Does Backlink Quality Matter?

Posted by Richard Stokes on April 20, 2010 to Link Building

Backlinks — incoming links to a website or page — are a topic of primary importance to search engine marketers. Incoming links not only bring visitors. They also help search engines to measure the quality and authority of a particular page, which in turn plays an important role in determining which pages are served (and in what order) on the results pages for search queries.

The number of backlinks is considered by many search engine marketers to be of primary importance in this regard. However, many experts hold that backlinks vary in their ability to influence search engine rankings. In other words, a backlink from a high quality page is theoretically worth more than one from a low quality page.

There have been various efforts to model "link quality". One of the most well known is Brin and Page's Pagerank algorithm, in which they explicitly stated:

Other mathematical approaches include:

The PageRank Citation Ranking: Bringing Order to the Web
Authoritative Sources in a Hyperlinked Environment [PDF]
Hilltop: A Search Engine based on Expert Documents
Combating Web Spam with TrustRank [PDF]

While these techniques have certainly been steps in the right direction, there is no denying the evidence that linkspammers are still able to rank well for desired terms. This has led to a widespread belief by some search practitioners that high quality pages can be easily faked. This, in turn, has led to a cottage industry which churns out splogs (fake blogs), comment spam, forum spam, and other various linking schemes with the intent of working around the well-known link scoring algorithms cited above.

It is our belief based on years of experience that the search engines - while far from perfect - have progressed farther than is widely believed in their ability to identify and discount spam links.

If true, then there should be some practical way to measure this effect. In April, 2010 we set out to determine whether high quality backlinks have more influence on search engine rankings than low quality backlinks. The answer has profound implications for search marketers. For every worthwhile backlink, there are thousands of low quality ones. Those who blindly pursue every available link without regard to quality incur high - and unnecessary - costs in doing so (even while raising the risk profile of their site.)

The parameters of our study was as follows:

• We recorded the top 1,000 organic rankings on Google US for each of 1,056 websites.
• These rankings were combined with current keyword search traffic estimates as well as standardized clickthrough rates to generate a measure of each site's prominence in the organic search rankings. This measure is known as Domain Strength. Domain Strength is measured on a logarithmic scale from 0 to 100, where 0 corresponds to 100 visits per month while 100 corresponds to 100 million visits per month.
• We then crawled a sample of 1,353,307 backlinks to these domains and assessed each on four scales: Trust, Spam, Social, Geo, and Other. The Trust and Spam scales measure high and low quality links respectively. Social signals measure links from popular social networking sites. Geo signals represent links originating from non-US domains (such as .com and .us) All remaining links are grouped in the "other" category.
• Finally, we mathematically analyzed the correlation between these different signal types and the prominence of each site.

Backlink Authority Study Results

The results of our study show that trust and spam links differ dramatically in their ability to influence search engine rankings. Across sites of all sizes, authority, and verticals, a trust link carries a minimum of 4.2 times more weight than a spam link. This estimate is a lower bound. In practice, a trust link is worth far more, especially for sites who have a well-established number of inbound backlinks.

We were surprised to learn that links from social networking sites have an even higher ability to influence search engine rankings. A social link carries 8.9 times more weight than a spam link and 2.1 times more weight than a trust link. However, this effect appears to diminish with higher traffic sites.

It is also important to note that links from social networking sites are transitory by nature. While they can boost search engine rankings, the effect is often short lived as the links age and disappear. In this respect, trust links are far more useful in that they appear to convey a lasting, long-term benefit to their target sites.

We also noted that spam links carried about the same weight as links that did not rate highly on any of the other measures ("Other" links). The difference between these types of links lies primarily in their risk profile. While both types of link carry similar weight, spam links are very likely to be discounted at some future date by the search engines. When this happens, sites who rely heavily on them will likely experience a significant reduction in rankings (and traffic).

Table of results

The following table shows how each signal type correlates to Domain Strength (a measure of a site's prominence in search engine rankings.) P-Value is a measure of the statistical validity of a particular variable. Simplistically speaking, a p-value of less than .01 indicates high confidence.

Limitations

• While generally true for all sites, the exact ratios and weights of various types of links vary enormously depending on the size of the site and the total number of pre-existing inbound links that a site may have. In particular, we noticed that spammy links tend to benefit small sites far more than large ones and that the law of diminishing returns is in effect.
• The techniques which we use to score links in the various categories are obviously proprietary (e.g. they comprise the fundamental algorithm behind our backlinks product.) Specifically, our spam scoring model is influenced heavily by Quality Bot, a learning algorithm which was designed to detect pages which search engine users would deem as "low quality". Nevertheless, our models are based in no small part upon well-known information retrieval spam detection algorithms.
• Due to the US-centric nature of our sample set, we were unable to measure the effect of "Geo" links.
• Relevance of a backlink to its target page was not taken into consideration. Backlinks to topically similar pages almost certainly have higher weights, but we did not attempt to measure this.
• The ratio of trust to spam is much higher in the real-world due to the fact that our algorithms filter the majority of spam links from consideration prior to crawling and scoring. If these links were included in the study, the influence of any single spam link would have been considerably lower.