Creating new AI-based features or integrating AI into existing products goes beyond simply adding an AI chatbot to your website. However, adding ✨AI✨ does not magically create value. In fact, in some cases, it’s a huge investment for a relatively small return.
This collection of resources describes measured approaches to deciding whether AI adds value in the first place and gives specific recommendations for designing AI features based on our research.
It’s easy to optimistically talk about AI as something that can swoop in and make anything easier. It’s also easy (and somewhat more gratifying) to skeptically talk about AI as completely overhyped and hollow. Neither approach is unilaterally true. We must give AI a real chance but be willing to call out whenever “the emperor has no clothes on” because it is not making a valuable difference.
The Strategy and Value Proposition of AI
Excited companies chasing after a new technology are not new. In 2021, well before the release of LLMs like ChatGPT, our former CEO, Jakob Nielsen, warned against overinvesting in the promise of the tech without a clear understanding of how it solves real user needs.
There is nothing wrong with seeing AI as a powerful hammer in your toolbox, but there’s something seriously wrong with trying to find a nail to hit with your favorite hammer rather than thinking about what you’re trying to build.
|
Number |
Link |
Format |
Description |
|
1 |
article |
Companies should lead with the value that AI implementations offer rather than counting on AI to create value. |
|
|
2 |
article |
AI implementations created for novelty rarely produce real value. Seek to solve real pain points. |
|
|
3 |
article |
Product teams must avoid rushing to adopt chat interfaces unless they meet real user needs. |
|
|
4 |
article |
Narrowly scoped AI features are easier to understand and have better user adoption. |
|
|
5 |
article |
AI’s four “superpowers” are content creation, summarization, basic data analysis, and perspective taking. |
|
|
6 |
Common-Sense AI Integration: Lessons from the Cofounder of Condens |
article |
Ask these 3 key questions to ensure AI integrations are valuable and appropriate. |
|
7 |
article |
Henry Modisett, head of design at Perplexity, shared advice for designing AI features. |
|
|
8 |
article |
Generative UIs could create custom interfaces for each user, focused on their desired outcomes. |
|
|
9 |
video |
||
|
10 |
article |
AI agents now interact with digital interfaces alongside humans, which requires rethinking what “user” means and prioritizing accessibility. |
|
|
11 |
article |
Users see little reason to use site-specific AI chatbots unless those chatbots solve problems that existing site features cannot. |
|
|
12 |
video |
Products deliver more value when teams start from the user problem rather than from the AI technology. |
Prompt Assistance
The power of LLM-based AI features is largely mediated by users’ ability to write good prompts. While newer models often yield useful outputs with underspecified prompts, they still cannot read people’s minds. To make matters worse, users don’t always know what they are looking for or what the AI is fully capable of. Thus, it’s critical to support users in their prompting efforts.
|
Number |
Link |
Format |
Description |
|
1 |
Overcoming the Articulation Barrier in Generative AI Using Hybrid Interfaces |
article |
UIs that combine prompt-based inputs with a graphical user interface (GUI) can improve the usability of image-generation tools. |
|
2 |
article |
Well-designed use-case prompt suggestions support learnability and creativity, helping users explore what AI tools can do while setting realistic expectations. |
|
|
3 |
article |
Prompt suggestions must be contextually relevant, personalized, and specific both to the task and to the user’s level of experience. |
|
|
4 |
Prompt Controls in GenAI Chatbots: 4 Main Uses and Best Practices |
article |
Prompt controls can increase AI feature discoverability, educate and inspire, set constraints, and facilitate follow-ups. |
|
5 |
The Most Exciting Development in GenUI: Buttons and Checkboxes |
article |
AI chats generate simple UI elements to gather context, which reduces typing and memory load and produces more personalized results. |
Product and Feature-Specific Recommendations
Once you decide that AI can add value to a product or feature, you’d better build it correctly. These resources highlight specific recommendations and observations we’ve made about designing AI-powered products and features well.
|
Number |
Link |
Format |
Description |
|
1 |
article |
AI summaries of product reviews are well-received when they are specific and transparent. |
|
|
2 |
article |
Generative-AI outputs often violate web-writing principles. They should still be concise, scannable, follow the inverted pyramid, and use plain language. |
|
|
3 |
article |
Amazon’s ”Rufus” demonstrates that even valuable AI features have low impact if they aren’t noticed. |
|
|
4 |
article |
AI integrated into intranets can boost productivity, support career growth, and tailor experiences. |
|
|
5 |
video |
||
|
6 |
article |
The sparkles icon is inherently ambiguous, but it is becoming increasingly associated with AI. |
|
|
7 |
article |
Five qualities make site-specific AI chatbots trustworthy: handoff willingness, flexibility, proactivity, emotional responsiveness, and transparency. |
|
|
8 |
article |
Different roles within an enterprise need different types of explanations for the same AI output. |
|
|
9 |
article |
Applying information-architecture principles to AI systems helps agents interpret information and produce better-aligned responses. |
|
|
10 |
article |
A study of Qwen’s AI agent yields four design lessons: support discoverability, reuse familiar patterns, handle personal data carefully, and protect user autonomy. |
|
|
11 |
article |
Helpful site-specific AI chatbots state their capabilities clearly, offer relevant prompt suggestions, and quickly signal what they can see. |
|
|
12 |
Less Chat, More Answer: Site AI Chatbots Need to Get to the Point |
article |
Users turn to site-specific chatbots for quick answers rather than conversation, so responses should be direct, scannable, and expandable. |
|
13 |
article |
Explanation text in AI chat interfaces is intended to help users understand outputs, but current practices fall short of that goal. |
|
|
14 |
video |
Product-specific genAI must follow common digital-writing practices to fit the way users scan. |
State of AI-Product Design
The resources in this section report on recommendations or cautions given at a specific point in time. While some highlight findings that still hold today, others have likely changed as AI’s capabilities have grown and its adoption in the UX field has evolved.
|
Number |
Link |
Format |
Description |
|
1 |
Initial Impressions of ChatGPT’s Agent: Successful, Shaky, and Slow |
article |
ChatGPT’s agent could successfully book a restaurant reservation, but the process was slow and error-prone. (2025) |
|
2 |
The Relationship Between Artificial Intelligence and User Experience |
video |
Focusing on user needs is more important than chasing a new technology. AI can potentially understand user intentions and make UX practitioners more productive. (2021) |
|
3 |
video |
Over-hyped marketing of intelligent assistants might ultimately hurt people’s willingness to use them if the UX is disappointing. (2019) |
|
|
4 |
video |
Jakob Nielsen observed that AI will be limited until it can understand both human language and intention. (2016) |
|
|
5 |
article |
As AI-generated content proliferates, users treat visibly human-made design as a signal of trustworthiness. |
Podcast Episodes
|
Number |
Link |
Format |
Description |
|
1 |
36. AI & UX: Innovations, Challenges, and Impact (ft. Henry Modisett) |
podcast |
Henry Modisett, head of design at Perplexity, shared advice for designing AI features. |
|
2 |
43. Responsible AI (feat. Paige Lord, Sr. Product Marketing Manager at GitHub) |
podcast |
Paige Lord offers some insight into ways UX professionals can help create ethical AI products and services. |
|
3 |
47. Designing AI Experiences: What to Consider (feat. Caleb Sponheim PhD, NN/g) |
podcast |
Designers tasked with creating AI-powered experiences must prioritize value and safety. |
|
4 |
57. The Next Decade of UX: Why Psychology Matters More Than Ever (feat. Thomas Watkins) |
podcast |
Designers must specify user intent so that the AI can create products that meet user needs. |
