Designing User Experiences for Generative AI

We’ve all heard of it by now — generative AI. It is reshaping how we interact with digital interfaces. It helps us search for and synthesize information instantly. It creates text, imagery, music, in any style. It analyses and interprets data. It can even generate code and integrate with software to create new user experiences.

For product designers, this has big implications. As Jakob Nielsen outlines, “AI systems are shaping up to launch the third user-interface paradigm in the history of computing — the first new interaction model in more than 60 years” (source). Currently, users have to interact with the computer using a language of boxes, dropdowns, input fields, buttons, etc. Now, users can interact with the computer using their own natural language. This presents an enormous opportunity for how we design and simplify user experiences.

So how do we design for these experiences? Here are some principles to keep in mind should you find yourself designing a generative AI tool for your software.

Principle #1: The design gives the user support

When I speak with users about generative AI tools, I often hear these common phrases: “What can I ask the AI to do?”, “Is it going to understand the way I speak?”, “I’m not sure what to say!”. This “articulation barrier is a usability challenge for [AI tools]” (source) and is something the design needs to solve for. Here are some tactics you could employ in your design:

• Provide prompt examples or a prompt library to guide users who don’t know where to begin.
• Offer instructional text on how to effectively communicate with the model, including tips on using specific keywords.
• Consider avoiding natural language prompts altogether — explore image prompts or pre-written inputs to ease users into the interaction.

Adobe Firefly provides scrollable prompt library

Pinterest Lens allows you to generate search results using screenshots

Principle #2: The design keeps the user in control

Generative AI models can be non-deterministic or unpredictable by nature. They can also “hallucinate”, which is when the AI produces incorrect or misleading results. And many times, the output just doesn’t meet the user’s expectations. This can make users feel apprehensive, which can impact engagement and retention. The design must work to give users a sense of control in this environment. Below are some ideas to explore:

• Provide disclaimers that the technology is experimental and can be wrong. This transparency can help buffer any eventual disappointment the user feels with the results.
• Offer the user multiple output options to choose from. This can enhance their feeling of control if/when results are imperfect.
• Build in a feedback mechanism to include user in the process of training the model.
• Include the ability to refine the model’s initial output. For example, create a conversational UX, allow results to be edited using an interface the user is already familiar with, or clearly label outputs as drafts. This gets users in the habit of understanding that the model’s output is just a first draft, not the final result.

Klaviyo’s AI segment builder provides a disclaimer, feedback loop, and ability to edit in classic builder

Canva provides multiple AI generated results alongside pre-made templates

Principle #3: The design cultivates a sense of wonder

The value of generative AI is rooted in its capacity to astonish and inspire users. It can produce results faster than we can, and sometimes come up with an output we never would have imagined. Its variable rewards (source), while sometimes unsettling, can also be exactly what hooks the user. Look for ways to highlight this magic through your design:

• Lean into micro-interactions, animations, and distinct UI elements like the sparkle icon that enhance excitement and allure.
• Create engaging loading states (which can be lengthy) that build anticipation in a positive way.
• Incorporate elements that build curiosity and encourage exploration of the AI’s capabilities like teasers and gamification.

ChatGPT loading state lifts the hood behind what the model is doing in live time

Inflection’s Pi offers subtle animations that make conversational UX visually engaging

In designing a UX for generative AI, the design’s focus must be on guiding users through this new landscape with clarity and excitement. Experiences must be both functional and captivating. Additionally, the field of generative AI is moving so fast that designers must keep a finger on the pulse with shifting user expectations and technological advances. The challenge is a lofty one for designers, but also an incredibly exhilarating one.

If you’re interested in diving down the rabbit hole of generative AI design, apply to work on the design team at Klaviyo.