A very practical set of recommendations for using AI responsibly in a research workflow
Watch this WaysConf session, then continue with related talks or explore the current programme.
In this WaysConf 2025 presentation, Kate Moran, VP of Research and Content at Nielsen Norman Group, addresses why users frequently ignore or reject newly introduced AI features. Drawing on qualitative user research, Moran identifies three primary barriers to adoption: a lack of genuine user value, poor findability and discoverability, and the steep learning curve of open-ended conversational interfaces. Using case studies like LinkedIn's retired suggested questions and Amazon's Rufus shopping assistant, she demonstrates how breaking established design patterns and anthropomorphizing tools with human names confuses users. To overcome these challenges, Moran advises product teams to stick to proven UI standards, prioritize narrow-scoped AI features that integrate into existing workflows, and rely on rigorous qualitative research with real humans rather than synthetic personas.
AI features built solely due to industry pressure or stakeholder demands without solving a real user problem will be rejected. LinkedIn's suggested follow-up questions are a prime example of a high-visibility feature that was quickly retired due to lack of utility.
Watch from 6:01Even highly valuable AI features will go unused if they suffer from poor findability and discoverability. In NNG's studies, no participants naturally discovered Amazon's Rufus assistant because it was hidden behind a non-standard icon and name in the global navigation.
Watch from 12:53Open-ended chat boxes require high cognitive effort, creativity, and prompt-engineering skills that average users do not possess. Most consumers default to simplistic, keyword-based search queries rather than writing detailed, context-rich prompts.
Watch from 25:04AI features that are tightly integrated into existing workflows and require zero user prompting are much more successful. Amazon's automated review summaries are highly discoverable and useful because they assist users directly within their natural shopping path.
Watch from 30:21Giving AI assistants human names, faces, or personalities often triggers skepticism and pushes users into the uncanny valley. Product teams should maintain transparency by avoiding anthropomorphism and clearly labeling AI tools as automated assistants.
Watch from 37:26These concise notes were edited from automatic captions and checked against the talk structure. They are not a verbatim transcript.
During the AI hype of 2023 and 2024, LinkedIn introduced a premium feature that suggested follow-up questions under feed posts. However, these AI-generated suggestions often asked questions that the posts had already answered. Because the feature offered no real user value, LinkedIn quietly retired it less than a year after its optimistic launch. This serves as a warning that users will quickly abandon and mock features that do not solve actual problems, regardless of how much money was spent building them.
Watch from 6:01Nielsen Norman Group evaluated Amazon's Rufus shopping assistant and found that while the tool provided genuine value by answering product questions without forcing users to leave the site, it suffered from terrible discoverability. In natural usage tests, not a single participant noticed or clicked the Rufus button on their own. Even when explicitly told a chat assistant existed, users could not find it because it was tucked away in a non-standard location with an unfamiliar icon.
Watch from 9:53Many design influencers claim that conversational interfaces are perfectly intuitive, but this ignores the reality of average users. Operating an open-ended chat box requires significant imagination, articulation, and a willingness to experiment. While UX professionals are comfortable with technology, regular consumers struggle with prompt engineering. They default to typing simple keyword queries, treating advanced AI systems like old-fashioned search engines and missing out on their full potential.
Watch from 25:04AI features do not always need to be designed as chat interfaces to be highly effective. Amazon's AI-generated review summaries are a massive success because they require zero user interaction and are placed directly within the customer's natural shopping workflow. Because users are already looking at product reviews, they naturally discover and benefit from these summaries without having to learn new prompting behaviors or navigate complex chat windows.
Watch from 30:21Giving AI assistants human names like Rufus or Alexa, or assigning them faces and personalities, frequently backfires. Qualitative research shows that users become highly skeptical and uncomfortable when they cannot tell if they are interacting with a real person or a bot. To build trust, product teams should remain transparent, avoid humanizing elements that trigger the uncanny valley, and clearly present the tool as an automated assistant.
Watch from 37:26Some companies react to low adoption rates by plastering AI assistants across every menu and toolbar, as seen with Microsoft's Copilot or Google's Gemini. This aggressive approach usually stems from relying solely on quantitative analytics rather than qualitative research. Shoving a feature in a user's face does not fix the underlying issue if they have already tried it, found it unhelpful, or are being interrupted at inappropriate times, such as being offered a document summary after writing only two sentences.
Watch from 43:02
