When Bots Go Bad: Common UX Mistakes In Chatbot Design

Mariya Yao
Published in
6 min readJul 28, 2016


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

Bots will one day sweet-talk their way into our good graces, but that day is not today.

Judging from recent debacles such as Microsoft’s Tay and the fembots of Ashley Madison, AI-assisted chatbots still have a long way to go before gaining genuine and socially acceptable conversational skills. Just visit YouTube and you’ll realize that giving bots an eerily human-like appearance is much easier than granting them the gift of gab.

That means you’ll have to wait just a bit longer before a real-life version of either Jarvis (The Avengers) or Samantha (Hers) will arrive to stir your intellect or your emotions.

The Rise of Chatbots

Nonetheless, chatbots have certainly made huge strides since they were first used decades ago. You’ll find them in many websites performing a variety of tasks — from giving customer support on Slack to flirting with users on Tinder. One award-winning chatbot named Mitsuku even keeps lonely people company 24 hours of the day with her surprising wit.

Many businesses realize the game-changing potential of chatbots, with some experts predicting that chatbots will dislodge apps in terms of ubiquity, usage depth, and importance in our everyday life.

They might be right. Major technology players including Google, Microsoft, Amazon and Apple have already placed huge bets on AI, leveraging big data and machine learning to get as close to human intelligence as possible. For many of these projects, chatbots serve as bridges between the AI’s algorithmic core and the people trying to communicate with it. Wherever a meaningful dialogue can occur, so will business transactions.

That is why demand for conversational interfaces is on the uptrend, propped by strong use cases. Aside from technology leaders, large companies from sectors such as finance (Royal Bank of Scotland), toy manufacturing (Mattel), food (Domino’s), media (Disney) and the automotive industry (Renault) now actively use chatbots to initiate dialogues and spur conversations with customers.

When Bots Do Well

The benefits of effective chatbots are easy to see:


A single chatbot such as Mitsuku can handle thousands of conversations at the same time. Such a bot can perform many of the tasks customer service and marketing teams usually perform but at a much lower cost, enabling a small firm to significantly bolster its profitability.


Chatbots promise unprecedented levels of convenience as seen in how consumers use the Assist chatbot to organize multiple apps to make everyday stuff — such as calling a ride, booking hotel accommodations, dining out, and sending gifts — very easy to do.


Chatbots prevent customer engagements from going stale, idle or unproductive by initiating dialogue and providing targeted options for consumers.


Chatbots can be programmed to build a unique and organic personal profile for each customer over time, enabling companies to deliver real “personalized” service.


Applications are endless. Chatbots have been known to function as lawyers, doctors, personal stylists, concierges, finance advisers, fitness trainers, teachers, tech support, pets, and even romantic partners.


Chatting is second nature to us since we primarily interact with each other through conversation. This makes our use of chatbots much more intuitive and easy than clicking on a bunch of buttons with a mouse in traditional user interfaces. Additionally, millennials and teens — who represent the bulk of tomorrow’s market — spend more time on messaging apps than on social media sites, creating a huge opportunity for businesses who want to reach them on those platforms.

When Bots Go Bad

While chatbots show promise, they can also inflict pain. As we’ve see in recent bot blunders, not all chatbots are created equal. A poorly designed chatbot can easily turn a potential customer engagement into a horrible user experience.

For example, some chatbots say exactly the same things over and over because of very limited vocabulary and response iterations in their database. Others can’t answer even the simplest questions because their designers overlooked the fact that humans ask stupid, unpredictable or silly questions all the time. Worse, some chatbots even fail at their own domain by misunderstanding what customers are actually asking for.

To prevent an epic fail from happening, here are some usability issues you should be aware of before exposing a chatbot or any assisted AI to your customers:

Trying to be everything to everybody all the time

Don’t try to design your chatbot to do everything people might want it to do. Cortana, Siri and Alexa might eventually develop that capability, but it’s better to deploy a specialized, purpose-driven bot to engage your target audience.

Solution: Set clear goals and identify the use cases for your chatbot. Don’t attempt to address problems beyond your scope. Instead, focus on achieving domain mastery and manage customer expectations by keeping the conversation within your comfort zone.

Poor or nonexistent escalation protocol

Most chatbots used in business still need human supervision and intervention to generate the best engagement outcomes. Never deploy a chatbot without establishing an escalation channel through which it can route customer issues it cannot adequately solve to humans who are trained to handle such issues.

Solution: Map out primary engagement paths so that your bot knows exactly what to do in common scenarios. Branch out these paths as your chatbot experiences new scenarios and collects data such as new queries. Build your chatbot’s knowledge base over time such that the need for escalation diminishes, giving humans in your team more time to optimize value-laden tasks.

Limited linguistic and natural language learning capabilities

This major flaw can take the form of many obvious shortcomings such as —

  1. Inability to answer unique, complex, personal, contextual, or unusually phrased questions
  2. Overuse of emojis and colloquial expressions just to appear human
  3. Poor response to linguistic elements such as onomatopoeia, sarcasm, humor, or insult.

For example, an early version of the weather chatbot Poncho struggled to provide precise weather information due to a limited understanding of natural language and poor ability to parse human statements that don’t match its language library.

Solution: Start by managing user expectations. If necessary, let your bot admit that it is not Webster and gracefully reframe the conversation within its linguistic zone. At the same time, bolster your chatbot’s language learning skills by organically building its language database so that it will always communicate better in the next session.

Mismatched or nonexistent chatbot personality

Only the nerdiest of nerds will enjoy talking to a robot that actually sounds like a robot. That’s why major tech players go great lengths trying to humanize their chatbots and establish a “real” connection with their audience. However, merely using emojis, slang and colloquial expressions won’t work.

Solution: Be sure to design a chatbot whose visual elements (icon/avatar), vocabulary, tone, and overall “personality” resonates with your target customers. After all, your chatbot serves as brand ambassador as much as the humans who create and sell your products/services.

Lack of integration with existing business systems.

A chatbot who doesn’t have a clue about the company they represent or the people they chat with is a serious liability. Customers value their time. Just like with human customer support associates, customers will likely lose their patience if your chatbot repeatedly ask for account credentials or isn’t aware of their purchase or interaction history with the company.

Solution: Enable your chatbot to access and optimize relevant customer database, biographical and contact information of key persons in your company, media marketing kits, brand assets, CRMs and other knowledge repositories that will help generate more value and reduce unnecessary steps each time they engage customers.

Lax security or privacy safeguards.

Twitter users easily corrupted Microsoft’s Tay into becoming a Nazi-loving, misogynist chatbot. Because your chatbot represents your brand, you wouldn’t want a similar thing to happen. Moreover, if you have already integrated your chatbot into your CRM, internal servers, or other critical systems, a breach can be devastating.

Solution: Adopt security best practices and ensure that your chatbot is adequately guarded from external attacks.

Join the conversation

Chatbot interactions can amuse customers into buying your product and loving your brand. Or they can piss them off the same way shitty customer support does.

As the full benefits of chatbots become more proven, most companies will deploy their own AI-assisted bots to augment their customer service, marketing, sales, product development and social media teams. Remember that user experience remains the key factor that will contribute to the success (or failure) of your chatbot strategy.

This article was originally published on TOPBOTS.

Join the TOPBOTS community to get the best bot news & exclusive industry content.



Mariya Yao

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