POST
Update: Long Term Shared Hosting Experiment
In September we published the results of our long term shared hosting experiment which generated a lot of debate in the SEO community.
The results showed that experiment websites sharing an IP address with spammy and/or 'low-quality' websites (defined as those we confirmed were using aggressive and suspicious link building strategies in adult, gambling or pharma niches) ranked lower than those hosted on a dedicated IP (which were set up on AWS).
In the initial study, we made sure to minimise outside ranking factors to more effectively measure the effect that the hosting had on the rankings by:
- Using a made up keyword that returned no results when searched in Google.
- Keeping content almost identical across all of the websites.
- Using brand new domains with no previous history or links.
- Running hourly page speed insight tests on all the sites to ensure that they were all, on average, as quick as each other.
- Running daily up-time tests to ensure that no sites went down over the course of the experiment.
- Alternating between shared and dedicated hosting sites when getting them indexed (fetching in Google Search Console from different accounts and IP addresses at different times).
- Making sure no one searched the target keyword and/or clicked any of the results for the entire duration of the experiment.
After the results were published, many SEO’s and SEO companies asked us to switch the websites hosting round (thanks to Cyrus Shepard again for observing the initial experiment and passing on requests to switch the hosts round) to further test whether it was actually this that was driving the rankings.
We agreed that this could be a good thing to test because a common critique of the study was that the keyword we made up for it, ‘hegenestio’, was too artificial to demonstrate any effects in an actual SERP.
After launching these kinds of experiments, the made up keywords essentially become real ones that could be more representative of an actual SERP. Some who read the results go on to search the keyword themselves, and click on various results.
Others will create content about the study and include the keyword in their articles.
When trying to isolate the effect that one factor can have on rankings, outside variables like this are not a good thing. This is why many SEO experiments focus on made up keywords that, before their test, return no results and are never searched.
In this case though, as others suggested, this could be used to further prove or disprove our hypothesis.
If the rankings reversed in order, it would support our hypothesis. If they didn’t, it would support the argument that something else was causing the dedicated IP address websites to rank higher than the shared IP address ones.
Switching Them Over
On 21/09/2020 our head of Dev Ops, Simon Hubbard, moved all of the experiment sites originally hosted on the shared servers to dedicated IP addresses on AWS (the same IP addresses previously hosting the dedicated sites).
He also moved all of the experiment sites originally hosted on dedicated IP addresses to the same shared hosts and IPs used by the originally shared hosting ones.
We continued tracking the rankings and can confirm that almost all of the originally shared IP websites (now on dedicated IP addresses) have started outranking the websites now on the shared IP addresses.
Our data analyst Niklas Jakobsen put together the below graph to demonstrate the changes in rankings.
Please note:
- The first grey area is where the experiment sites were still indexing.
- The first dashed line is when the initial experiment officially ended.
- The second grey line is when we switched the hosting round (we moved the originally shared hosting sites to dedicated IPs, and the originally dedicate IP sites to shared hosting).
- The final grey area is when the recent core update was rolling out.
The graph above shows the average ranking of both the shared IP and dedicated IP (AWS) sites. You can see the second grey line showing where we changed the hosting over meaning, at this point, the green line now actually shows the shared hosting sites and the red one shows the dedicated IP sites.
As the graph shows, the experiment sites now on a dedicated IP address gradually outranked those now on the shared ones. In both versions of the experiment, the sites sharing an IP address with low-quality and potentially toxic domains ranked lower than those hosted on a dedicated IP.
Considering we didn’t change any other factors on the site (no content changed, none of the sites picked up any links, the speed and uptime of each one remained consistent), these changes further support our initial findings.
It is important to note that the increased searches of the keyword and behaviour generated as a result of that could also have influenced the second set of results.
Google Search Console Data
We checked the Google Search Console properties of the experiment domains and can confirm that, on average, the now shared hosting sites recevied slightly more organic clicks and impressions over the past 16 months (because they were ranking on the first page after the inital publication of our results when they were still hosted on their original dedicated IPs).
The shared hosting sites actually had a slight edge when it came to Total Download Size (bytes), speed and crawl rates remained near enough equal across all of the experiment sites.
Effects of the December 2020 Core Update
Interestingly, all of the sites reacted strangely to the recent core update. For the few days that the update was rolling out, the experiment sites changed in rankings dramatically (the now shared hosting sites previously on dedicated IPs ranked top again before returning to their previous positions). There is no way of knowing why this was the case but it was interesting to watch in any case!
