A new approach to Information Architecture in the age of AI?
It's not enough that we need to monitor and offer a reasonable experience to bots — we are now entering an era where we need to work alongside them. This comes with many exciting possibilities, such as cognitive offloading and increased productivity, but it also raises important questions about the future of design and UX. How will this collaboration reshape our creative processes and deliverables?
I'm curious to explore what this change means for deliverables. While AI offers powerful new tools, creativity, taste, and decision-making remain human strengths. This shift may allow us to focus on those strengths, becoming more strategic with our time and expertise.
I've been exploring UX and design resources on this topic, specifically AI-first Information Architecture. The content below is based on a few recent references that I will list below (in case you prefer to read the original work):
- Modern Information Architecture for AI-first Applications by Greg Nudelman & Daria Kempka
- AI Can't Give You Good Taste by Elizabeth Goodspeed
- AI and cognitive offloading: sharing the thinking process with machines by Tetiana Sydorenko
Here are a few key takeaways that resonated with me from these articles:
AI-first Information Architecture framework
AI-first Information Architecture (IA) isn’t just about adding LLM-powered search. Instead of overhauling existing IA principles, the author suggests adapting them to take advantage of modern AI capabilities. Nulman outlines a framework based on five key page types: Analysis Overview, Category Analysis, LLM Search Results, Item Detail, and Q&A Maintenance.
The examples in the article focus on content personalized to the user, which means lots of faceted pages and variations depending on the individual. I wonder if canonicals and robots.txt rules could help manage this new level of personalization. We wouldn't want to overwhelm bots with all these individual-level pages being created.
Here are some content ideas—though they don't seem particularly innovative. I suppose that's just the reality of commerce; perhaps other use cases are more captivating.
Analysis Overview: The Analysis Overview page could be applied to an e-commerce homepage, featuring LLM-generated text summaries tailored to seasonality and shopping incentives, such as limited-time discounts, free delivery, and other promotional highlights. All content generated or updated by LLMs.
Category Analysis: The main change to category pages is to return results based on what the customer is more willing to buy and also offering complementary products based on the user history of shopping. The hero image could also be generated in a customized way by an AI tool such as Midjourney.
LLM Search Results: Here, we see a significant shift driven by machine interpretation of the query, allowing for a more conversational tone on the page. The content could return top products based on various factors such as user purchase history, product reviews aligned with the user's main objectives, and the preservation of facets with detailed tags to enhance search precision further.
Item Detail with Contextualized Search: The goal here is to create a more conversational experience, moving beyond constant scrolling. By focusing the search within the product, users can explore specific details—such as, “Are these sneakers good for pronation?” or “Is it waterproof?”—or even match the product with previous purchases to ensure it complements what they already own. Again, placing a greater emphasis on personalized content and using the search box rather than relying solely on scrolling to gather information about a product.
Q&A Maintenance: Lastly, a conversational tone in Q&A is essential. I appreciate responses from other users when asking about a product, but perhaps a machine-summarized answer based on what's available in reviews could be helpful?
The Importance of Taste in the Age of AI
Do we want content created by machines to lead our experiences? I’m not convinced this is the best approach. While AI can assist with certain tasks, especially when trained models demonstrate accuracy, as Goodspeed highlights, curation, and judgment are qualities that machines cannot fully replicate. Additionally, a non-algorithmic experience, free from hidden agendas, may be what users are looking for. User-personalized content created by machines will not eliminate the need for incognito pages, which will still require human oversight and regular updates. Companies will continue to need to manage canonicals effectively. Infinite content, especially when generated by machines, is not sustainable and risks becoming spammy. The value of content curation has never been greater. In a world of infinite scrolling, having a trustworthy source is invaluable.
Conclusion
At this point, it's helpful to utilize AI platforms and LLMs for research and refining the grammar of our content. AI features can be leveraged to improve user experiences, but not to the point of redundancy or where AI simply replicates existing knowledge (as highlighted in the Strava discussion on AI). I anticipate revisiting this article as new and innovative use cases arise.
What is AI-first Information Architecture?
AI-first Information Architecture (IA) adapts existing IA principles to leverage modern AI capabilities, rather than overhauling them. It involves structuring and labeling information in a system, whether digital or physical, to ensure easy retrieval. It is not just about adding LLM-powered search. The goal is to tell the story in the context of the customer's need, using their language and understanding the problem they are trying to solve.
What are the key page types in Nudelman's AI-first Information Architecture framework?
Greg Nudelman outlines a framework based on five key page types: Analysis Overview, Category Analysis, LLM Search Results, Item Detail with Contextualized Search, and Q&A Maintenance.
How does AI impact the role of taste and curation in content creation?
While AI can generate content and optimize for engagement, it cannot fully replicate human qualities like taste, curation, and judgment. In a world of infinite AI-generated possibilities, human taste becomes the essential differentiator and a crucial skill. The focus shifts from execution to curation, where humans make critical judgments about what feels right and what truly matters.
What are the potential downsides of AI-generated content?
Infinite content generated by machines risks becoming spammy and unsustainable. User-personalized content created by AI will still require human oversight and regular updates. There is also the risk of AI-generated content lacking a non-algorithmic experience and potentially containing hidden agendas.
How can AI be used to improve UX design?
AI can assist in UX design by brainstorming ideas, drafting wireframes, creating user interfaces, and predicting user behavior. It can also analyze large volumes of user data, optimize design prototyping and testing, improve UX and product copy, increase design accessibility, customize the user journey, and enhance UI design. However, it's important to balance AI insights with the personalized, human touch that defines exceptional UX design.