Rand Fishkin still doesn’t understand NoFollow

UPDATE: Rand let me know he had revised his post to clear up the ambiguity about the effect of using rel=’nofollow’ in his remark.

I hate doing this but enough people follow SEOmoz that if Rand Fishkin posts something wrong, the chances are pretty good that a lot of people will start repeating the erroneous idea and then we have yet another war against SEO ignorance to wage.

Begin Emily Latilla Voice – Never Mind

Today Rand wrote: “Nofollowing a link from a page makes it pass more juice and ranking power through the links that are followed.”

Really? So, what we should infer from this remark is that of three possible permutations, only the third (in the list given below) will produce optimum outbound PageRank for your links:

1. You place six followable outbound links on your page
2. You place five followable outbound links on your page
3. You place five followable outbound links on your page and one NOfollowed outbound link on your page

If we’re to believe Rand, option 3 gives your five outbound links more juice than option 2. Is that really what he meant to say? I hope not.

Using rel=’nofollow’ on some of your outbound links in no way gives more juice or power to your other outbound links. All it does is tell the search engines that honor nofollow to not count those links when evaluating links. Masking outbound links from search engines is indeed, as Rand points out, an old black hat SEO trick. It’s also about as effective as pouring a glass of clear water into the ocean.

There are many billions of Web documents in the major search engines’ indexes. Some of those documents pass PageRank-like value to other document. Some don’t. In the overall scheme of things, you’re not in any position to influence PageRank-like value distributions by using or not using rel=’nofollow’. You’ll do better in the long run by including valuable expressions in your meta tags.

End Emily Latilla Voice

Rand also wrote: “…to say that no one should ever [use rel='nofollow'] because it doesn’t work, or because they’re too dumb to figure out how to do it properly sounds like you’re either in denial or don’t want competition.”

Sounds to me like Rand doesn’t want to let go of a failed idea. Supposedly (I cannot find the blog post where I read this because at the time I didn’t think anyone would be bold enough to say something pro-nofollow so soon after the conference) someone asked an SMX Advanced audience (last week) how many of them had tried sculpting PageRank with NoFollow. Most people in the room raised their hands. They were then asked how many of them felt it helped. Most people in the room dropped their hands.

If you really could work some benefit from using “rel=’nofollow’”, an identifiable trend would have emerged by now. Despite the fact that people like me, Shari Thurow, and others pointed out just how much of a waste of time it would be for people to implement nofollow on internal links, many of you tried it anyway. Most of you appear to have reached the conclusion that it didn’t do anything for you.

It doesn’t take much of an intuitive leap to see that if you cannot improve your PageRank by telling search engines to ignore your own content, you’re not going to improve it by telling search engines to ignore other people’s content.

That’s called PageRank Hoarding, which won the Most Stupid SEO Idea of the Year for 2005 (PageRank Sculpting won Most Stupid SEO Idea of the Year for 2007). You cannot hoard PageRank for the same reason you cannot sculpt it: you cannot see it, measure it, or control it.

You don’t get to decide which of your pages obtain PageRank and you don’t get to decide which of your pages pass PageRank.

I think it’s also interesting that Rand put words into Matt Cutts’ mouth by writing “Matt Cutts said All Links to a Domain Help Every Page on that Domain Rank Better”.

I’ve read several reports about Matt’s comments and Matt Cutts said no such thing. Of course, Rand admits in his follow up comments that he is putting words into Matt’s mouth but that’s just morally and ethically wrong in my opinion.

In fact, statistically Rand only has about at most a small chance of being right. Matt reportedly said “The original PageRank was purely a page-level document.” (Source: SmallBusinessSEM Blog – Cf. previous link)

If Matt was implying that PageRank now works differently, here are some possible variations:

1. PageRank could be assigned on a page-zone basis
2. PageRank could be weighted on a page-zone basis
3. An Ask-style context-sensitive PageRank might have replaced the old PageRank
4. PageRank might now only be a “Main Index page-level document”
5. PageRank might be filtered by neighborhood or Web zone

I can think of many more ways to get away from the original PageRank distribution model, several more likely (based on patent applications and technical papers) than Rand’s assumption that there must some sort of domain-level PageRank.

There is no statistical evidence for domain-level PageRank assignment. Nor is there any anecdotal evidence for it. There is plenty of speculation, of course.

Can you test for domain-level PageRank? Well, first of all, you have to be able to measure PageRank. Since we have no way of measuring PageRank, we have to find it the same way that astronomers find black holes — we have to look at PageRank’s effects on the Web space around it. How do you do that?

Example 1: There is a fairly well-known wiki-style site that links out freely to other sites. I have obtained some links from that wiki-style site for various sites. Some of the links appear to pass value. Some of the links appear NOT to pass value. All of the destination sites pass value through their own links. Conclusion: During the period of observation (lasting several months), the wiki-style domain did not pass value through all of its outbound links (which were on separate pages).

This is a strong indication that no domain-level PageRank was working, but it’s not proof.

Example 2: I have several value-passing domains that all link to a specific destination (call it MyTestSite). Links on the destination page don’t appear to pass value. My value-passing domains consistently help me launch new content into the major search engines’ indexes. The new content consistently obtains the ability to pass value through its own links. But MyTestSite doesn’t pass value through its links. Why?

Conclusion: Simply pointing links from trusted domains to a new domain does not guarantee the new domain will be trusted.

This is a strong indication that no domain-level PageRank is at work, either for the linking domains or the destination domain.

Example 3: There are several popular social media sites that people use for link placement. I have created profiles on some of those social media sites to test my theory of “child inheritance” (which I discussed in On The Googleness of Being in early 2005). I believe that “child inheritance” works in a limited fashion and has more to do with trust than with PageRank. Still, if there were such a thing as domain-level PageRank, one would expect that a profile page on a PageRank-rich domain should have some minimal impact in search results.

My test profiles did not even appear in search engines’ indexes for many weeks. When they eventually did, their links accomplished nothing. Now, more than a year later, those links still accomplish nothing.

Conclusion: Simply creating pages with outbound links on link-rich sites does not create any value for you.

This is a strong indication that no domain-level PageRank is at work, but it’s not proof.

However, if we assume for the sake of discussion that Google does have some sort of domain-level PageRank, they don’t have to use it for weighting links or adjusting relevance scores in search results. That is, a domain-level PageRank-like mechanism could be used to establish trust thresholds, to identify neighborhoods, to create filtering processes that help Google figure out where Web spam is, where new fountains of legitimate content have sprung up, etc. That is, by mapping relationships between domains, Google can analyze the Web without actually using that analysis to directly impact its linking valuations or search results.

Some people actually crawl the Web and collect terabytes of data without publishing that data. These home-crawl indexes are neither professionally designed nor professionally managed. These home-crawl indexes are used for “analysis” that is, in my opinion, shoddy at best.

One of the most common mistakes people make with their home-crawl analysis is that they treat their database as if it is similar to the major search engines’ database. This is about as stupid an idea as using Yahoo! SiteExplorer to do Google link analysis.

Another common mistake is that the home-crawl analysts don’t refresh their snapshots. Since they don’t have the resources to recrawl the Web once a month, they just lump all their old Web site images together with their most recent Web site images. What they end up analyzing is an Elephant Man metaphor for the Web that bears no resemblance to the real Web.

Some home-crawl analysts try to duplicate seed-based crawling. That is, they may start crawling the Web from Yahoo!’s directory or the Open Directory Project in the hope of duplicating the major search engines’ crawl patterns. However, without access to the major search engines’ filters, penalty lists, and ban lists the home-crawl crowd cannot help but index radically different versions of the Web from what the major search engines index.

Now, doing your own crawl is not necessarily a bad thing as long as you don’t try to second-guess what the major search engines are doing based on your own experience. If you become so bloated with home-crawl arrogance that you start making sweeping generalizations on the basis of your unsubstantiated analysis, you risk saying some pretty stupid things.

I seriously doubt we’ll ever get the SEO pundits to stop advocating the use of “rel=’nofollow’”. Every time someone claims to have accomplished real results with it, desperate people in the audience will go away and think, “If it worked for him it can work for me”.

The truth is that if you’ve concluded that using “rel=’nofollow’” has somehow helped you, then you’ve missed something BIG. Nofollow is not a solution. It’s just a band-aid that covers up a problem.