Top-down IA work focuses on the home page, section main pages, and landing pages, the structural navigation, organization schemes, and labeling that anticipate what a visitor needs the moment they arrive. It asks a specific set of questions at each level: where am I, what can I do here, is this the right place, how is this organized. Bottom-up IA work focuses on the detailed structure of individual pages, content modeling, chunking, sequencing, asking a parallel set of questions once someone has actually landed on a page: where am I now, what's here, what should I look at next.
Both are necessary, and neither substitutes for the other. A site with a beautifully organized top-level structure can still fail users if individual pages are a disorganized wall of text once they arrive. A site with perfectly structured individual pages can still fail if the overall navigation never gets them there in the first place. When stakeholders have already locked in the top-level structure before an IA gets involved, bottom-up work on page-level structure may be the only place left to add real value.
Getting someone to the right neighborhood and getting them to the right house are two different jobs, a site can nail one and completely fail the other.
The first complaint is top-down; the second is bottom-up. Guessing which top-level section a task belongs to is exactly the "where am I, is this the right place" failure that top-down work, organization schemes, section structure, labeling, exists to prevent. Scattered content and buried details on a page a user has already found is a page-level structure problem, which is what bottom-up content modeling and chunking exist to address.
These two failures require genuinely different fixes because they occur at different points in the user's journey. Reorganizing the top-level navigation categories to better match how users think about their work would help someone find the right page faster, but it would do nothing to fix what happens once they arrive, the scattered, poorly chunked content on that page would still be scattered and poorly chunked regardless of how they got there. Conversely, restructuring individual pages into clear, well-labeled chunks would make each page easier to scan once reached, but wouldn't help a user who's still lost trying to figure out which section to click into in the first place.A reasonable corrective approach: address the first complaint by revisiting the top-level organization scheme, likely through user research such as card sorting to understand how the actual audience categorizes their own tasks. Address the second by applying content modeling to the page templates, decomposing pages into well-labeled components with a consistent structure, so related information doesn't stay buried in unstructured paragraphs. Both need to happen; neither one is optional just because the other exists.