On being relative: rel=”nofollow” and semantic drift

For those that may have missed it, Google weighed in a week ago with a proposal to reduce comment spam. The proposal says that search engines should not include links qualified with the rel="nofollow" attribute in the page rank calculations for their destinations—a move done in concert with MSN, Yahoo, Six Apart, Dave Winer, UserLand, Technorati, and all the other major blogging platform providers. Basically, the move is designed to remove the incentive for comment spam by insuring that URLs entered by anyone other than the blog’s author won’t get a boost in rank, and that search engines won’t follow and index the page marked with the nofollow link (unless, of course, it was linked elsewhere on the web with an unornamented link).

Pretty cool, potentially, as a solution. As discussed, it provides a neat way for blog tool providers to implement the recommendation automatically, by automatically “neutering” links entered by guests and commenters. And it uses an existing part of the HTML machinery, the rel attribute, to do the job.

Except… if you look at the allowed list of types in recent HTML specs (4.01 and 4.0), you don’t see nofollow anywhere. What you see is a list of relations that include both structural instructions to the browser (e.g. alternate and stylesheet), navigation instructions (e.g. start, next, prev), and part-of instructions that show how the link relates to the text (e.g. appendix, glossary, contents). The spec asks you to include a profile declaration in your <head> tag if you’re going to extend the list.

I’d like to see Google provide the canonical profile document for this usage, because it represents a new semantic category: search engine instructions. I don’t have a problem with that per se, since the intent of this link type is that it extends the weak protection afforded by robots.txt for the document into a stronger protection inside the content itself. (Note that the other emergent uses for the rel attribute, including Vote Links and tag links, also represent new semantic spaces that need profiles, and the Technorati developers have provided a Vote Links profile.)

See also Rogers Cadenhead’s contrarian take on the proposal.

(Updated 25 Jan to link to the official UserLand announcement of support for nofollow.)