I was in Munich for the past two days: yesterday, Bing showed us upcoming features and changes in the Allianz-Arena and today, Stefan celebrated the 10th anniversary of suchmaschinentricks.de with about 300 guests (and thanks to vigorous support from Bing). Having been a forum-user of the first hour, I was happy to be do a presentation together with Uwe Tippmann from Abakus on the subject of backlink-evaluation. After the presentation, there was the request from the audience to publish the numbers and the ideas behind them and so I would like to do just that in this post.

My presentation took on the subject of link-evaluation-factors and what SEO's do with them. The word “Trust” that we use often originated from the paper “Combating Web Spam with TrustRank”. In it, Google explains how to recognize spam, based on a manually selected seed-list of trustworthy domains. This can be roughly explained in that they take the click-interval between URLs on the seed-list and those URLs that they are checking for. Many assume that there are many .edu-top-level-domains on that list – seeing how these are some of the oldest and most trustworthy domains on the Internet.

Additionally, there's the public information. A surprising number of SEO's generalize this fact to the end that they believe all .edu-domains will generally have a high trust in Google and so they will put a lot of energy and money into acquiring these obscure links – the main thing is that they come from a .edu-domain. In our database, I found 4.824 .edu-domains. I then took the IP-popularity, which is the amount of different IP-addresses that have links pointing to that domain, as a rough indicator and was able to show that only 472 of those domains have an IP-popularity of more than 1.000 – which is less than 10%. Here a slide to make this even clearer:



For my second example, I looked at the composition of the backlink-structure. In the patent “Methods and systems for identifying manipulated articles“ (US Patent 7,302,645), Google depicts how stark deviations of certain signals from the norm can be used to pinpoint unnatural backlink-profiles. This should be general-knowledge for SEO's and there is probably no SEO that would disagree with these remarks. If we take a closer look at some backlink-structure though, it becomes apparent that objective and reality are not always exactly on the same page. This example shows a domain that has more than 1 million backlinks – but all of those are from only 11 different IP-addresses. Here we might not have as many links, but every link comes from exactly one domain. Natural-growth looks different, which should have hopefully been nicely demonstrated by the base-numbers forspiegel.de.

Uwe showed an analysis of a backlink-set of more than 10 million links and presented us with some interesting averaged values, which also continued the second part of my presentation in more detail. I was under the impression that most of the attendees were pleased with the event and I am looking forward to the 20th anniversary.