SEO Is There Top Level Domain Bias in Search Engine Indexing and Ranking?

[bhartzer]

Ever since Google's Exact Match Domain (EMD) update back in September 2012, they've always said publicly that all TLDs are created equal: all of them have an equal chance to rank in their search results. I've been skeptical--especially when the New gTLDs were launched.

So, I thought I'd test it, using several ways to look at the results: SEO Testing software, manual searches at Google, and Google Search Console data.

I chose 15 TLDs, including New gTLDs, .COM, .NET, .ORG, and a few ccTLDs as well:

• .ICU
• .TOP
• .XYZ
• .SITE
• .CLUB
• .ONLINE
• .VIP
• .COM
• .NET
• .ORG
• .DE
• .CO.UK
• .CO
• .LONDON
• .IN

The results were interesting, and I learned a lot about how Google handles indexing and ranking of brand new sites. For example, .DE and .CO domains are both considered to be in German and in Spanish (.DE in German and .CO in Spanish), despite absolutely no indication anywhere what language the site is in. No German or Spanish content and no hreflang tags on the sites.

Also, the .XYZ site got indexed much quicker than any of the other TLDs. Some were really difficult to get their pages indexed.

Each site was a unique 25 page site, 375 pages total. I chose a unique word (namescon spelled backwards) to register each domain (i.e., nocseman.icu, nocseman.com, etc.). The word "nocseman" doesn't appear anywhere on any of the sites: yet Google read the keyword in the domain name and, at the end of the test, we ended up with the .NET site ranking #1 for 'nocseman'.

Each site was also assigned a unique keyword, as well. And I followed the indexing and rankings of each site during the testing process and wrote about it all.

The highlights are all written up here if you're interested: https://www.hartzer.com/blog/tld-bias-search-engine-indexing-rankings/

I'm happy to answer any questions about the testing or the results.

 

[bhartzer]

>> I guess I was surprised that given how smart Google is at not being fooled by content that is not really content, that they got indexed as well and quickly as they did.
Not a surprise to me, recently someone ranked lorem ipsum type content in Google--and it continued to rank well despite being nonsense text. That was detailed in a Search Engine Journal article recently.

>> I agree it would be interesting to see how the Google new gTLDs that are secure spaces .page, .app and .dev would do in this.
Hopefully will be able to test those other TLDs soon.

>> In another thread someone pointed out differences in delay time for DNS access that seemed to depend on the registry backend.
Yes, I did see the Bunny CDN testing, and actually it turns out that we just proved that DNS response has nothing to do with indexing and rankings. In fact, DNS response was very slow for .XYZ in the BunnyCDN test but in fact it was #1 in our test for indexing.

>> How many other sites from your experiment ranked for "nocseman" on the same results page and in what order did they appear.
I'm not sure what you mean, but I think that you meant that for the made-up word 'nocseman', before the test how many results were there? There were only 16 results/pages that mentioned that word. It was the newspaper site and the fold3 site that had some pages indexed. All the others are new. Namepros is expected to rank because that word is used in the content and because it's overall trust and authority. Same goes for Hartzer dot com, as well. So just ignore those pages in the search results. The sites are listed and rank only because of the word in the domain name.

>> It's possible that the Google's algorithm (we better call it Google AI from now on) is smart enough to recognize that "nocseman" is backwards for NamesCon and as such associated it with Networking and found .Net to be the best fit (just guessing).
I think it's technically possible, but you're giving Google's AI too much credit. Personally I think that's highly unlikely that they see it that way. Especially nowadays that nonsense text can rank in Google, as well as lorem ipsum content, as well.

 

[branding]

I'm perfectly happy with this unbiased data set: 25 page sites and 375 pages of content total. The point was to see indexing rates and initial rankings data.

My only 'regret' here (if you can call it that) is that more TLDs weren't tested. I went with the most popular TLDs based on # of domain registrations. I fully plan on testing more TLDs in the future and comparing those results with these results.

I haven't read the full report but judging from the blog post there are some basic errors from an academic research point of view in your approach. I don't want to ruin your party but whatever the outcome of this 'study' is, it holds no value and simply cannot be used as a reference for anything SEO related. Although it does make for an interesting read.

Like I said, nice experiment. Do this for 100-1000 domains (different before the dot, with multimple tlds after the dot) and I would agree with you if that shows a certain tld gets ranked faster and better. This could be an outlier, who's to tell? One of them was bound to get indexed first.The moment that happenend that set back the ranking of every other TLD indexed afterwards. That all domains are hosted on the same IP address obviously doesn't help.

Also, if you want to experiment further, set a global language for all domains, even the ccTLDs. It's common knowlegde that unless a language is declared Google will use the ccTLD to detect the site's language. This can be of major influence on your ranking if your site is actually in english.

What would be interesting to see if you scale up this experiment to at least a hundred Keywords is if Google would look at the extention when it comes down to ngTLDs. They say it doesn't matter but at some point they will have to adjust their algorythm to make rent.homes rank righer than rent.fun based purely on the TLD when content is quite similar.

 

[bhartzer]

>> I haven't read the full report but judging from the blog post there are some basic errors from an academic research point of view in your approach.
Then I highly suggest that you read the full report and look at all the data. An overview blog post is just that--a quick highlight of the real data that's presented.

>> Do this for 100-1000 domains
More domains/sites will be tested in the future, that's the plan.

>> One of them was bound to get indexed first.The moment that happenend that set back the ranking of every other TLD indexed afterwards.
That's actually not the case! The indexing and rate of indexing was based on one unique keyword per domain name. Each domain name had its own keyword, and the rate of indexing was based on a search for that unique keyword for each domain. So having one indexed before the others had no influence whatsoever on how indexing was reported.

>> It's common knowlegde that unless a language is declared Google will use the ccTLD to detect the site's language.
It may be common knowledge to you and me (I've always suspected this to be the case) but to see it in action and proven that's helpful. So I point it out. I can understand the .DE but with so many USA-based websites using .CO now I think it's important to point this out.

>> What would be interesting to see if you scale up this experiment to at least a hundred Keywords is if Google would look at the extention when it comes down to ngTLDs.
I totally agree... I do want to scale this up and test all TLDs and not just a few domains on each TLD. The only thing holding the testing back is budget. It takes multiple hours of time to set up the proper testing and report the results.

 

[oldtimer]

Especially nowadays that nonsense text can rank in Google,

It seems like anything can rank in Google initially, but after couple of days the nonsense stuff gets removed or sent into the abyss.

Added: also even though you used a unique keyword for each site, but I am interested to know which ones ranked for the keyword in the domain itself (nocseman) and in what order did they appear on the same results page.

IMO

 

[bhartzer]

It seems like anything can rank in Google initially, but after couple of days the nonsense stuff gets removed or sent into the abyss.

Yep, that's how it works.

Google only has a certain amount of data initially, so they have to go with what data they do have. For example, it takes them some time for certain parts of the algorithm to kick in.

 

[branding]

> I haven't read the full report but judging from the blog post there are some basic errors from an academic research point of view in your approach.
Then I highly suggest that you read the full report and look at all the data. An overview blog post is just that--a quick highlight of the real data that's presented.

>> Do this for 100-1000 domains
More domains/sites will be tested in the future, that's the plan.

I would've if you had tested 100 keywords spanning, lets say, 10 tlds. Should've even included not so popular TLDs to avoid bias. To me this reads like, I listed Keyword.something in 15 different TLDs, the .net sold first and for the highest amount, therefore it's the best TLD to invest in. There's no logic in that conclusion given your dataset contains of nothing more than one sample.

I applaude you for the time and effort though. It's easy being critical. Upping the domains and generating a representative sample would be the way to go to be able to get anything out of this study.

>> One of them was bound to get indexed first.The moment that happenend that set back the ranking of every other TLD indexed afterwards.
That's actually not the case! The indexing and rate of indexing was based on one unique keyword per domain name. Each domain name had its own keyword, and the rate of indexing was based on a search for that unique keyword for each domain. So having one indexed before the others had no influence whatsoever on how indexing was reported.

Agreed. A small oversight on my side. The choice of using an unique keyword per domain does pollute your dataset though. I'm not sure how to avoid this as when all content would be exactly the same it'd be harder to track the indexing and ranking and you would probably suffer more from duplicate content penalties.

>> It's common knowlegde that unless a language is declared Google will use the ccTLD to detect the site's language.
It may be common knowledge to you and me (I've always suspected this to be the case) but to see it in action and proven that's helpful. So I point it out. I can understand the .DE but with so many USA-based websites using .CO now I think it's important to point this out.

Agreed. Especially as .co is effectively a repurposed ccTLD nowadays. Good info to take into concidertation if you operate a .co. Same goes for .io and whatever else the popular kids are using these days :P

>> What would be interesting to see if you scale up this experiment to at least a hundred Keywords is if Google would look at the extention when it comes down to ngTLDs.
I totally agree... I do want to scale this up and test all TLDs and not just a few domains on each TLD. The only thing holding the testing back is budget. It takes multiple hours of time to set up the proper testing and report the results.

I feel you. Seems like a costly adventure to reg at least 100 keywords on 15 tlds. You should be able to automate a lot of tasks but yeah, it would also add a significant workload for you to interpreted the data and report on. Buy hey, don't give up on your dream ;P

 

[bhartzer]

I would've if you had tested 100 keywords spanning, lets say, 10 tlds. Should've even included not so popular TLDs to avoid bias.

For this test, 1 keyword per domain was chosen. We didn't know which page on each site was going to get indexed first, so we had to go with 1 keyword per domain. Easier to check, and the main goal was to check indexing rates. The rankings data was a bonus.

There were some less-popular TLDs included. Future testing will certainly include a lot more TLDs and maybe these TLDs as well.

 

[Pay.My.id]

Many factors why some sites/domain fast indexed. Natural traffic based some keyword, quality of backlink ,and content quality. There are more 200 factor to do SEO well based google algorithm, personally I would like generating revenue than struggle on ranking. Good ranking never same with good revenue in some situation.

 

[bhartzer]

Many factors why some sites/domain fast indexed.

Yes, there are many factors, actually over 1,000 factors (not 200).

However, in this test, these were brand new domain names. NO links. The keywords had no prior history because they are made-up words. The content quality was, well, there was no content quality, just made-up, jibberish content.

So, the only reason could be because of the TLD. Every site had an equal chance to get indexed.

 

[Dondomainer]

That's a good question... it's only speculation. I personally don't think it's because Google's parent is .XYZ.

I may have to do with the number of domains registered on the TLD and the overall crawling activity of domains/sites on that TLD.

I think we can only speculate--I'd love to be able to come up with some sort of test to try to figure it out. But for now we do know that based on these test results they indexed the XYZ pages a lot quicker than any other TLD in the test.

We need more detail, like

domain vs tematic, #post published, html or php, #links, etc

 

[bhartzer]

We need more detail

All the detail is in the blog post and published in the PDF file that you can download for free:

https://www.hartzer.com/blog/tld-bias-search-engine-indexing-rankings/

 

[bhartzer]

Quick update... now seeing the .ICU domain rank for 'nocseman', the made-up word that is only in the domain name. Rankings have been changing, interesting to see which TLD is doing better now. Totally different than it was when the testing was initially complete and before 'nocseman' as a keyword was made public. The .net had initially "won" but that's not the case anymore.

 

[sbweb]

OK you said each site had a different keyword to rank for. Were the content topics/industries the same for all sites?

 

[bhartzer]

OK you said each site had a different keyword to rank for. Were the content topics/industries the same for all sites?

All the keywords were essentially in the same "industry"... all fictitious words. All were nonsense words.

 

[bhartzer]

I've been following the rankings of these sites (they're still live). We're back to the .net at the top of the rankings for "nocseman" keyword (which is ONLY found in the domain name):
.net
.com
.co.uk
.vip
.icu

Originally, the .net site ranked #1 really quickly... then it dropped off. But over the longer term (about 2.5 months later), the .NET is back at the top. Then .COM.

 

[bhartzer]

It's been over 6 months since the 'nocseman' test sites have been live (and indexed in Google). The current ranking are interesting, and have changed quite a bit since the domain testing was started. Now the rankings are as follows:

.com
.icu
.net
.in
.site
.top
.org

This is if you remove the pages that are about the testing itself. The made up word "nocseman" is still only in the domain name. So, looks like Google prefers the .COM keyword in the domain the best, followed by .ICU and then .NET. At least that's what it looks like after about 6 months.

 

[Jimmysun]

Lol, it’s not about bias in Google thread but bias against .xyz, I now can confirm how potential it has again.

 

[fiero]

Lots when you Google for reacher on the topic,see @JB Lions posts above.

I get what you're saying. Ranking seems to be random though.

For me it ranks XYZ, London, co.

Results will be different for other people. Once logged out it ranks XYZ club com top.

The concept of your experiment is nice but I don't think it provides usable data. The dataset is way too small to be able to draw a conclusion.

i totally agree, just 1 test is not enough to say that .net its faster than other tld. good test anyway