Don’t Outsource Findability to AI
Findability by browsing or searching is high stakes, whether you oversee a website, a learning platform, or an app. Digital content should be written, organized, labeled, classified, and cross-referenced in ways that are logical to its target audiences, so they can find what they need and engage in valuable ways. This is true whether audiences encounter your content on a website, in an app, or in organic or generative AI search.
Content structure, navigation menus, link names, and classification taxonomies (e.g., used to filter, tag, and cross-reference content) are hugely important to making content findable. Traditionally, creating information architecture and taxonomies for complex, content-rich environments has been a painstaking process. AI tools can now assist in some of the tasks information architects and taxonomists do to ensure findability, making these professionals more efficient. But human governance and judgment is still needed; the tools should not be in charge of the process, or the results.
AI is a tool, not an approach. AI can help achieve content findability, but humans must be in charge. They know the unique audiences (or should). They have the domain expertise. They can identify risk. They are ultimately responsible for the user experience and the success of the digital product.
Putting it bluntly, if you’d be held responsible for a failure, don’t leave your work to AI.
Achieving Findability
The key to usable navigation and content classification has always been to understand the content and how your target audiences think about and use it—and then present content in a way that makes sense to them.
If I’m organizing a website for a scientific organization, I need to work closely with subject matter experts who understand the subject domain. I need to talk to the scientists who use their standards documents and attend their conferences and read and publish in their journals, in order to understand what’s important to them and why. (As an aside, the “why” is something we can’t get from analytics; only humans can tell us that.) I also need to get my client’s team in alignment about their goals as an organization and their goals for the website or digital platform, so we don’t stray from these guardrails, for instance, emphasizing something audiences want but that isn’t aligned with the client’s mission or goals.
When creating or revising information architecture, it’s best to focus on—
your audiences’ most important tasks that they can do on your website or app, and
how they describe both the tasks and the content they need to achieve the tasks.
This typically means doing audience research to inform your project. You can also find insights in the search queries (questions, keywords, or phrases) your audiences are putting into search engines. This knowledge should inform the organizational structure, labeling, and classification of content as well as decisions about what content is needed and how it should be prioritized.
For taxonomy, keys to success include harmonizing any redundant or overlapping term sets, rationalizing similar terms, and validating the taxonomy with real content and subject matter experts.
Making a Mess at Scale
The tried and true way to mess up navigation and taxonomy in the past has been to misunderstand the content, not understand the audiences, be unaware of usability best practices, or disregard what you know and follow some other logic—like what your company wants customers to want, or what the new marketing executive says your main navigation categories to be.
Humans can make mistakes and create findability problems, but AI can accelerate these types of problems. AI is good at making poorly rationalized suggestions very quickly, and sounding wholly confident about them. Multiply a misalignment between your digital platform and your intended audiences by thousands of pages of content, or millions of potential users, and you have a serious problem.
There’s a large company that cut its user experience team and entrusted much of its design process to developers and AI. They ended up with a flat navigation menu of close to 100 items in their workplace productivity app. It was hard for users to find what they needed or know what some of the items were from their labels. When subject matter experts were finally brought in to optimize the unwieldy menu, they created logical groupings 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.
Your safest path forward in using AI in information architecture and taxonomy creation or refinement is to rely on human expertise and guidance, a solid framework of iterative prompts, and validation checks along the way. AI tools can operate under misunderstandings, hallucinate or fabricate, or make decisions that are out of alignment with your target audiences and/or your company’s strategy. If you don’t have checks and guardrails built into your AI-optimized process, you may not see the problems until late in your project or, worse, after launch.
AI is not an approach, it is a tool. A tools designed to recognize patterns in language (that’s what LLMs do, after all) should not be entrusted with important, bottom-line decisions about your company’s digital presence, or how people and search engines should discover your content. It shouldn’t decide what concepts or content should be prioritized on your website. It shouldn’t make the final decision about whether you should have a flat, hierarchical, or polyhierarchical classification taxonomy and what terms should be in it. It simply doesn’t know what humans know.
When the aim is to create a usable digital experience and make content findable, AI tools—as any tools—should be wielded responsibly by humans who understand the context, the content, and the audiences involved. Hint: You’re not actually solving findability problems by merely sticking a chatbot on a website and hoping users can find what they need.
Using AI for Findability
What’s the ideal approach to creating navigation and taxonomy with the help of AI? The exact steps in the process can vary, but in the tradition of user-centered design (UCD), it should involve subject matter experts who lend their insights and test your approach with real content and real-world scenarios. It should involve audience research with your end users, to gather insights early and later validate your plans.
There should always be humans in the loop—project leaders, internal experts, and external stakeholders. The humans and their expertise in the business context are your safety net. Technology tools should never make decisions when your company is responsible for the outcomes.
Your AI governance process for any kind of digital design or strategy project should include—
your prior knowledge (without making any sensitive information public)
audience insights (while guarding their privacy and your competitive advantage)
a series of focused prompts that allow you to iterate toward your desired outcome
validation checks, including source citations (where did the AI tool get its information, and can it be verified?)
testing with real humans—business stakeholders and your target audiences
(If you need help with AI prompts or digital governance of any kind, we’re happy to assist.)
The approach to achieving findability is nothing new, only the tools are. But when so many organizations are putting all their trust in tools—adding AI and hoping for the best—it’s a good time to step back and remind ourselves that navigation and taxonomy, like all aspects of user experience design, need to be intentional and tailored to their audiences in order to work.
Want to talk about making your content findable, or AI governance for communications? Get in touch.
Photo: A physical card sorting exercise to get subject matter experts’ ideas on organizing content. Thanks to our clients at AAMVA.
