There are two ways to work with this book: follow along and learn the principles and practices and use one of the OpenAI playgrounds, which is typically a no-code approach, or use the APIs provided by ChatGPT.

We have examples in our book’s GitHub repository. If this is your first time using GitHub, it is a place where we will store any materials needed to download to complete the examples in the book. It is an online folder of resources.

GitHub: Repository for book materials (https://github.com/PacktPublishing/UX-for-Enterprise-ChatGPT-Solutions/)

GitHub is the repository for all files from the book. Click the download button, highlighted in Figure 1.1, to download the file to the desktop. There is no viewer for most of the files on the GitHub repository.

Figure 1.1 – How to download a file from GitHub

Make sure you have a ChatGPT account.

Website: OpenAI Chat (https://chat.openai.com/)

That is easy...

Interaction design intersected with AI well before the LLM revolution. This history is helpful to appreciate when applying design principles to the latest conversational experiences. Any discussion of AI at least mentions Alan Turing and the question posed by his article in Mind (a peer-reviewed academic journal).

Article: Computing Machinery and Intelligence (1950) (https://redirect.cs.umbc.edu/courses/471/papers/turing.pdf)

This is routinely referenced as the Turing Test. The ability of a machine to seem intelligent and be indistinguishable from a human.

Article: Wikipedia on the Turing Test (https://en.wikipedia.org/wiki/Turing_test)

When this was published in 1950, we were still far from a computer being indistinguishable from a human, at least in a text-only interaction. We must skip ahead to the mid-1960s before we see something that appears to engage in discourse.

If we try out one of ELIZA’s conversational interfaces...

Does ChatGPT even need an introduction at this point? The innovative model developed by OpenAI is in a new class of LLMs, which are trained on billions of data elements from the internet’s vast supply of articles, books, and knowledge. It achieved over 100 million users in about two months. It can generate human-like conversational interactions in text or voice in many languages and converse on vast information. And it does it pretty darn fast.

ChatGPT has undoubtedly come on like a firestorm. Unique, fun, fast, intelligent? However, when designing solutions for your business or enterprise, they should be accurate, have the most current business-related information, and communicate to customers in the voice, style, and tone expected from the business. So, how do you take such a fast-paced moving target and wrap it into a product that exceeds customer’s expectations? Can one ensure that it doesn’t give random answers that...

Every coin has two sides (okay, plus the edge!). Typically, we see two sides to UX design. Those with visual backgrounds and those who come from a science perspective. Schools are now overwhelmingly delivering visual and graphic artists to meet demand, and with conversational AI, there is some art to the experience but only sometimes visual elements. The introduction of generative AI impacts every facet of interaction design. The design roles will adapt or die. Adaption is the better option. As graphical user interfaces (GUIs) adapt to include conversational elements, the role of a visual designer will still be relevant, even if only to create the correct prompts to help them generate the look and feel that aligns with the organization’s goals. The side of the equation for the science of design remains vital to requirements, understanding, and communication. Even when writing this book, ChatGPT provides some good answers related...

So now we have a sense of the people and tools we need. One more piece of the puzzle is the content that will make a ChatGPT solution valuable. There are many ways to include content in an LLM, such as ChatGPT. However, we assume a focus on unique, proprietary content hidden behind a secure paywall or available only to authenticated users. If company answers are already out in the world and answered by the basic ChatGPT, ask yourself, How would my custom version of ChatGPT provide value? We can help get that answer by focusing our content discussions on building a private model and including company data without sharing it with the world.

The idea of enterprise assistants is precisely that. Ensure company data is only exposed to customers and that adding it to an instance of ChatGPT does not expose the data to the world. Only some people were this careful when ChatGPT was first introduced. Don’t add sensitive data to a model used by competitors...

At this point, this only scratches the surface. We will go into how to develop a user research plan and apply standard research methods to the conversational world, then dive deep into user interface patterns, guidelines, and heuristics to help guide and improve the quality of an enterprise ChatGPT solution. Once we have something, it is time to apply our care and feeding approaches to monitor and improve the quality of customer experiences, extending it further into more customer use cases with the quality they expect. We will wrap it all up in a nice bow so this newfound expertise can be shared with your friends in development and product management. No person is an island. It does take a team.

Next, let’s cover basic research and tools we can use to understand our users and how we can learn what our customers need. A few suitable research methods can go a long way to bridging the gap for unknown unknowns…

“Reports that say that something hasn&...