Column: Structuring Success

In 1992, Ward Cunningham introduced a metaphor that gave engineering leaders a new vocabulary. Technical debt, he argued, is what you take on when you ship code that is good enough for today but wrong for the long term. Taking a shortcut is not free. Interest accrues. Eventually, you must pay the principal back with all the compounded charges or the system would collapses under the weight of decisions that no one wants to revisit.

Information architecture (IA) and other UX design practices work the same way, but few organizations use this metaphor for IA debt, which is a problem because IA debt is real, measurable, and is likely growing inside every organization that has been running a digital product for more than a year. IA debt accrues quietly, hidden behind small editorial decisions and well-meaning content drops. IA debt compounds because teams must slot every new piece of content into a structure that has already drifted away from coherence. An organization pays back IA debt eventually, in the form of redesigns, lost conversions, and customers who quietly stop showing up.

Organizations must take IA debt seriously. In this column, I’ll explain where IA debt comes from, what it costs, how to audit it, and how to pay it down without setting fire to your IA roadmap. Read More

In this installment of my column, I’ll write about what happens to information architecture (IA) once a meaningful share of a Web site’s traffic no longer navigates the site at all. A growing number of users are asking an artificial intelligence (AI) assistant a question, then getting an answer that an AI has assembled from the site’s content, providing a citation link that drops the user somewhere in the middle of the site, or the user may never visit the site at all.

Information architects have always built sitemaps for visitors who arrive at a site’s front door. Now increasingly more visitors are arriving through a side window—when an AI assistant drops a user into the middle of the site.

Now an AI model often reads a Web site before a human does. People aren’t actually experiencing the site’s navigation, labels, or hierarchy—the things that IA practitioners have spent their careers refining. This is not a hypothetical shift that we can plan for later. AI is already changing referral patterns for content-heavy sites, and this is changing the role of a good information architecture. Read More

Card sorting has been a cornerstone of information architecture for decades. This method works well because it lets real people show you what they think. You can hand participants a set of labeled cards representing pieces of content, ask them to group the cards in whatever way makes sense to them, then study the patterns that emerge. The result: a window into users’ mental models that no amount of internal brainstorming could replicate.

However, the conversation around card sorting is changing. Artificial intelligence (AI) tools can now generate category structures in seconds. Large language models (LLMs) such as ChatGPT and Claude can sort a list of 40 content items into plausible groupings without your needing to recruit a single participant. Some teams have started asking whether traditional card sorting is still worth the effort. Others have gone even further and entirely replaced participant research and information architecture with AI outputs.

Both reactions miss the mark. AI does not make card sorting obsolete, but it does change how smart teams should use it. In this first installment of my new column Structuring Success, I’ll share what AI can and cannot do for card sorting, illustrate the differences that AI can make with real product examples, and offer a practical framework for combining human insights with AI’s speed. Read More