For much of the past twenty years, search strategy has started with keywords. A business would identify the phrases people searched for, estimate how many searches took place each month, assess how competitive those phrases appeared to be and then produce content designed to rank for them. There was nothing inherently wrong with that approach. In many respects it remains entirely sensible. If people are actively searching for something you offer, understanding that demand still matters. Search volume remains one of the clearest indicators we have of market interest.
What has become increasingly apparent, however, is that search demand alone does not adequately describe how people explore complex subjects. Particularly in B2B markets, the path from an initial question to a commercial decision rarely follows a neat sequence of keywords. People investigate problems, compare approaches, encounter unfamiliar terminology, seek reassurance, evaluate risks and gradually build confidence before they ever become a lead. The search terms that appear inside keyword tools often represent only fragments of that wider process.
This has become even more visible as generative search products have begun to emerge. Systems such as Google’s AI Overviews or conversational search interfaces do not simply look for an exact phrase and return ten blue links. Instead, they attempt to understand the broader context of a question and then explore related subjects, supporting concepts and adjacent questions before constructing an answer. The query itself effectively expands into multiple lines of enquiry, much of which may never have been searched for before.
One of the more interesting implications of this shift is that search demand increasingly becomes topic-led rather than keyword-led. The question is no longer simply, “What do people search for?” but also, “What would somebody reasonably need to understand in order to answer this question properly?” Those two things overlap considerably, but they are not always identical. A topic can contain dozens of related questions, concepts and dependencies that traditional keyword research may only partially reveal.
This is one of the reasons I have become increasingly interested in content structures rather than individual articles. A single piece of content rarely establishes expertise in isolation. Businesses demonstrate expertise by covering subjects in sufficient depth that both people and search systems can recognise a coherent body of knowledge. Rather than publishing isolated articles, the objective becomes building clusters of connected content that reinforce one another and gradually establish a position around a particular area of expertise.
The emergence of AI-assisted search does not diminish the importance of conventional search in this process. In many respects it reinforces it. Large language models are extraordinarily capable at generating language, but they are not libraries containing a live copy of the web. When current information is required, generative engines increasingly retrieve information from traditional search indexes before synthesising responses. The quality of those responses therefore depends heavily upon the quality of the information available to retrieve.
This retrieval process is sometimes described as retrieval augmented generation, although the underlying principle is relatively straightforward. Search engines continue to identify, evaluate and organise information. Generative systems then use those sources to help construct responses. Businesses that are visible, authoritative and well structured within conventional search therefore place themselves in a stronger position to appear within AI-generated answers as well. Traditional search and generative search are not opposing systems. Increasingly, they appear to be working together.
For that reason, I remain relatively optimistic about the future of search. Much of the public discussion surrounding AI assumes that expertise becomes less valuable as machines become more capable. My suspicion is that the opposite may prove true. If systems become increasingly effective at summarising generic information, then genuinely useful expertise, practical experience and informed opinion may become even more important. The challenge is no longer simply producing information. It is contributing something worth retrieving in the first place.
