ChatGPT for Customer Service: Limitations, Capabilities, and Prompts

Customer service can stand to gain from AI phenomena like ChatGPT. However, your business’ approach to ChatGPT, as with all new technologies, must be calculated. There are those who still underestimate AI’s capabilities, who stand to lose by eschewing it entirely. There are those who overestimate AI’s capabilities, who stand to lose by rushing in prematurely. 

Only 50% of companies say their teams are well-versed in AI.

Customer acquisition and retention are on the line since customer expectations will quickly outpace those slow on the uptake. And we know that missed customer opportunities are critical, especially during a recession. 

In the first five days of its launch, ChatGPT reached 1 million users. It is widely understood that we are on the precipice of even faster evolving AI functionalities – and every business and leader should be questioning how they can seize this opportunity to their advantage. This is not about striking the iron while it is hot – more about keeping up with the curve.

Stick with us the whole way to also discover unique Chat GPT prompts for customer service leaders, as well as a way to completely reinvigorate your CSAT responses. 

Before you wade in to experiment, look in the mirror to reflect on what problems you might want to solve. Here are some of the most common challenges for customer service teams:

• Meeting customers expectations
• Improving quality at scale
• Maintaining consistency across teams
• Keeping up with technologies.

Now, let’s talk about ChatGPT for customer service and whether it might help with these issues.

What is ChatGPT?

ChatGPT is a chatbot developed by OpenAI that uses a variant of the GPT (Generative Pre-training Transformer) language model to generate human-like text. It is designed to engage in conversation with humans and answer their questions or requests.

For example, it coughed up the above definition of itself (how meta) during research. But language generators like ChatGPT don’t (yet) contain enough nuance to think of and describe their actions as meta.

The system is trained using machine learning techniques, using a large dataset of text from the internet. Ease of use is one of its pre-eminent advantages. 

It was made public on November 30, 2022, and is currently in the research review phase for users to test it free of charge. The chatbot itself is free to try for individual users, hosted by OpenAI, but the company also sells access to the AI models to businesses for use.

AI for customer service is already multifaceted, but does this dawn a new era for how we speak to customers?

First, let’s look at the capabilities ChatGPT for customer service

You might be wondering how this is any different from existing chatbot services on the market. The above four benefits are all selling points for the chatbots that have become standard for answering basic customer inquiries. 

ChatGPT is far more advanced than its chatbot predecessors. 

Chatbots have always aimed to mimic human conversation as closely as possible through AI. They could answer frequently asked questions and provide set information about products and services. However, it could not engage in any more complex two-way dialogues. As you may have experienced, users had to be exact in their questioning to avoid confusing the bot. This can be frustrating for customers who have more complex queries.

The new GPT variant is much more proficient at simulating human language and is able to respond to more natural user input. It also has an extensive knowledge base and is able to recall previous conversation points and even call out a person for lying.

Businesses thus have a path to more natural conversations with customers or clients through ChatGPT.

Um, but there are limitations of ChatGPT for customer service in this format

Let’s dissect these disadvantages.

Filling in knowledge gaps

ChatGPT is good at solving problems and skillfully navigates conversational discussions. Just like many humans. But where ChatGPT’s knowledge falls short, it does not have the awareness to admit this and instead fills in the gaps. Just like some people would during a discussion. 

But a discussion, or conversation, is not quite what you would describe the interaction between a customer service agent and customer. This is more transactional – the setting entails accurate information exchange. 

Thus, if a query becomes too complex for ChatGPT, it will fall back on its powers of making conversation rather than its powers of knowledge. Customer service agents should never try to fill in gaps in their knowledge in the context of their job.

The need for deliberate, articulated prompts

There are several reasons why a customer may not want to tailor their line of inquiry (to get the best answer from ChatGPT, or any chatbot):

1. The customer may be experiencing frustration or stress due to a problem or issue they are trying to resolve. They may not have the patience or inclination to carefully craft their questions in order to get the best answer.
2. The customer may feel that the answers they need already exist and that it is the responsibility of ChatGPT or the customer service representative to provide them – rather than having to spend time and effort trying to extract the information they need through careful questioning.

This is especially true if the customer is unable to assess the accuracy of the chatbot’s responses.

ChatGPT’s inability to provide personal information

There is also the issue of personalization, which is a priority, considering that 80% of consumers are more likely to buy from a company that provides a tailored experience. In the past, automation has come at the expense of personalization. And although customer service chatbot advancements have come in leaps and bounds from first generation bots, it cannot give customer-specific data. 

If a customer wants to know account information, for example, ChatGPT falls short.  

The issue of putting a customer in front of a ChatGPT-powered bot is that you are asking too much of a customer and not giving enough in return. If a customer wants to put in the effort to find the answer themselves, they will search your knowledge base, or Reddit, or YouTube. When they come to a chat, they want a direct answer and have likely already exhausted the more proactive, self-serve means of support. At this point, they desire a ‘real’ customer service agent. 

How Generative AI can help you improve CSAT response

Customer feedback is a treasure trove of (often untapped) potential, the benefits of which extend beyond the confines of the support department. Using the voice of the customer to influence strategy, product, marketing, etc., will help your company drive loyalty among existing customers and attract new ones.

A customer satisfaction survey is one of the easiest ways to extract this feedback. But CSAT response rates across the board are uninspiringly low. 

You want a higher response rate – yet, you also need comments from which you can derive context for their sentiment. 

What does this have to do with Chat GPT? 

Well, this is where we start tooting our own horn. We’ve built our own large language model to help companies make it easier for customers to provide feedback. 

KlausGPT works twofold to boost your CSAT response rate:

• Turning on the function gives your customers an automatic conversation summary of their support interaction. Many support issues take time to solve – succinctly summarizing their communication to jog a customer’s memory and help them remember precisely what they are rating.  
• It also suggests personalized comment snippets, helping them construct feedback to give you context to their score.

One of the most common problems plaguing almost all surveys is low response rates and the need for free-form human feedback. But when a customer leaves a review and comments to provide context to their sentiment, it’s often golden information for your support team (and potentially other departments).

By adding Klaus GPT, we’ve acknowledged and removed these issues to make CSAT surveys 10x better.

Mart Objartel

Klaus

✨ Find out more about this pawsome feature✨

7 Customer Service Prompts for ChatGPT

ChatGPTs strengths lie in its ability to mimic human conversation when you feed it prompts. People leap to question whether it can serve as a proxy for customer service agents and jump-frog its other uses for customer service.

Below are several more ‘behind-the-scenes’ ChatGPT prompts to help customer service leaders manage their support teams. 

These more creative applications help you save time through training teams, bolstering proactive support, and unifying communications. All of these things will help with the afore discussed customer service challenges: 

• Meeting customers expectations
• Improving quality at scale
• Maintaining consistency across teams 
• Keeping up with technologies.

1. Prompt to provide examples for training soft skills

This gives agents go-to phrases and helps them work around different ways to convey the same sentiment.

2. Prompts to help you easily use code to automate tasks 

3. Prompt for creating templates to answer common customer queries

You can feed any relevant product information and policies into each individual prompt.

4. Prompt to help new agents adapt their tone of voice

Some businesses are more colloquial, some more formal, some use lots of cat puns (meow, we’re Klaus).

5. Prompt to act as a proxy for customer chats to test agents

6. Prompt to repurpose knowledge base articles

Reduce the time spent on common questions by helping agents make the best use of existing material.

7. Prompt to create knowledge base articles

Reduce the time spent on creating a help center database to help customers self-serve.

What’s next? 

When we look at artificial intelligence as a whole, its functions are to augment, perfect, and accelerate how we work as humans. 

In all of the recommended uses for ChatGPT, as with almost all AI implementation, there still needs to be a human in the loop – fact checkers, documentation managers, customer service trainers. And of course customer service agents themselves. 

Top-line customer support will, for the foreseeable future, entail human-to-human interaction. Customers still expect that, for their most complex inquiries, there will be a human to talk to somewhere down the support path. So, the customer service industry is not going to change overnight.

The increasing capabilities of machine learning, natural language processing, and language models will likely lead to the development of more advanced and accessible AI tools for businesses of all sizes. As demonstrated by ChatGPT.