How to use ChatGPT for customer service

What is ChatGPT?

ChatGPT is a chatbot that can reply to a wide range of questions and prompts with human-like responses. OpenAI developed the tool and released it to the public for free on Nov. 30, 2022.

The chatbot sits atop OpenAI's GPT-3.5 LLM -- a type of generative AI technology that its developers trained on billions of pages of text from the internet. Developers stopped the tool's training in September 2021, so it cannot answer questions about events that took place after that date.

Unlike simpler chatbots, ChatGPT can understand language in context, remember past conversations and quickly generate creative content like poems, short stories, essays, articles and email responses. Many people use the free tool for entertainment purposes, but business professionals can also use it to automate work tasks.

How can ChatGPT improve customer service?

Contact center managers can use ChatGPT to reply to customer complaints and reviews, enhance customer-facing chatbots, summarize and translate inquiries and create virtual assistants.

1. Reply to customer complaints and reviews

As service agents engage with customers on various channels, such as email, social media and product review websites, ChatGPT can help them quickly formulate responses to complaints and reviews.

For example, a service agent can ask ChatGPT to write an email response to an angry customer, and the chatbot will do it -- typically with the professional and empathetic tone that organizations expect their agents to use.

ChatGPT also lets users dictate the length of responses, which can help them craft short-form responses for social media and product review comments. For instance, a user could ask ChatGPT to write them a response to a negative customer review in 500 characters or fewer.

Despite ChatGPT's ability to generate human-like responses, users still need to edit, personalize and fact-check responses. Developers trained the tool on internet data and, therefore, it contains biases and inaccuracies that can be found online.

Additionally, this tool doesn't have access to organizations' specific business policies, so users must ensure responses don't promise discounts or other forms of compensation their organizations cannot offer.

2. Enhance customer-facing chatbots

Since the early 2000s, customer service departments have used customer-facing chatbots to help answer frequently asked questions, lighten agents' workloads and reduce hiring costs. However, most have been rule-based chatbots, which offer preprogrammed responses to limited sets of keywords and phrases. These chatbots function more like a Google search than a true conversation, and natural language easily confuses them.

Conversely, ChatGPT and other generative AI chatbots use advanced LLMs to engage in realistic dialogue. Although organizations cannot use ChatGPT itself as a customer-facing chatbot, OpenAI offers an API for its GPT-4 LLM that an organization can use to train the model on its knowledge base. Organizations can then incorporate this customized LLM into their customer service chatbots, which can then offer human-like interactions and answer questions specific to the business.

3. Summarize customer inquiries

Quality customer service requires agents to understand customers' problems and frustrations. Agents must read complaints and review past interactions between customers and the contact center, so they don't have to repeat themselves. These actions can enable better customer support, but they can also take considerable time for agents.

To help agents more quickly pinpoint customer problems and understand their perspectives, they can use ChatGPT to summarize inquiries and past interactions. For example, an agent could copy and paste a complaint email into ChatGPT and ask it to summarize the main points in a few short sentences.

Agents must remove any personally identifiable information -- which includes customer names, addresses, phone numbers and email addresses -- from the content they enter in the chat. ChatGPT stores a transcript of all user interactions, which cybersecurity professionals fear hackers may find ways to access.

4. Translate inquiries and responses

Many organizations serve customers around the world and therefore must offer services in various languages. Contact center agents can use ChatGPT to translate content, such as emails and social media comments, into over 50 languages.

Yet, misspellings and colloquial language can confuse ChatGPT and lead to mistranslations. To prevent miscommunication, contact centers can use ChatGPT in combination with multilingual staff. For example, an agent that can't speak Russian could use ChatGPT to translate an inquiry from a Russian-speaking customer. However, if the translation doesn't seem to make sense, the agent could transfer the customer to an agent proficient in Russian.

5. Create virtual assistants

Customer service departments can use generative AI technologies like ChatGPT to power virtual assistants for their employees. For instance, organizations can use APIs to train GPT-4 on their knowledge bases and integrate it with existing tools, such as contact center platforms and scheduling systems.

As agents interact with customers across channels, a generative AI virtual assistant could help in the following ways:

• Suggest detailed responses to inquiries in real time that agents can use, edit or ignore.
• Automatically summarize lengthy inquiries and past customer interactions.
• Sort customer inquiries into categories such as sales questions, cancellation requests or technical issues.
• Use sentiment analysis to understand customers' emotions and route them to the most appropriate channel and agent.
• Automatically schedule customer appointments.

A custom virtual assistant offers many of the same benefits as the free ChatGPT tool, such as email generation and summarization. However, these assistants can create more detailed answers because organizations can train them on their knowledge bases. They also offer more convenience because organizations can use APIs to embed their features directly into the tools agents work with every day.

ChatGPT's public release has sparked a lot of hype around generative AI. Many technology vendors, such as Microsoft and Salesforce, have since announced partnerships with OpenAI, whereas others have built or plan to build their own generative AI tools.

Contact center managers can experiment with ChatGPT for free, but they may want to wait before investing in a paid generative AI tool. Generative AI is still new, so products may need many more years of fine-tuning before they become truly effective for customer service.