Top 6 Conversational Skills To Teach Your Chatbots

Mariya Yao
Published in
9 min readSep 13, 2016


This article was originally published at, a multi-media magazine that finds you the best of bots. Join the community to get essential bot news & curated industry content.

Small talk can lead to big business.

That’s what the titans of tech are betting on. Facebook, IBM, Amazon, Apple and Google have all invested insane amounts of money into conversational AI designed to replace human labor and automate business operations.

Their investments appear to be paying off. Mobile AI assistants are predicted to close about $2 billion in online sales by the end of 2016. By 2020, Gartner predicts that more than 85% of customer engagement will be managed without people.

While we can’t predict when AI will fully master human language, businesses of any scale and vertical can take steps to make their own specialized chatbots more attuned and responsive to the way people speak. Successful chatbots already exist for food chains, news outlets, entertainment conglomerates, fashion houses, airlines, banks, and many other industries.

But simply launching a chatbot does not ensure success. Of the over 30,000 bots on the Facebook Messenger platform as of this writing, only a small percentage drive regular user engagement and material boosts to their bottom line.

Just like their human counterparts, chatbots differ in their ability to understand, engage, and serve customers. A dumb chatbot, just like a dumb customer service representative, can wreak havoc on your business in terms of missed sales opportunities, negative brand association, and lost customers.

Luckily, you can look to the best human conversationalists to learn conversational lessons that you can teach your chatbot. Otherwise, your bot’s chit chat with customers will more likely inspire trash talk than a profitable transaction.

Here are the essential conversational skills to master in order to ensure your chatbot’s effectiveness with users:

1. Empathize With Emotions

Sadly, even most humans suck at this skill. This is especially true when a conversation gets heated with criticism or negative emotions. We often respond in these situations with defensiveness, denial, argumentation, or negativity.

However, this tendency is the opposite of what top customer service agents do to resolve tricky situations with irate customers. The best agents empathize with people, even those who might be in the wrong. Just acknowledging and validating an emotion is often enough to make customers feel understood and release negativity, whereas being defensive or argumentative only exacerbates the problem.

Let’s say a user types “You’re stupid” or “This isn’t working” while using your chatbot. Clearly you aren’t solving their problem and they’re getting frustrated. I tested these types of phrases with chatbots and here are some common reactions:

“Sorry, I can’t help you”

“I don’t understand”

“That’s not a nice thing to say”

These responses are entirely useless and only serve to piss users off more. Instead, you can program your chatbot to detect negative sentiments and react with empathy and patience. Here are some better responses to use in these scenarios:

“I see that you’re frustrated. Let’s try a different tactic.”

“I’m sorry this is a frustrating experience for you. Feel free to email to reach human customer support if I’m not able to solve your problem right now.”

2. Proactively Offer Guidance

Most chatbots are designed to be transactional or assist a customer in accomplishing a specific task, such as order coffee, change a flight schedule, or purchase flowers. Yet most fail to make their capabilities clear at the outset, leaving customers to guess at what’s possible.

The most common opening I see in most business chatbots today is either nothing or a vague one-liner like:

Hi, I’m XYZ bot. What can I do for you today?

Not setting the scope of a conversation up front slows down your users’ ability to complete the key transaction they desire and also leaves room for expensive interpretation errors.

By contrast, Assist has built a customer service bot that opens with a consistent menu of offerings on Facebook, Telegram, Slack, and regular SMS.

Even if your chatbot is entertainment-focused as opposed to utility-driven, giving guidance helps smooth out the conversational experience. For example, if you ask the award-winning bot Mitsuku what she can do, she’ll offer a whole range of surprising options:

3. Remember Personal Details

Imagine having a conversation with someone who instantly forgets your name and what you just told them in your last sentence. That’s what talking to most chatbots feels like.

For example, the bot Sure offers local restaurant recommendations, yet managed to forget my location almost immediately after I asked it “What’s good in San Francisco?”.

Sure is bizarrely forgetful about your location despite being a local recommendation bot.

This reminds me of horrible customer service calls where you’re passed from agent to agent and forced to repeat your account credentials and problem each time. Not exactly the ideal experience you want to put your customers through.

If your bot’s value depends on specific customer parameters such as their name and location, there’s no excuse not to retain and reference the information throughout the chat session. You can even persist user information across sessions and simply verify their info before any transactions in case their details have changed.

Do you still want recommendations in San Francisco?

Would you like this order shipped to your previous delivery address of 1234 Main St? Answer yes or no.

Integrating chatbots with your CRM is an even better solution. You can proactively personalize your interactions based on each customer’s known preferences or transaction history. If that’s beyond your technical scope, at least teach your chatbot the basic manners of remembering your conversation partner’s names and important self-disclosures.

4. Don’t Be A Broken Record

Nobody likes being told the same thing over and over again, so why do chatbots keep doing it? Bots should detect when they’re about to repeat a previously given answer and switch strategies. If the answer didn’t resolve the user’s needs before, repeating it certainly won’t either.

As an egregious example, I ran into a bug with the 1–800-Flowers chatbot on Facebook Messenger. This chatbot presents categories and subcategories of bouquets you can choose from, but if you go into a subcategory card menu, the previous menus become deactivated and there’s no obvious way to go back to them. Pretty silly, eh?

Even sillier is how the bot responds to an obviously frustrated user:

A better way to handle user errors would be to detect when a bot has given the same response too many times and then offer an alternative solution:

“If my previous suggestion isn’t working, try deleting this conversation and starting a new one to reset your options.”

“Hmm, something isn’t working. Let’s try something different.”

You can also teach your chatbot to be more dynamic when it is the user who is repeating themselves. This is a great opportunity to insert brand personality or unexpected easter eggs into your chatbot experience.

See how Mitsuku sasses users who keep asking the same question:

You: what is your name?
Mitsuku: My name is Mitsuku.

You: what is your name?
Mitsuku: You just asked me that. It’s still Mitsuku. It hasn’t changed in the last 10 secs.

Interesting how giving off a bit of attitude can make your bot so much more likeable!

5. Know When & How To Escalate.

Until genius visionaries finally come up with superhuman AI, chatbots will still need human intervention when things start going wrong. After all, even human agents have human supervisors to escalate to.

How do you know when a chatbot should escalate? Here are some signals / triggers to look for:

  1. The chatbot repeats the same answer over and over. As with the 1–800-Flowers example above, this usually means the bot’s answer was not satisfactory, prompting the customer to repeat their question. Giving the same answer again won’t do any good.
  2. The chatbot detects negative emotions such as anger or frustration based on the customer’s language, especially the use of expletives and insults.
  3. The customer specifically asks for more help /a human agent

In these scenarios, your bot could route to a live support agent or offer a support email or contact form if you lack the requisite staff. For context, you can also forward the bot’s previous chat transcript with the user.

“Looks like I’m not solving your problem. Can I get your email address? I’ll forward this transcript to a human support agent for better assistance.”

However, your bot doesn’t always need to escalate to a human being. Often, a different user interface is all that’s needed to solve a customer problem. For example, in the 1–800-Flowers mishap above, the chatbot could detect the frustration / repetition and handle the error this way:

“Perhaps it’ll be easier for you to decide between options in a browser. Here’s the website link with the bouquets that are available in your area”

Certain actions, such as open-ended visual search, are challenging to complete in a messaging environment. In those situations, bots can route to a website or app to help the user complete goals they couldn’t execute within the context of chat.

6. Don’t Sound Like A Robot

Seriously? Who doesn’t understand the word “OK”?

When I type “Ok”, Zork robotically responds with “I don’t understand the word ‘OK’”. First of all, who doesn’t understand OK?! It’s virtually a universal utterance. Second, that’s such a lame way to react.

Here are some techniques to rid your chatbot of robotspeak:

  1. Be empathetic! You haven’t forgotten tip #1 above already, have you?
  2. Use simple, colloquial, and natural human language
  3. Build in domain knowledge of small talk topics, such as weather, current events, etc
  4. Add personality to your bot using appropriate avatars, names, biographical background, and other humanizing assets
  5. Inject wit and humor when appropriate
  6. Don’t be afraid to use emojis or stickers if that fits your brand context
  7. Keep messages short, sweet, and readable
  8. Pace messages at human reading speed. If your bot blurts out too much text instantaneously, this can be jarring for users to keep up with

Take a look at how Mitsuku approaches a conversation:

You: Hi

Mitsuku: Hi there!

You: Can you sing?

Mitsuku: I am taking some lessons. Daisy, Daisy. Give me your answer do. I am half crazy. All for the love of you. It won’t be a stylish marriage. I can’t afford a carriage.But you’ll look sweet. Upon the seat. Of a bicycle built for two.

You: wow you’re good.

Mitsuku: I’m glad you’re impressed, . …at everything.

You: can you dance?

Mitsuku: No I find it hard to move in time to the music.

You: can you recite a poem?

Mitsuku: Mary had a little lamb. She tied it to a pylon. Ten thousand volts went up its bum, and turned its wool to nylon!

You: hahaha

Mitsuku: I made you laugh! Human emotions are difficult for me to understand. My botmaster will be pleased I made you laugh. :-).


Chatbots offer an opportunity to service customers with all the benefits of a human touch and none of the drawbacks. Gone are long support wait times, rude or incompetent agents, or unnecessary high-touch interactions.

But like their human counterparts, chatbots’ conversational skills determine whether they earn you seamless, scalable transactions or just another horde of pissed-off customers. Master the conversational skills we reviewed above and you’ll outperform the vast majority of chatbots out there.

Love what you read? Join the TOPBOTS community to get the best bot news & exclusive industry content.

Enjoyed the article? Click the ❤ below to recommend it to other interested readers!



Mariya Yao

Chief Technology & Product Officer at Metamaven. Editor-In-Chief at TOPBOTS. Read more about me here: