Ah ok, that makes it a bit more solid.

edit: couldn't resist temptation to play around with it :)

I see how it works now, that's nifty. So, basically you're taking the search query, you retrieve the search terms and then you fish up a bunch of urls on the local server to try to capture the user longer.

This should have a measurable effect on time-on-site and bounce-rate stats.

You should do an A/B test for sets of pages with and without the widget to see if those variables change measurably (outside of statistical noise range).

That way you can get yourself a unique selling point.