How to break a chatbot: 10 tricks
When you create a chatbot, you want them to be as efficient as possible for your customers. However, certain replies or nuances in human speech can throw the bot off track.
1) Ask to Reset or Start Over.
When a customer tells a chatbot to reset or start over, it can cause confusion. For the bot, it’s like hitting a dead end when it cannot comprehend the customer’s demand. In such situations, you might have to shut it down and start again from the beginning.
This can be extremely annoying for customers, especially when they need to restart even if they have already provided a lot of information to the bot. Thus, it is better to add a simple way to start a new conversation such as using a specific word or phrase to direct it to certain actions.
2) Using Filler Language.
When people converse, they often use filler noises such as “umm” or “ohh” and some people tend to use this in texts too.
If a customer puts any of the above instances of filler noise into the chat, the chatbot would probably reply with a “Let’s talk about that” or “Tell me more about this”. In such cases, the customer might have to explain what they meant to get the bot back on track.
3) Questions related to display buttons
The next example of “how to break a chatbot” is related to the display buttons on the screen. When you open a chatbot window, you might come across preset options for interaction.
A majority of the chatbots are trained to respond by initiating a set of questions. But when a customer chooses to type in one of these options, the bot is thrown off track even if the question is similar to the options given.
It might ask you to either choose from the options given or ask for clarification to understand how to handle your input.
4) Request for Assistance or Help
Interestingly, one way of “how to break a chatbot” is by asking for help or assistance. Many chatbots are not programmed to handle such requests effectively, as companies often don’t make it clear what you should type to get the chatbot to guide the conversation. You might try typing “help,” “agent,” or “what can I do” to initiate assistance.
If the chatbot is equipped to handle these requests, it will typically display a short menu of pre-set options. Selecting one of these options should direct you toward a solution to your original question.
5) Answering with non-traditional answers
Chatbots prefer unambiguous “yes” and “no” answers. This is how they are programmed, and failing to respond with these options may cause a chatbot to malfunction.
For example, if the chatbot asks whether you want to look at x product and you say “nope,” the chatbot will most likely respond with an error message.
Companies could consider adding variations to typical “yes” or “no” responses to make the bot replicate human speech patterns.
6) Greetings at the end
Once you’ve ended the chat, try saying something like “goodbye” and see how it responds. The majority of the bots will prompt for further clarification.
Most companies do not think of this possibility, leading to further confusion, Ideally, the chatbot should be able to comprehend such simple answers and respond with a greeting or ask if there is anything else it can do for you after you say “goodbye”.
7) Asking Odd Questions.
Depending on the chatbot’s intended usage, odd inquiries such as open-ended, hypothetical, or rhetorical questions can become questions that break AI systems.
This can be done when you ask questions that have nothing to do with the product or the service that the brand provides. Since the bot cannot understand the context, it might ask you for clarification or even inform you that it did not understand your query.
For example, ask a bot “I see the garden, do you?” and check what it says.
8) Ask Emotional Questions
If you’re wondering how to trick AI chatbot, you can simply ask “How are you feeling?” If there’s a human behind the screen, they will respond to this question in various ways.
Chatbots, on the other hand, are designed to respond to generic or frequently asked questions. Thus, such small talk or conversation holds no meaning for them in the same way that it does for people.
It might respond with a “I’m a machine, I do not have any feelings” which is the most accurate response.
9) Try Requesting for Better Phrasing
To understand “how to break an AI chatbot” you can simply keep asking it to rephrase its responses. When you frequently ask “What does this mean?” it exposes the chatbot’s limitations in responding outside of pre-set responses.
Additionally, this also highlights the need to review and improvise the chatbot’s scripts to ensure it shares the right information without jargon.
10) Asking for a Personal Story
Another trick for “how to break an AI chatbot” is by asking it a few personalized questions such as “How did you apply for this job?”
While humans can share personal experiences in detail, chatbots lack personal experiences and can’t provide such responses. Instead of trying to program responses for every possible personal question, it’s better to have the chatbot deflect these inquiries and bring the conversation back on track.
This authentication flow raises questions about its efficiency. Virtual assistants can process customer data, so it would be more streamlined for Nike’s bot to collect this information directly rather than through a form.
Additionally, requiring multiple pieces of customer data seems unnecessary, as a single piece of identification should suffice to provide the necessary assistance. Simplifying this process could enhance user experience and efficiency.
User Journeys of Nike’s Bot
1) Track My Order
In the “Track My Order” journey, Nike’s bot starts without a welcome message, immediately addressing order status inquiries.
The bot requests the customer’s name and email address again, despite this information being provided during authentication. This redundancy is inefficient; a single piece of identifying information, like an order number, should suffice.
If customers don’t remember their order number, the bot should guide them on finding it or offer another method to access the order. Instead, incorrect entries result in a fallback response and the original prompt repeating, causing confusion.
2) Ask About a Product
When users ask about a product, the bot asks if they are searching for a new or existing product, which is often unclear.
For inquiries about popular products, users are frequently transferred to a human representative instead of being provided with relevant information or directed to a product page. This shows the bot is not adequately trained to recognize product names.
Other Issues with Nike Chat Support
1) Queue and Typing Indicators
When users first join the chat, they are instructed not to refresh the page and informed they are in a queue, suggesting they are waiting for a human representative, not a bot.
2) Language Inconsistency
The language used throughout the interaction is formulaic and inconsistent, mixing formal and informal messages. This inconsistency is a missed opportunity for Nike to create a cohesive and engaging user experience.
3) Chat Timeout
If users pause for a moment, the chat times out after 90 seconds, which can create feelings of panic. Ideally, a bot should wait indefinitely for the user to respond, accommodating common distractions.
4) Outdated User Interface
The user interface appears outdated, resembling a design from 2005. Considering Nike’s prominence, a more modern, simple, and clean UI would enhance user satisfaction.
5) Task Switching Issues
If users want to switch tasks, such as asking about a product instead of tracking an order, they must exit and reenter the chat. This extra step slows down the user and complicates the process.
6) Main Menu Navigation
Users might think they can return to the main menu from a drop-down menu, but this option is hard to find and doesn’t function as expected. Instead of returning to the main menu, the bot prompts users to reconfirm their request, offering only the original prompt and an “other” option.
7) Saving Chat History
Users can save their chat history for future reference. However, the options to save the chat—via a .txt file or by printing—are unusual and could be more user-friendly.
Recommended: How chatbots can improve customer experience?
