Why the best chatbots don't pretend to be human

This happens 4 times out of 5.

Almost every time, when I hop on a call with a potential client early on in their chatbot discovery phase to plan their build, we end up talking about the age old topic:

Should my chatbot pretend to be human?

In fact, more often than not, it comes up in the form of "Oh no, I think my chatbot shouldn't say it's a bot".

With this article, I hope to put this rookie (but understandable from a client's point of view) chatbot building mistake to bed.

 

chatbot-building-mistake-human.jpg

 

No, your chatbot is not human

Let's start with the obvious: your chatbot is... a chatbot.

It's not human. It doesn't act or react like a human. It has been programmed to process specific queries and answer them with pre-set responses.

It is a bunch of 1's and 0's put together by a bunch of geeks.

I know this is extremely obvious to everyone involved but, trust me, it is sometimes lost in the excitement.

 

You only have one chance

For me, the most important reason your chatbot should not pretend to be human is because:

You only have one chance to engage with your chatbot users.

If you decide to pretend your chatbot is a human, at the first sign of failure or weirdness, you lose the user.

It is as simple as that. As soon as your chatbot messes up and your user realises it has been duped; poof. Gone.

 

Deception is a rookie chatbot building mistake that will hurt your brand

Not only is deceiving your users bad for retention (because, you know, poof), it is also bad for your brand.

Imagine the PR nightmare of having to deal with duped users talking on social media to lay out the embarrassing experience they've just had of trying to speak to a machine.

Remember when people used to make their answer machine recording start with saying "Hello?".

Super funny.

If you've ever been on the receiving end of that 'joke', do you remember how foolish (and maybe angry) you felt when you realised you had been caught out?

This is how your chatbot pretending to be a human would make your users (your customers!) feel.

This little trick only works in The IT Crowd.

 

 

 

Your users will understand

Amazing things happen when you are upfront and honest.

In this particular case, users tend to be more understanding and accepting of a chatbot 'messing up' (not understanding the intent, answering the wrong question, etc.) when it is upfront.

Not upfront? People get super angry really fast.

Upfront? People are more understanding.

Chatbots are still new technology for 99.9% of the population. No one expects your chatbot to be perfect and answer everything like a human can.

In another article, I talked about our "three strikes and you're out rule", which is a perfect example of this.

After three times that the chatbot does not understand its user, it very openly says it (e.g. "Sorry, my chatbot abilities only take me so far. Not sure I understand what you are trying to tell me!") and offers a series of commands he or she can use that work.

 

You need to contain interactions

One of the initial phases of building a chatbot is figuring out its goal. What will it do?

Once we figure this out, we can start defining what it is going to talk about. Which topics should it be able to answer? Defining topics is important because they help contain interactions.

Let's say you are planning on releasing a chatbot that helps you choose wine. The goal is to help users figure out which type of wine they should buy to go along with tonight's meal.

The topics here could be:

  • wines (being able to discuss wines around the world, major brands/crus),
  • taste (being able to suggest wines based on the likes/dislikes of the user), and,
  • food association (being able to pick a wine based on the ingredients of a dish).

Now let's compare user experiences.

 

Winebot pretends to be human

User: Get started
Winebot: Hi, I'm Jodie, how can I help you?
User: I need a wine for tonight.
Winebot: Great, that's what I'm here for! What are you having?
User: Well, I'm not sure yet but I am going to my cousin's house and he tends to love red meat so potentially that. Although, I did mention that my daughter is coming along and she's vegetarian, so I would need something that suits both.
Winebot: I'm not sure I understand.
User: What? I need a wine that goes with red meat (lamb I think?) but also with a vegetarian dish like humus.
Winebot: I'm not sure I understand.
User: *quits*

 

 

Winebot is honest about being a bot

User: Get started
Winebot: Hello there! I'm Winebot, nice to meet you. My goal is to help you find the best wine to go with your meal. I can help you if you tell me the ingredients you are having.
User: Red meat
Winebot: What sort of red meat will you have?
User: Lamb
Winebot: Great. With lamb, I would recommend a young Bordeaux or Merlot. Would you like me to send you some options to review?
User: We may also have hummus, not sure, my daughter is coming and she is vegetarian.
Winebot: I'm not sure I understand. I am just a bot after all! Tell me the ingredients are you having with your meal and I will help you.
User: Hummus
Winebot: With hummus, I would recommend a sauvignon blanc or a pinot noir. Would you like me to send you some options to review?

 

By being upfront about its limitations in understanding, the chatbot is able to contain the conversation and even bring users back to them main goal without making them feel silly.

From the user's point of view, it is expected that a chatbot will not understand a long-winded paragraph about the tribulations of a dinner party and guests with differing dietary requirements.

 

Why does all of this matter? Build your chatbot for retention

Building a chatbot is only half the battle. Once you release it into the wild, you need to attract users, engage them, and deliver value.

If done properly, you should not experience any of the terrible retention some chatbot building platforms receive.

As enticing as it is to make it seem like you've built the best chatbot in the world, once it (inevitably) fails to deliver, you'll lose all credibility.