Whilst a relatively small study, for the time-poor/less patient, here are the top five key takeaways:
- This test needs to be run under many different conditions (and many more times) for these findings to be really exhaustative.
If you are only going to read to here – please skip onto the “limitations” below to at least understand where the gaps are here. Whilst we’re confident we can ask some pretty worthwhile questions off the back of the study, it would be a bit premature to come to any solid conclusions at this stage.
If you were relying on weekly, or daily rank tracking – the “norm” – testing the speed element here may prove challenging because of the inaccuracy – give or take a day here isn’t quite good enough! However, since recently building a UI for our experimental hourly tracking (more below), it seemed like a great opportunity to see if we could get a more accurate measure of how different it may be.
More on Hourly Rank Tracking
We have been running this test off the back of hourly rank tracking, something that my colleague Simon and I have been working on and off for nearly a year now. It is pretty much as the name suggests, and has been developed more as an experiment – we’re starting to realise some of the potential behind this frequency of tracking results now and are opening it up to a very limited number of people – you can read more about it here.
To do this should be straightforward:
- Create two identical new pages
- Each containing 250-300 words of unique copy, 1x H1, 1x <h2> + optimised title
- Page #1 has the content in the source – visible on page load
- Both pages published (but orphaned from main navigation) at the same time
- Both pages fetched and rendered by Googlebot (and crawl requested) at the same time
- Monitor both keywords hourly and observe the time-difference between the two
This process needs to be repeated for each of the sites in question, we’ve picked three domains which we own, are ‘clean’ of legacy SEO and – most importantly – would not cause a problem if the test went wrong.
Here are they ran through Majestic:
Again, to make the test “fair,” all the following were common through each:
- All WordPress
- All low-authority
- Fictional content/keywords – selected because of no exact match results
Below is the tag we used to implement the content (triggered on the test page only):
We ran through this process with all six of the pages (three tests – three controls across three domains).
It’s easier if we review each test result by the domain we tested it on. If nothing else, it will help highlight how each one reacted differently.
Control: content was indexed within the hour of it being fetched and rendered. Note the 8th – 11th of February; the server, which runs the tracking (half) restarted which meant we lost data across the board across everything during that time.
This seemed like a far bigger time-delay than we’d have imagined – two weeks! Yet, something seemed wrong. On further inspection (thanks to Kieran of The Web Shed), there was a mixed content error on the page (rookie mistake!), which was brought about by a WordPress plugin that needed updating.
As soon as I fixed the error, I ran another indexing request and the page was ranked within the hour (second line below).
So, in this instance, the unsafe scripts appeared to have stopped Google from indexing the content. As soon as we fixed the error, it was indexed.
Control: As per the first test, the content was indexed within the hour.
With another indexing request, the content was picked up again and also within the hour.
It is still ranking now and we’ve not been able to recreate this.
Control: As per the first two tests, the content was indexed within the hour.
The main limitation here is that the study itself has been very small in size and there’s a lot we haven’t tested for.
This has been the intention in some respects – we’ve wanted to control for as many variables as we can (a difficult task in SEO!), but it does mean we need to be careful with how sweeping we are with the findings.
For final reference, here are some other limitations of the study we’ve identified which we want to outline – in the event anyone else wants to plug some gaps.
- All test sites were WordPress, but other platforms could introduce further variables, which could change the results. Ecommerce sites with crawl budget issues may find these reliability challenges are even worse.
- All sites used are relatively low in TrustFlow/CitationFlow, although we haven’t tested for how authority may impact this (if at all).
- GTM was used as a deployment method, but there are many different ways of doing this; some potentially more stable, others far less.
- Requesting indexing via Google Search Console is convenient for testing, but is likely a poor proxy for crawling/indexing “in the wild”, undoubtedly an area we need to test more thoroughly.
We need to test further to duplicate the results above across more different websites/ scenarios. What we could do with is more people participating or running their own versions of this – ultimately to try and disprove what we’ve said here. Anyone who is inclined to do so, contact me on Twitter, you don’t need to user hourly to run the testing, but you can always register your interest here if you want to be let onto the platform soon.
Please let me know your thoughts; do you see any glaring flaws in our methodology, want to debate any findings, or have some ideas for next time? Twitter’s best to catch me on, otherwise email also suits me, chris ‘at’ strategiq.co