.