So you implemented autocomplete in your search box, but suggestions are off-topic? Or people complain the suggestions are irrelevant? It’s not the user’s fault, even though the average internet user puts relatively small efforts in accuracy when typing.
Autocomplete suggestions speed up searches, since users have to type less when suggestions are accurate. But the autocomplete algorithm then has to understand what the user is searching for, even when the first few characters contain typos.
TellusR’s autocomplete suggester analyzes continually using modern techniques like n-grams, and both prefixes and complete words are scrutinized in and out of order. This «complete autopsy» of the search phrase ensures you get relevant suggestions even for queries with several misspellings.
You can try out the autocomplete suggester in the TellusR/Central GUI and make sure your configuration works on your data before you release it to your live production search box. Then lean back and see the search re-vitalize. A typical effect of improved autocomplete suggesting is both a decrease in search time and an widening of the spread of visited pages on your site.