BMW Germany February 2006 · Manual action
Doorway pages, and the domain left the index
What was donebmw.de served search engines a keyword-stuffed text page and redirected human visitors to a Flash site. Matt Cutts announced publicly that the domain had been removed from Google’s index.
What was rightThe response. BMW removed the pages quickly, said what had happened, and the site was reinstated within days rather than months.
What went wrongThe technique itself: showing one thing to a crawler and another to a person is cloaking, and it has been against the guidelines since long before 2006. A brand of that size had no reason to need it.
What should have happenedServe one version to everybody. If your important content only exists inside a plugin or a script, the fix is to publish the content as text, not to build a shadow page for robots.
What the right plan does to the economicsOur own arithmetic on what a removal costs: a site earning 200,000 organic visits a month at a 2% conversion rate and 180 dollars average order loses roughly 720,000 dollars of monthly revenue while de-indexed. Four days out is about 96,000. The doorway pages were built to save a redesign that would have cost a fraction of that.
Google Search Central blog, «Ramping up on international webspam», February 2006 · revenue arithmetic ours, on a modelled site J.C. Penney February 2011 · Link scheme
Number one for everything, until the New York Times counted the links
What was doneJ.C. Penney ranked first for a very wide set of commercial terms. The New York Times investigated and found thousands of paid links pointing at the site from unrelated pages.
What was rightNothing about the tactic. What is instructive is how visible it was: the pattern was findable by a journalist with a backlink tool, which tells you how findable it was to Google.
What went wrongThe links were bought at scale from sites with no topical relationship. Rankings that arrive faster than the content improves are the signal, and somebody inside chose not to ask.
What should have happenedJudge a placement by whether real people read the page it sits on. If the page has no readers, the link is decoration with a penalty attached. Mark anything paid as sponsored.
What the right plan does to the economicsOur own arithmetic: after the manual action the site fell from position 1 to around position 70 on its main terms. At a 28% click share for position 1 against under 1% past page five, a term worth 50,000 monthly searches drops from roughly 14,000 clicks to under 500. Across a portfolio of such terms that is the whole channel, and rebuilding trust takes quarters, not weeks.
The New York Times, «The Dirty Little Secrets of Search», 12 February 2011 · click share and revenue arithmetic ours Any content site ongoing · Structure
Fifteen pages, and two of them nothing points at
What was doneA site grows to fifteen pages over a few months. Everything is linked from the header and footer, so every page shows healthy inbound link counts in any crawler.
What was rightThe publishing itself was disciplined and the pages are individually good. This is not a quality problem, which is exactly why it goes unnoticed.
What went wrongCounting all inbound links hides the only ones that carry meaning. Template links appear identically on every page, so they cannot say that one page is relevant to another.
What should have happenedStrip header, footer and navigation before counting. Anything left is a real signal. Fix pages with zero by adding links from inside sentences on pages that have a genuine reason to mention them.
What the right plan does to the economicsWe ran this on our own five sites. An average page had 8.2 inbound internal links, of which 5.6 were template and only 2.6 came from inside body text. Two pages of fifteen had zero contextual inbound links, including our own blog index, which showed fifteen inbound links in any standard tool. The fix was two sentences on two pages and cost nothing.
Our own measurement across five sites with tools/linkgraph.py, method described in the piece
Events and dates under a company name come from the source listed beneath that case.
Revenue and click figures are our own arithmetic on modelled sites and are labelled as such.