David Ellis identified eight recurring patterns in how people seek information: starting (an initial search for relevant sources), chaining (following links or citations from one source to another), browsing (scanning topics of potential interest), differentiating (filtering sources by quality or type), monitoring (periodically checking sources for updates), extracting (systematically working through a source for specific information), verifying (checking that information is accurate), and ending (wrapping up the search). Not every search uses all eight, and they don't happen in a fixed order.
Marcia Bates observed that real searches rarely follow a single, static query from start to finish. Instead, a person's understanding evolves as they go, each new piece of information reshapes what they're looking for next. She called this berrypicking, comparing it to picking huckleberries scattered across bushes rather than gathered in one bunch: the searcher collects relevant bits one at a time across many sources, and the query itself keeps shifting along the way.
If your IA only supports a single, complete, correctly-worded query typed once: you're designing for a search that rarely happens in practice.
Monday, Starting. An initial search to survey what's available is the defining move of Ellis's starting pattern.
Tuesday, Chaining. Following a reference list from one source to others is chaining by definition.
Wednesday, Differentiating. Discarding sources based on quality or reliability is filtering, differentiating.
Thursday, Bates's berrypicking model. This is the key moment in the week. Elena's query itself shifted as her understanding grew, she didn't just find new information, she found a better question. That's berrypicking, not any single Ellis pattern.
Friday, Extracting. Working systematically through trusted sources to pull specific information out is extracting.
Thursday matters because it's easy to mistake for an error, a mentor might say "you changed your topic," when in fact this is exactly how real research works. An IA that only supports Monday's kind of search (a single clean query) would have no way to help Elena on Thursday, once her actual need had moved past what she knew to ask for at the start.Features that support an evolving query include related-content links that surface adjacent terms she hasn't thought to search yet, faceted filters that let her narrow within a broad starting category, saved search history so she can retrace her path, and suggested or auto-complete terms drawn from what other searchers with similar starting points eventually searched for. All of these help a searcher move from a vague starting point toward the specific thing they didn't know they needed until they got there.