User Needs Specific to Generative AI

Deducing Good UX Design for LLM Technology

5 min readJul 4, 2023

When we design for Generative AI, as UX professional we need to be aware of the impact brought by generative AI technology. The foundational principle in UX Design is focusing on the user's needs. User needs will vary according to the domain and goal of the product. But to understand user needs specific to generative AI Technology, we need to start with understanding the common user journey in generative AI.

In any generative AI product, this will be the typical user journey.

Phase 1: Users send instructions & their required input as Natural Language in the given interface.

Phase 2: Users Validate the received output.

Phase 3: Users correct the received output If needed.

Phase 4: Users utilize the output for their professional or personal purposes.

Phase 1: Input + User-Made Instructions (Prompt)

In traditional programming, the user gives input and selects the predetermined rule or instruction made by programmers, Then the input is processed according to the rule, and output is delivered.

But in generative AI, the user not only gives the input but also creates the instruction or rule in natural language which we all popularly call “Prompt”. The boundary of these user-made instructions is very broad, not very specific like programmatic instructions.

Writing a prompt with clear well-defined instructions is not a cup of tea for everyone. Many of us might underestimate this problem. Writing is more challenging than reading and understanding.

The recent NN Group article also highlights that “Based on recent literacy research, I deem it likely that half the population in rich countries is not articulate enough to get good results from one of the current AI bots.”

Especially in Asian countries like Korea and India definitely, who are non-native English speakers are likely to have articulation problems.

Need 1: User needs assistance/guidance for prompt writing

Phase 2: Users Validate the received output.

This phase is very crucial for any task productivity. Majorly there are three scenarios associated with this phase.

Scenario 1: If the users have the expertise to validate the output, then they might reject or utilize the output with some customization.

Scenario 2: If the users lack the expertise to validate the output, then they might take extra efforts to validate the credibility of the output. it might be a simple Google search or even a directly discussing with a subject matter expert.

Scenario 3: If the users skip the validation due to ignorance of generative AI limitations, then they are risking task productivity and their individual credibility.

Need 2: User needs the nudge & assitance to validate the AI output.

Phase 3: Users Correct the received output.

Generative AI processing logic is probabilistic. Imagine it as a big collection of text with interconnected patterns. When you give it some text as input, it looks for patterns in that input and tries to find the most likely matching text based on those patterns. It doesn’t follow strict rules or deterministic instructions like traditional programs. So it will give a probabilistic output.

Due to this probabilistic nature, AI lacks task specificity in design. For the same input and the same prompt, it is more likely to give different outputs with varied accuracy, especially for complex and medium tasks. So if the user gives the instructions and input to the AI system, they are more likely to get output that needs correction. Currently, popular Generative AI user interfaces don't provide this option

Need 3: User needs efficient correction options to edit, refine or recover the AI generated Output

Major User Needs Specific to Generative AI

Need 1: User needs assistance/guidance for prompt writing.
Need 2: User needs the nudge & assistance to validate the AI output.
Need 3: Users need efficient correction options to edit, refine or recover the AI-generated Output.

However, As Discussed earlier User needs will vary according to the domain and goal of the product. I like to highlight that the above needs will arise when you implement Generative AI Technology in your system or product.

Potential Solutions already available in the Market

User Interfaces that support & guide prompt writing.

1. Easy Google Search option 2.Mentioning original Sources & References of the Output.

In conclusion, addressing the user needs specific to generative AI technology requires a combination of research-based insights and creative solutions from UX professionals. By conducting thorough UX research, designers can gain a deep understanding of the challenges users face at each phase of the generative AI journey. By doing so, they can bridge the gap between users and generative AI technology, providing experiences that are both user-centric and valuable in meeting the evolving demands of the digital landscape.

Feel Free to connect with me on Linkedin! https://www.linkedin.com/in/isanthoshgandhi