rel="nofollow" links

The major search engines Google, Yahoo, and MSN Search as well as several blog vendors/communities like Six Apart have announced support for a link attribute specifically meant to combat comment spamming.

The idea is to add rel attributes with the value nofollow to untrusted, user-submitted links (such as the links in blog comments or guestbooks), for example:

<a href="http://www.example.com/" rel="nofollow">example</a>

Search engines are then to ignore such links in page rank calculations, reducing the incentive for comment spamming. Similar mechanisms have been proposed by many web authors, particularly bloggers. Unlike some previous proposals, this is valid HTML and XHTML markup: rel attributes are already allowed in anchors (the a element).

I'd like to share some observations about this initiative:

• The rel attribute is supposed to describe the relationship from the current document to the anchor specified by the href attribute. nofollow does not describe a relationship, but rather asserts that no relationship is recognized for a certain class of user agents (namely search engines).
• It's unlikely to have much of an effect on the amount of spamming. Experience with email spam has shown that spammers will keep at it even if the perceived benefits are miniscule, because the costs are even smaller.
• It will get abused to influence page rank in ways not originally intended. Various limitations in page relevancy calculations (like the way redirects are handled) are already being widely exploited, and this is an officially sanctioned way to manipulate page rank.
• We should expect a fairly dramatic redistribution of page rank for certain kinds of pages. Decreased page rank for spammed sites might not be the most noticeable change.

In Google's Quality Guidelines the first basic principle is make pages for users, not for search engines. This is sound advice for several reasons, but perhaps most importantly it is because the search engine exist to assist users in finding content. In order to do so a search engine's model of the content needs to match that of the users, which is impossible if the search engine spider is unable to view pages like human visitors do. The rel="nofollow" initiative violates that principle.

I'm open to describing relationship semantics with rel, but links always imply some kind of relationship; to break the relationship, remove the link. A link is not necessarily an endorsement, and page relevancy calculations need to take this fact into account. But littering the web with explicit instructions to ignore certain links isn't productive. It does not solve anything in the short term, nor does it help search engines in the long term.

21 January, 2005