With the development of Open AI’s ChatGPT, the discussion about a new role of ‘prompt engineering’ has started coming up more often. Prompt engineering is a way of communication with the large language model, a relationship between a human and machine to create personalised, fast, and informed answers.
If you want to become a prompt engineer, you must learn to give specific and smart directions to get the best solutions. In this article, we will explore what prompt engineering entails, what we know about it for now, and its potential impact on CX.
What is prompt engineering, and who is a prompt engineer?
A prompt engineer is responsible for crafting effective instructions or prompts to guide language models, like ChatGPT. Learning to give directions or prompts effectively involves understanding the capabilities and limitations of the language model. What does this mean?
Language models like ChatGPT pull out knowledge from a large database, but can’t always differentiate what information is correct and is not free from biases. This further means that the prompt engineer has to develop critical thinking and invest time in fact checking, rewriting, and editing answers.
We asked Milan Nikolic from Awards International who recently started actively using ChatGPT in his work to tell us what it means to have a virtual assistant:
‘I feel like I’m an editor asking an employee to write some content for me. The employee is great but doesn’t know my taste quite yet. However, these virtual assistants take direction really well and do their work really quickly and professionally.‘
Our observation is that people start developing a relationship with a virtual assistant. Interviewees for this article told us that they sometimes ‘argue’ with the ChatGPT if unhappy with the answer, some are always polite and show it by adding ‘please and thank you’ at the end of the requests they have.
People see generative language models as ‘their robots’, new colleagues, employees, or assistants with whom they should develop a relationship that works to their benefit. We expect more discussion about anthroposophical characteristics of robots in the future. The conclusion for now is that robots shape us, but we also shape them back.
How does prompt engineering impact CX?
In the context of CX, prompt engineering refers to a proactive and efficient approach of addressing customer queries, concerns, and feedback. It involves the strategic use of technology, data analysis, and streamlined processes to deliver swift responses and resolutions to customers. It aims to reduce customer wait times, and enhance satisfaction.
To execute prompt engineering successfully, organisations must have a robust infrastructure in place. This includes:
- adopting modern communication tools
- employing AI-driven chatbots
- creating knowledge bases
- empowering customer service representatives with the right resources to resolve issues quickly
In one of our CXM guides, we wrote about how generative AI can impact the development of video feedback. We predict that prompt engineering might impact CX by addressing two fundamental aspects: response time and problem resolution.
a) Faster response time. Prompt engineering enables businesses to reply to queries, provide answers, and create alternative and customised offers to their users. For this, employees will have to be trained to understand the business context, search functions, and how to write good prompts.
b) Efficient problem resolution. Prompt engineering streamlines the customer support process, enabling agents to access relevant information swiftly and provide accurate solutions. For this to be true, companies first have to train large language models to understand and analyse huge amounts of unstructured data. A lot of manual work and labelling has to be done by developers and other employees. It will be on them to contextualise data and make sure to avoid biases.
Content creation is where we see prompt engineering being applied the most
Prompt engineering is mostly used for content creation purposes – answering emails, writing copies, and articles, brainstorm efficient titles, etc.
Here’s how content creators can use prompts effectively:
- Topic ideation. Content development experts can use prompts to brainstorm potential topics and themes for upcoming issues. They can create a list of thought-provoking questions or statements related to current trends, events, or issues. However, language models can’t generate ideas based on real life experiences, encounters, and often ideas suggested won’t go beyond obvious observations. Therefore, human factor absolutely has to stay in equation.
- Content guidelines. Editors can provide specific prompts or guidelines to writers, outlining the type of content they are looking for. For example, a prompt might request a personal narrative, a research-based article, or a listicle on a particular subject. The challenge with this is that editors need at least one person to edit content and fact check everything that is generated by a language model.
- Provoking titles. One of the ways brands can use language models to engage their audience better is through their weekly or daily newsletters. To increase open and click rate, they can play with prompts until they are happy with the results. This works for article titles and any content brands want to be engaging.
- UX writing. As mentioned earlier, we talked to Milan Nikolic who used ChatGPT to write copy for his company website. The language model did great work informing users about the company programme. But the authenticity of the content had to come from the human writer. To give good prompts, one has to understand what knowledge the UX writing role contains. Especially related to usability, accessibility, diversity, targeted audience and their preferences.
Milan explains that the more precise prompt you give, the better results you get. Here is an example of a prompt:
‘I want to write content for the home page of my clothing store. It has high-end clothes quality made from locally sourced material, ethically made, ships anywhere, but most importantly affordable, and multiple colours and designs. It should be up to 500 words, the tone should be exciting while telling a story. ‘
What you get as an answer is up to your creativity and further guidelines you will give to your chatbot assistant.
A look to the future
ChatGPT is far from being the last big generative AI model to gain this many users, attention, or prompts. We’re now in the age of AI assistance, so we must adapt, and learn how to use them effectively.
Prompt engineering will become a key skill in this. Perhaps it will become part of future education curriculums, job application requirements, qualifications, etc. And as the HBR School professor Karim Lahkan said in a recent interview ‘AI won’t replace humans. But humans with AI will replace humans without AI’. If this is true, we should be ready just in case.
No matter the reason, learning how to successfully engineer an AI prompt for best use is something we should all be looking into.