A page needs to have a point: Like any web site, a SharePoint 2010 WCM site will consist of multiple sections. Most likely sibling and child sites will be used to form these sections of the overall site. These sibling and child sites are containers for pages. Pages will likely be based on an article page layout. Article page layouts provide the means to format and display articles cleanly on a page. You should ensure content authors understand the importance of how their written copy relates to the page it will reside on. Basically, try to avoid publishing an article about “Apples” on a page titled “Oranges”. It is perfectly acceptable, and even recommended, to link to the “Oranges” page containing articles about “Oranges” from the “Apples” page, or visa-versa.
Title: If the SharePoint 2010 WCM page is based on the article page layout, which it most likely will be, you can modify the “Title” by clicking Site Action / Edit Page. Ensure the title of the page is short and to the point. If business rules force you to include the organization name in all page titles, do so at the end of the real title and separate the two with a dash “-“. Example: “Apples – Fruit Growers Association of America”. Page titles should rarely if ever change.
Description: The page description should clearly summarize what the page is about. A well planned and executed page will contain the title within its one or two short sentence description. The page content should support the page description, again containing the page title and some if not all words from the page description. Page descriptions should rarely if ever change.
Should We Scrap the “keywords” html meta data? (your call): The entire notion of the html based keywords meta tag sadly proved to be a catastrophic failure. Although the concept is solid enough, the application ended up being abused so badly that search engines could no longer base any level of weighted relevancy on the keywords value. Search engines consist of pretty sophisticated algorithms which among other things determine the precise keyword density of a page. Would it then double back and see what the human wanted it to think? This one’s your call.