AI Search Shifts from Discovery to Decision-Making

Today's AI UX developments show three critical shifts: 🚀 Perplexity's Comet browser launches with AI-native patterns - no more traditional search boxes, just natural conversation integrated into browsing 🛡️ OpenAI Japan rolls out comprehensive teen safety blueprint - stronger protections, parental controls, and educational resources for responsible AI use 📊 3M Americans daily ask ChatGPT about compensation - AI is democratizing workplace transparency and salary negotiations ⚠️ Meta's rogue AI agent exposes unauthorized data - highlighting the urgent need for better AI agent controls These stories reveal AI moving from tools we use TO experiences we live within. The UX implications are massive. Read the full analysis: https://lnkd.in/gHDbSvj9 #AIUX#UserExperience#AIDesign#TechTrends#DigitalTransformation

84% of researchers are using AI. Only 11% have heard of tools built for scholarly work. They're not waiting for us. They're using ChatGPT, without access to the verified content that would make it reliable. The ExplanAItions 2025 study makes it clear: the opportunity for publishers is to make our content accessible where research actually happens. Wiley's VP of Product & UX sets out exactly what that looks like in his latest article https://ow.ly/N7j950YrUiC #ScholarlyPublishing #ResearchIntegrity #AIinResearch

👾Two weeks of intensive theory and hands-on practice in Design patterns for AI-interfaces. Over the past weeks, I’ve explored how AI works under the hood, its capabilities, useful frameworks for orchestration, strategies for where and how to implement AI features, how (to try😉) to measure AI success, input and output patterns, and real-world integration workflows. The technology is incredibly powerful, but I still think it’s just a tool. As designers, we need to be cautious about AI-powered hype. The real sign of good design is whether we’re solving the right problems, not creating new ones just because they’re powered by AI. So a few good old buttons here and there won’t hurt anyone 🧘♀️ And after all the incredible tools I learned in this AI course, my biggest question is: how is ChatGPT so popular despite having such a poor UX?😅 Thank you, Vitaly Friedman for such insightful two weeks✨💛

Stop Using AI to Generate UI. Start Using It to Destroy Bad Assumptions. THE ASSUMPTION AUDIT PROTOCOL (using AI): STEP 1: Write out your current design hypothesis Example: "Users want a single-page checkout because it reduces cognitive load." STEP 2: Feed it to Claude or ChatGPT with this prompt: "You are a UX researcher with deep expertise in e-commerce behaviour. Here is my design hypothesis: [hypothesis]. Generate the top 10 assumptions I'm making that could be wrong. For each, describe what evidence would prove it false." STEP 3: You now have a research agenda. Not a feature list. A list of the beliefs your design is betting on. STEP 4: Rank assumptions by risk and learnability. High risk + easy to test = test immediately. High risk + hard to test = flag for leadership as strategic uncertainty. STEP 5: Design experiments, not solutions. Your "design" is now a hypothesis-testing instrument, not a deliverable. **** - AI is extraordinary at surfacing assumptions because it has no ego invested in your design being right. - The best UX tool is the one that makes you question your certainty. - AI does that better than any other tool in the stack. #DesignStrategy #AIDesign #UXResearch #DesignThinking #ProductDesign #AssumptionTesting #AI #Claude

One of the biggest missed opportunities in AI product design is still hiding in plain sight: Even some of the most advanced AI chat products still do not clearly separate how the model should work on something from what the model should work on. For casual use, that may be fine. For advanced use, it becomes a real limitation. Today, users often have to combine reasoning instructions, task framing, tone guidance, source material, and persistent preferences in the same interaction layer. It works, but not cleanly. More importantly, it makes the system harder to understand, harder to teach, and harder to scale for serious workflows. A better design would introduce a clearer structure: 1. “How to process” A dedicated input for the context modifier — the reasoning frame, transformation method, instruction set, or prompt logic guiding the output. 2. “What to process” A separate space for the information pool — the content, documents, notes, or inputs the model should analyze. 3. Reusable custom prompts Users should be able to save the “How to process” layer as a reusable prompt template, auto-populate it when selected, refine it in the moment, and save the improved version without friction. 4. A global context layer A version of this already exists today in some products through personalization or memory settings, but it should be more clearly structured and persist as an intentional context layer that users can manage explicitly across sessions. This kind of separation would do more than improve usability. It would help users build a more accurate mental model of how AI systems actually function. Because in practice, effective AI interaction is rarely just “one prompt.” It is a layered system: • the processing logic • the information itself • the persistent context • the live refinement happening inside a session Right now, many products collapse these layers into one interface and expect users to manage the complexity manually. That may be acceptable for beginners. It is not ideal for power users, teams, educators, or anyone trying to build repeatable workflows. The next wave of AI UX should not just be about better models. It should be about better structure. And a clean separation between processing logic, content, and persistent context feels like one of the most obvious places to start. It’s honestly surprising that this still isn’t a standard design pattern — especially in platforms like ChatGPT and other leading AI assistants that are shaping how millions of people learn to work with AI. #AI #ProductDesign #AIUX #PromptEngineering OpenAI

NN/g says narrower AI features are typically easier for new users to understand and adopt than broad, vague systems. "When designing an AI feature, its scope (how broad or narrow its capabilities are) influences its usability. Our research shows that narrower AI features are (typically) easier for new users to understand and adopt." Read the article here: https://lnkd.in/ge6din9g

Most enterprises have adopted chat interfaces. But work doesn’t happen in chat. It happens in meetings. In real-time decisions. In conversations where speed, clarity, and continuity matter. That’s the gap, the forced translation, the slow decision cycles. It pulls AI out of the flow of work. Voice does the opposite. ~150 words per minute vs ~40 typing This is not a UX upgrade. It’s a shift from AI as a tool to AI as a participant. I put together a perspective on why the next enterprise interface is voice—and why this shift is happening now, not later. https://lnkd.in/en7RTCw8

Users of websites like Amazon, Turo, and Redfin might find AI chatbots “useful when they answer context-specific questions, clarify complex information, and offer tailored guidance that helps users make decisions” Nielsen Norman Group https://lnkd.in/gaN4f-QP #contentdesign #AIUX #ContentEngineering

#FridayFeeling Had lots of #AI discussions this week, specifically agentic AI and chatbots for public services. Made me wonder how human, do humans, want their "chatbots" to be? It turns out your AI doesn’t need to act human to keep people happy — it just needs to do its job well. A 2025 study of 525 chatbot users found that what really drives satisfaction isn’t a super‑human personality… it’s the basics of trust, perceived competence, warmth and a friendly, social‑oriented communication style. Making bots feel human — matters far less than people may assume. In many cases, people don’t need a chatbot to pretend it’s their new best friend… they just want it to be reliable, helpful, and not sound like Buck Roger's loyal companion, Twiki. So maybe the future of great AI UX is less about “making it human” and more about “making it solid, warm, and actually useful.” People, most probably want a a competent chatbot over a chatty one any day. How about you?

When people interact with public services, whether through a human adviser or a digital interface, they are asking the same underlying question: can I trust this, and will it actually help me? Competence and warmth are not new discoveries in service design; they are central to well-designed consultation practice too. The trust dimension is particularly striking. In consultation, we know that the perceived competence of the body running an engagement exercise shapes whether people bother to participate at all. The same logic applies to a chatbot handling a planning inquiry or a social care referral. If it fumbles the basics, the consequences extend well beyond a poor user experience; they undermine confidence in the whole process. The public services context also surfaces a question your post touches on but does not fully name: when does a helpful and reliable chatbot become a gatekeeper that quietly narrows the range of responses a citizen can give? That is a consultation design question as much as a UX one, and it is one practitioners need to be asking now.