His point is that there is a gap between actual search functionality and people's understanding of it. Users expect to be able to type in something similar to how they would describe a product to a store clerk (e.g. "I need a pink, impact resistant iphone 5 cover."). But search fields don't understand language patterns or sentence meaning. Search fields work off key words, ranked by priority. These key words match item titles, descriptions, tags, categories, etc. It also isn't capable of doing word substitution, so it may not know you meant "case" instead of "cover". So if "cover" doesn't appear in the title or description, the search won't rank it that highly (if it's a "case").

In this case, putting the query as something like "iPhone 5 case pink" would have probably returned results that were more aligned with what the user was looking for, but most people aren't going to think that way.

When you understand how a search input actually works - text input, query statements, database structures, ranking algorithms, etc. - you think about searches in a completely different way. Most people don't understand these things, so they think about searches in terms of what they would ask an actual person for.