Search features I depend on:
* Non-English characters. Handle multiple encodings of web pages and URL-encoded characters in search queries.
* site: to search only within a domain. This is often a national domain, such as "site:co.uk" to search only British sites.
* Minus: Begin able to block certain words, or sites.
* Plus: A word prefixed with a plus is required.
* Quotes/hyphen: Searching for exact phrases. "Java class file" is different from "Java File class".
Where current search engines are lacking:
* If there is a period between the words then they do not belong to the same phrase. (A search for "Hello Google" should not return "Say Hello. Google for it." as its top result)
* Use word order in search query to weigh how important a search term is. Rank pages higher wihen those words are closer together.
* Don't correct my spelling by default, assuming that my search query is in US-English. (I am speaking to you Duck-Duck Goo!). I can spell, and I do not always write English. If I misspell then that is my mistake, and sometimes I search for a brand name that was misspelled intentionally.
* When indexing a web page, identify what is the important text on the page and ignore the rest. For instance, on an internet news site, the text in the articles is most important. On a forum text inside the comments. On this forum, articles followed by comments. What people have written in their signatures is not important. Slashboxes are not and ads are definitely not.
It is aggravating when you use Google on a collecting site and you get every other page on that site in every search result because members have listed their collections in their signatures.
If I search for the word "review", I don't want every page on every web store that has a Reviews tab.
Pages on a site often follow a certain pattern - find that pattern to find which text on each page that is the most unique.