What It Is
Google recently released its Google Instant search product. It is basically a recommendation engine on steroids. The way it works is that, as you type, search terms are recommended; starting with the first letter. For instance, typing "W" will present you with the weather forecast based on your geographic location. That's because it assumes the word you are typing is "weather." This is indicated by gray text filled in after what you are typing. If you continue typing, the predictive text is updated as well as the results presented. As an example, if you continue to type "ork," Google predicts you are typing "Work of Art" and presents you with results about the Bravo Network reality show.
The official Google video is a pretty good introduction:
Why do I say this is backing into semantics? Well, two key challenges to semantic search are understanding context and disambiguation. Normally, this happens through complex language processing. In Google Instant, it happens on the fly.
As far as context goes, Google knows when and where you are typing the query. Try typing "quizn" into the search box and you're presented with a map of Quiznos restaurants in your area. Type "linds" and you're presented with some recent news articles about Lindsay Lohan. Even though it's not much info, knowing where and when you're asking the question lets Google take a guess at the answers.
The disambiguation point is a little more subtle. In traditional semantic search, the technology does language processing of not just the question you asked, but the answers it knows. It will base its assumptions on the surrounding text whether you meant "Turkey" as an animal, a country, or a word to follow "Jive." I've seen some pretty impressive semantic search engines in action and they are monsters of indexing. They understand that "J2EE" is a term related to "Java" simply due to their proximity in the content indexed. So, if you then type "J2EE" into the search engine, it will return results containing "Java" just because it has seen them paired a number of times. Google circumvents this analytical complexity by using real-time feedback. When you type "turk," the search engine begins to show you results based on a number of meanings as well as prompt you for more specific queries it has seen. So, as you type Google will ask, "Did you mean X, Y, or Z?" If you meant neither X, Y, nor Z, you'll continue typing and Google will refine its feedback as you refine your query. Disambiguation is achieved without complex ontological processing.
What It Means
Traditional keyword strategies are founded on the principle that you can anticipate what the user will type when looking for whatever it is you provide. You look at your content, product, or service and place your bets on a coherent constellation of keywords that will drive traffic to your web properties. Then, you test for success by seeing how high you reach in Google's page rank algorithm. If you're not getting the results you want, you adjust your keywords, and so on, and so on...
With Google Instant, it's entirely possible that when Google asks, "Did you mean X, Y, or Z?" it will lead the users away from the keywords upon which you've placed your bets. As a result, you need to revisit your keywords and test them within the Google Instant environment. As you begin typing your keywords into the search box, you need to review the recommendations. If one of the recommendations is similar enough to your pre-determined keywords to be equivalent, then you need to swap out your current keywords for the recommended ones. Google has effectively taken control of the query language, and you need to adjust to reap the rewards.