Don’t Outsource Findability Decisions to AI
Why Information Architecture & Taxonomy Are Important
Every digital platform should have a navigation scheme and labels that its target audiences understand, so they can find what they need and engage in the ways that bring value to them and to the organization who owns the digital property.
Achieving clarity means organizing content and features in ways that are logical to your audiences, and using words they use. Whether you oversee a website, a learning platform, or a collection of services or functions in an app, reflecting your audiences’ organizational logic and use of language are important to usability and adoption of digital properties.
‘Hands Off’ Is Risky
The key to usable navigation and content classification (taxonomy) has always been to understand the content and how the target audiences think about and use it. The tried and true way to mess up navigation and taxonomy in the past has been to not understand the content, not understand the audiences, or forget what you knew about those entities and follow some other logic—like what your company wants to sell, or what the new marketing executive wants to call your main navigation categories.
Today there are tools at our disposal to help classify and organize content—but they can also mess up navigation and taxonomy at scale, if you let them operate under misunderstandings, or have giant blind spots, or make decisions absent real understanding of your target audiences and/or company’s strategy.
It’s very easy to just let the tools do their thing. Content management systems have embedded AI. External classification tools (semantic technologies) and commercial LLMs (large language models) enable us to derive navigation schemes and classification taxonomies in a fraction of the time it used to take using manual processes.
But tools that produce helpful first drafts because they recognize patterns in language (that’s what LLMs do, after all) should not be entrusted with important, bottom-line decisions about how your audiences and search technologies should be able to discover your content. Or what concepts should be prioritized on your website. Or how and whether you should have a flat, hierarchical, or polyhierarchical classification taxonomy for the scientists using your journal archives.
For example, I recently saw a presentation about a company that entrusted too much of its design process to developers and AI and ended up with an menu of close to 100 functions in a workplace productivity app. When subject matter experts were brought into the project, they created logical groupings for the menu items, based on how people use the app in their jobs. The menu was greatly reduced in length and became more user-friendly. Only humans with real-world job experience could have made these improvements.
The tools available now to assist information architects are amazing. But you still need the information architect, and you still need to know how your target audiences think about what they need from you, how they expect to find it, and what they call it. A misalignment can lead to alienated audiences, poor engagement, usage, and conversions (clicks, purchases, registrations, etc.) for your digital property.
Humans in the Loop
What does the ideal approach to information architecture and taxonomy look like? The exact steps in the process can vary, but a user-centered design (UCD) approach involves subject matter experts who lend their insights to the organization and classification, and test out your taxonomy with real content.
It involves user research with your end users, to gather insights early and later validate your proposed navigation and classification.
It means always having humans in the loop—project leaders, internal experts, and external stakeholders—and not letting any technology make important decisions for you.
This ideal balance of human and automated decision making is not always put into practice. Lately, organizations have entrusted an awful lot to AI-endowed tools, including website navigation, content classification, search filters, button language, brand messaging—even “audience research,” if they do it. Instead of optimizing a user-centered design process with the help of AI tools, many companies have truncated their design processes and cut out real users entirely, relying on what AI knows from training data and the artifacts the companies can supply. A fully mechanized approach cedes human control and invites missteps. Involving actual end users and subject matter experts in the design process is a safety net.
It’s best to start with what your audiences’ most important tasks are and how they describe both the tasks and what they need to look for to achieve them, and apply that learning to your information architecture and taxonomy. This knowledge should also inform what content is actually needed, and where your content gaps may be.
To understand how your audiences think about relevant tasks and content (their “mental models” of information), it’s helpful to—
Do user research or audience research to inform your project. Ask audiences about their most pressing needs that your website, app, or digital ecosystem can fulfill.
Get input from users about what they look for when they are trying to answer a question or solve a problem. (You can look at search query data for some of this information, and use feedback tools, surveys, and interviews to get these insights.)
Make sure page names and document names are standardized and human-readable.
Make sure pages use relevant keywords in the title and body content.
Clearly identify the purpose (associated tasks), dates, and intended audiences for content in its metadata and copy; this will help get the right content to the audiences that need it.
Make sure all content contains the appropriate markup schema so it is accessible and visible to search engines, including generative AI search (AEO, GAIO); for instance, use article schema for various types of articles, FAQ schema for your FAQ page, news schema for news, etc.
None of this is new—we’ve been organizing and classifying digital content appropriately for audiences for decades. But when so many organizations are throwing their content into a CMS, adding AI, and hoping for the best, it’s a good time to step back and remind ourselves that information architecture and classification taxonomies need to be INTENTIONAL and tailored to their audiences in order to work.
Getting input from users and target audiences and adjusting your digital content accordingly should in fact be part the governance plan for any digital property.
Want to talk about using audience insights plus AI to improve your navigation and/or classification for browsing and search? Need help with content governance and performance measurement? Get in touch.
Photo: A physical card sorting exercise to get subject matter experts’ ideas on organizing content. Thanks to our clients at AAMVA.
