Beginner’s principles: Design for AI Text Generation

This is episode #1 of what user experience principles to use for text-generative AI products.

As text-generative AI models become more advanced and accessible, it’s crucial to consider the user experience (UX) when implementing these technologies. A well-designed UX can make the difference between an empowering and frustrating interaction. Here are some best practices to keep in mind.

Transparency and expectations

Users should clearly understand the AI’s capabilities and limitations. Could you explain how the system works, what it can and cannot do, and set appropriate expectations? In addition, give the user an alternative that can potentially solve their question or task.

Claude AI is clearly communicating why it cannot perform a certain action and tells what it can do.

Human-centered Design

Always prioritize the user’s needs. Customize the interface and interactions based on their mental models, goals, and pain points. Continuously gather user feedback and make iterative improvements to enhance the experience.

Note. This sounds obvious and that’s because it is. So one of the core parts of gathering user feedback is to do a two-factor research:

1. Let the user use the conversational AI for a set amount of time and interview them afterward. This type of AI product needs to become part of their daily workflow and you want to understand where your product fits in. And optimise from there.
2. Interview them with questions like ‘What were you trying to achieve? How successful or unsuccessful were you in achieving this?’. Why? Because you want to know how they are using the product, so you can validate you are on the right track.

Contextual Guidance

Offer helpful prompts, suggestions, and instructions to guide users through the text generation process. This reduces cognitive load and increases the chances of successful outcomes.

Perpexlity AI suggests specific suggestions to specific related keywords and includes third-party sources to each answer to guide the user through the process.

However, you can only refer to those elements if you can put your money where your mouth is. For example, if the suggestions are wrong, this will affect their trust in your product or feature and they will stop using it.

Quality leads to trust

In the design process, review each (new) feature with the question if the data quality has a minimal set standard. This is essential because if the data is outside the context of the problem you are solving or lacks information, users will start to distrust the outcome of the generated information. And it takes a long time to build trust, and one bad mistake to lose it all. So although you should be somewhat risk-averse, we need to aim for high standards here.

Google has created design principles on how to validate data that’s (ir)relevant for the user. https://pair.withgoogle.com/guidebook/patterns#good-data-practices

Ethical Safeguards

Implement robust guardrails to prevent the AI from generating harmful, biased, or inappropriate content. Communicate the ethical principles and boundaries that the system adheres to.

By following these principles, you can design a UX for text-generative AI that is intuitive, transparent, and tailored to the specific needs of your users. Embrace the power of this transformative technology, but do so with a user-centric approach that sets your solution up for long-term success.

👾 Interested in more content around Product Design for AI and awesome tools that can help you build, scale & iterate your product?

→ Follow me on X / Twitter
→ Tools & templates for product managers in my shop

👾 Product by Pim is free today. But if you enjoyed this post, you can always support me via my Buymeacoffee