I posted a blog post about thinking about UX for AI (mainly LLM) in the past, and I actually conducted a study with those who contacted me. The study was conducted a total of 5 times, and there were a total of 4 participants. I think it would be good to increase the scale a bit next time, but for now, I will talk about what I organized.
I have many shortcomings, but when it comes to planning, I divide it into 5 major stages. Since it's a crappy philosophy, any objections are welcome, and if you have a good methodology, please let me know. When planning, I proceed considering these stages, just like the customer journey that many people do. Stage 1: Intent (understanding user intent and goals) → Stage 2: Wayfinding (user understanding of the product) → Stage 3: Using (using the product/service) → Stage 4: Feedback (user feedback) → Stage 5: Retention (user retention and churn prevention). I will try to organize each stage.
I have covered this before on my personal blog. If we think of it as replacing it with artificial intelligence, we can organize it as follows. Most of the services that are currently available can be classified into the categories below.
Focus: You want a specific outcome and have a clear plan for getting there (e.g., editing to change the tone or format of your content).
Navigation: A situation where you have a goal, but are not sure how to get there (e.g., you want to find the answer to a specific question, but do not know how to enter the information).
Synthesizing: The act of deriving conclusions or insights from a known data set (e.g., finding trends or insights from multiple data sets).
Browsing: A creative experimental mode where you try out different inputs and explore the results (e.g., try out different tokens in image generation to see what happens).
Step 2: Wayfinding (User Product Understanding)
When a user first encounters a product, they need to observe and understand how to use it. This process is more heuristic than you might think and moves reflexively, so there are cases where various psychological effects are used to read the user's psychology and hit them first or provide them with friendly guidance. Psychological effects frequently used in UX can be found in the document below.
Guide method
It is not easy to write a perfect prompt from the beginning when interacting with AI. The 'Follow-up prompting' technique is gaining attention as a solution to this. This method is that AI does not respond immediately to the user's initial request, but goes through an intermediate step of asking for additional information.
1.
Simplified user experience : Break complex prompts into multiple steps to reduce user burden.
2.
Build trust : Instill the perception that AI is trying to accurately understand the user’s intent.
3.
Empower users : Empower users to actively participate in the process by requesting additional information.
Merit
Improved accessibility : Easier to use for users with limited AI experience.
Improve accuracy : Get more accurate results by checking details.
Increased user satisfaction : Transparent processes increase trust and satisfaction.
Application Cases
Advanced techniques such as Retrieval-Augmented Generation (RAG) can be used to further improve the user experience. These techniques allow users to provide direct references or parameters to the AI instead of simple text input, saving time and effort. This is the most common and easy-to-use method.
The service that uses this best is Perplexity.ai . It extracts tail questions or related questions and provides a good user experience.
Nudge method
AI tools offer a variety of features, but many users only utilize a few of them. 'Nudges' is a strategy that naturally introduces users to the hidden features of AI and encourages more effective use of AI.
1.
Introducing features gradually : It’s difficult for users to learn all the features of an AI tool at once. Nudges introduce features at the right time, in the right context.
2.
Context-based recommendations : Understand the context of the user’s current task and suggest AI features that fit that context.
3.
Improve user experience : Let us know when AI can take over or assist with repetitive or complex tasks.
Merit
1.
Assist with achieving goals : Introduce features that directly help users with the task they are currently performing. Example: Recommend automatic summary feature while writing a document
2.
Grow into an advanced user : Increase your users’ ability to leverage AI by gradually introducing complex features. Example: When analyzing data, AI shows the process of creating a pivot table and guides the user to practice it themselves.
Precautions when applying
1.
Avoid information overload : Instead of introducing all features at once, provide them in stages based on the user's needs and circumstances.
2.
Personalize your experience : Deliver different levels of Nudges based on your user’s skill level.
3.
Optional Enablement : Gives users the option to turn Nudges off or adjust them if they wish.
Application Cases
When using Notion or Github , it is typical to encourage the use of artificial intelligence in the form of ✨. Personally, I think this part can be used well as a browser extension, etc. (An area where Grammarly was good in the past)
Proposal
Suggestions is a strategy where AI systems suggest possible prompts or next steps to users. This helps solve the 'Blank Canvas Syndrome' that users experience when interacting with AI, and creates a smoother conversation flow. Think of it as a way to make Cold Starts easier.
How is it different from the guided method? Guided approaches focus on accuracy and depth of understanding, while suggested approaches focus on ease of use and ongoing conversational flow. In other words, guidance can be thought of as the way AI asks for additional information in response to the user’s initial request, while suggestions can be thought of as the way AI suggests possible prompts or next steps to the user.
Key features of the proposed method
1.
Format : Typically presented as 3-5 clickable text options.
2.
Location : Usually displayed near the text input field.
3.
Function : When you click, the selected suggestion is automatically entered or an AI response is generated right away.
4.
Contextual reflection : Suggestions are generated taking into account the flow of the conversation and the user's previous behavior.
Advantages of the proposed method
1.
Improved usability :
It reduces the time users spend thinking about what to ask.
It allows you to interact with AI without having to write complex commands or prompts.
2.
Introduction to system functions :
Naturally introduces users to the various functions of AI systems.
It provides users with the opportunity to discover and leverage features they didn't know about.
3.
Keep the conversation going :
Helps users when they have trouble continuing a conversation.
Improve user experience by maintaining a natural conversation flow.
4.
Personalization possibilities :
By learning your preferences and behavior patterns, you can make increasingly more relevant suggestions.
Application Cases
If you use chatGPT or GPTs, this is a natural way to see it. It is a way to extract good questions or good answers.
Template method
Templates are a powerful tool for simplifying the use of complex AI systems and helping with user onboarding. They provide a predefined structure that allows users to perform complex tasks with minimal input.
Features of template method
Structured Framework: Define the basic structure of your work in advance.
Flexibility: Use a mix of parameters, suggested prompts, open-ended text, etc.
Automation: Based on user input, AI automatically completes the rest.
User friendly: Allows you to leverage advanced features without having to write complex prompts.
Advantages of template approach
It allows you to leverage advanced AI capabilities without requiring specialized knowledge.
Even those with little experience can expect high-quality results.
In Korea, Luton is systematizing this in its own way. Overseas, Jasper.ai and others have shown good approaches, and recently, they have shown various possibilities such as economic activities through prompt templates. Personally, I think it is similar to the PPT and Notion template market.
It's getting longer than I thought, so I'll continue in the next post.
📚 Welcome to Haebom's archives. --- I post articles related to IT 💻, economy 💰, and humanities 🎭. If you are curious about my thoughts, perspectives or interests, please subscribe. Would you like to be notified when new articles are posted? 🔔 Yes, that means subscribe. haebom@kakao.com
