Has Google’s A.I. Penalized Itself For “Search Engine”?

30
Mar

While performing a UX review of my.reach.ai private beta (which is due for a major release in a couple of months), I noticed Google.co.uk has been quite visibly struggling to rank the UK for the query: “Search Engine”. Having stumbled across this quite by chance, it’s raised a few questions arounds the ranking algorithms or perhaps they’ve been hit by penguin?

The graph shows Google.co.uk taking it’s second dive this month… it’s now outranked by twitter & search engine land

Google UK, query Search Engine

If you think that’s odd, take a look at Bing.com below…

Google vs Bing - UK results for: Search Engine

Bing.com has also taken a hit for the query: “Search Engine”.

As I mentioned, I only noticed this yesterday and following on from Nate Dame’s post on Search Engine Land this raised a question about Google using AI to improve spam detection.

Having spent the last 3 years gathering data and investing in technology in order to teach A.I. to identify filter thresholds, Nate’s article was of particular interest.

A supervised machine learning algorithm requires input data, in the assumed case of Google, this equates to confirmed cases of spam, which are used as data points to train a learning algorithm. (more on Machine Learning)

False positives are more likely to occur in the early stages of training but over each iteration its accuracy improves. The Perceptron Convergence Theory states that:-

If there is a set of weights that correctly classify the ( linearly separable ) training patterns, then the learning algorithm will find one such weight set, w* in a finite number of iterations.

Looking at this case in more detail, we can see Google has been struggling to rank itself for some time, bouncing between two canonical variations of the homepage (http and https) and an SEO starter’s guide PDF.

11th March – Position 2 – URL: http://www.google.co.uk

search_engine_-_www_google_co_uk_-_UK_-_STOP__Blind_Panic_Report_-_my_reach_ai_-_reach_ai_Limited

16th March – Position 15 – URL: https://www.google.co.uk

search_engine_-_www_google_co_uk_-_UK_-_STOP__Blind_Panic_Report_-_my_reach_ai_-_reach_ai_Limited_3

26th March – Position 20 – URL: https://www.google.co.uk/webmasters/docs/search-engine-optimization-starter-guide.pdf

search_engine_-_www_google_co_uk_-_UK_-_STOP__Blind_Panic_Report_-_my_reach_ai_-_reach_ai_Limited_2

29th March – Position 24 – URL: https://www.google.co.uk

search_engine_-_www_google_co_uk_-_UK_-_STOP__Blind_Panic_Report_-_my_reach_ai_-_reach_ai_Limited_1

In the above chart we can clearly see the last drop on March 29th occurs when the green band indicating daily standard deviation begins to expand in thickness. This indicates a large volume of variance within the SERPs, usually indicating an algorithmic update.

When I observe a drop as notable as this either during or in particular at the beginning of a period of high deviation. This would ordinarily indicate that the latest update has classified the site in question as failing to meet the updated standards. Or in other words… you’re busted.

Either way, something isn’t right here…

Perhaps Mr Cutts or Pierre Far could offer some insight?

PS… Claim one of 20 more invites for my.reach.ai private beta