After you publish your WebHelp project the indexer scans all the files found in the HTML output directory and creates an index file. This index contains a list of all the words found and their locations in the files in your project.
When the user accesses the search function for the first time the index file is downloaded by the user's browser. All subsequent searches during the current session then access this local copy on the user's computer.
It's important to understand that the indexer scans all the files in your output directory! This means if you add your own files to the directory before you publish they will be scanned too.
If you don't want your own files to be included in the index you must add them to the output directory after you publish your project – or upload them to your server separately so that they cannot be indexed when you publish.
Normally Help+Manual doesn't delete anything in your HTML output directory. Since every topic is a separate HTML file in WebHelp the files for deleted topics are still in the output directory when you publish a new version of the project that no longer contains these topics.
To prevent this it's a good idea to use the Delete all files in output folder function in the Publish dialog when you publish a build for distribution. See Publishing for more details on this.
Actually, you can. However, to do this you need to set up your search platform to use server-side searching based on either the PHP or ASP platform. This requires support for PHP or ASP on your server and can't be tested locally, because it is a server-only technology.
This can be combined with wildcard characters. For example, "cat*" will find catalog but not advocate.
ASP and PHP platforms:
These platforms support searches for exact phrases that are enclosed in quotes.