Jump to content
Sign in to follow this  
mki

Understanding Topics In Google

Recommended Posts

Posted here, because Yukon pointed this out.

Many people are still stuck on trying to rank keywords in Google global searches.

This often results in people targetting odd keywords or longtail keywords and this can work, but it's more difficult than you think because Google actually works differently than most people think it does.

Most people think Google returns search results based on the keyword and it doesn't.

Google actually interprets the topic that would be most appropriate for the search query and you will actually find similar search results on LSI keywords.

So for each topic, there will be a cluster of different keywords the content will rank on.

If the keyword is an LSI keyword, you can figure out the parent topic by looking at bolded words in the search result. If the bolded words are the keyword you are targetting, it's not an LSI keyword. But if they're not, then that's the actual topic you want to write about to target that keyword.

There's one big thing to know here though. Google will still evaluate click data per search result, not per topic. So the results on the LSI keyword will usually be similar pages, but the order might be completely different. This is because the click data (and some other factors) are not identical for each SERP.

Example: The keyword "beard designs" is a relatively low competition keyword with about 5k global searches monthly. But, when you look at the bolded words in that search result, you notice that the topic is actually "beard styles." If you Google "beard styles" you get the exact same search results in a similar order. That keyword gets approximately 324k global searches monthly and is very similar in difficulty.

So let's pretend you had a brand that sold beard styling products and you wanted to target the keyword "beard designs" for traffic to your ecommerce store. This used to work (I believe it was the hummingbird update that changed this,) it used to be that if you targetted "beard designs", that would be much easier than "beard styles," because the pages that are targetting "beard styles" do not have the word "design" on the page at all.

That hasn't worked for a long time but people are still making that mistake. If you target the keyword "beard designs", you are actually more likely not-to-rank-period if you do that since Google might think that it's a different topic entirely. The correct plan of action here would be to realize that if you target the keyword "beard designs" you are really targetting the keyword "beard styles" and it makes much more sense to target the correct keyword.

Proof: Google "beard designs" quoted and compare that to "beard styles" unquoted. It's completely different results.

Special thanks to Yukon, Google really does tell you the LSI keywords in most cases.

My theory on how Google figures out the topics is by analyzing the words on the pages that are in the index and then it compares that to click data that it already has. (I always assume it's not Aliens, AI, or that Google spent 1,000,000 hours building some insanely huge LSI keyword database which contains every keyword variation in every language. I have a degree in computer programming so I always think "What's an efficient way to do this without writing more code than I need to?" I'm pretty sure the programmers at Google think the same way. I'm sure it's more about a novel and efficient algorithm, rather than analyzing neural meshes (AI), which would probably require a supercomputer for every single search query...)

So, if you correctly targetting the keyword "beard designs", it would be a page about either, a designer named Beard, or creating designs based upon beards. If you do that, the content on the page will be extremely different than the words on the pages that are targetting "beard styles." So Google knows it's not the same topic and will likely not rank it, unless the trend it analyzes in click data changes over time.

Note: By relatively easy, I'm talking about "adults that do SEO." I'm not suggesting that it's easy for black hat spammers or something... Figure 1 week to produce a skyscraper and you need to earn something like 30-40 quality links from outreach marketing. So figure 4% effectiveness per email address and 4 follow-ups per contact, so that's something like 1,750 prospects to grind out 875 contacts, which will take 2-3 weeks to send the initial contact emails and the follows up will be over an additional 4 weeks. Unless you do crank or something, then maybe you could smash the emails out in a single 100hour session. Disclosure: I have no experience doing SEO in the men's grooming niche. It might be way easier, or it might be way harder.

Cost there would be something like time + image/media licenses (no idea) + $100 for Ahrefs + $50 for Voila Norbert+ $50 for hunter.io and I would just use a VPS to send the emails since the volume isn't high enough to worry about spam scores.

Return would be something like (figure 30k UV/monthly) $300 in Adsense revenue monthyly (not worth it) or if 1% opt-in to your ecommerce mailing list for a coupon, and those leads had an average earning of $5 per lead over time, that would be something like $1,500 monthly (worth it in my opinion) and the SEO campaign (if successful) would return $36k over 2 years, after it ranks. I'm not factoring in social shares here.

If you want to know how much crank that would take, I'll have to ask SEO Cat, since I personally don't take drugs.

fatcatonline-shutterstock.jpg

Share this post


Link to post
Share on other sites

SEMrush

True, Google will show you the LSI keywords, well, a small glimpse in the SERP description.

I searched for the keyword "buy car insurance", cleaned up the duplicates for almost 300 search listings and below is the condensed list.

Nothing amazing but it does return a few decent LSI keywords. The word "get" is interesting in this case, looks like it's the "buy" LSI keyword.

 

  • auto
  • auto insurance
  • bought
  • buy
  • buy auto insurance
  • buy car
  • buy car insurance
  • buy insurance
  • buying
  • buying auto
  • buying auto insurance
  • buying car insurance
  • car
  • car buying
  • car insurance
  • car insurance buying
  • cars
  • get
  • get auto insurance
  • get car insurance
  • get insurance
  • insurance
  • insurance vehicle
  • purchase a
  • purchase a car
  • vehicle
  • vehicle insurance

 

 

 

Share this post


Link to post
Share on other sites

Yeah and that's the correct way to find LSI keywords.

If you go to LSIgraph, the keywords it spits outs are indeed LSI keywords, but they're not necessarily the real LSI keywords that Google uses.

One of the keywords it gives you is "local independent insurance agents" and the SERP is completely different results compared to "car insurance".

The root keyword is "insurance agents", not "car insurance".

It appears that thegeneral has picked up on that and is targetting both terms correctly.

Share this post


Link to post
Share on other sites

Check this keyword out.

 

  • best nylon strap watches

 

Google favors some watch styles, models and brands over others as LSI keywords even though those devalued keywords back on the source webpage clearly include nylon watch straps.

Example, the "Times: Weekender" watch is bold while the "Invicta" brand and watch model aren't bold.

...and some of those bold keywords in the SERP descriptions (keyword list below) are anchor text for followed Amazon affiliate links.

 

 

3LNGwPl.png

 

best
best nylon
best nylon strap watch
best nylon strap watches
best straps
best watch
best watch strap
best watch straps
best watches
chronograph
eco-drive
emporio armani
good
nylon
nylon strap
nylon strap watch
nylon strap watches
nylon straps
nylon watch
nylon watch strap
nylon watch straps
nylon watches
nylon webbing
seiko 5
seiko men's snk809

strap
strapcode
straps
straps nylon strap
tag heuer
timex
timex men's t2n647 weekender
timex unisex t2n651 “weekender
tissot

top
watch
watch strap
watch straps
watch straps watches
watches
webbing
weekender

 

 

 

 

 

Share this post


Link to post
Share on other sites

Frequency in the top 10, (does the word exist on the page.)

 (tools used: sublime text and excel)

best    10
best nylon    2
best nylon strap watch    0
best nylon strap watches    0
best straps    0
best watch    3
best watch strap    2
best watch straps    2
best watches    1
chronograph    8
eco-drive    2
emporio armani    2
good    7
nylon    10
nylon strap    7
nylon strap watch    4
nylon strap watches    1
nylon straps    3
nylon watch    4
nylon watch strap    1
nylon watch straps    1
nylon watches    1
nylon webbing    0
seiko 5    1
seiko men's snk809    1
strap    10
strapcode    1
straps    8
straps nylon strap    0
tag heuer    1
timex    7
timex men's t2n647 weekender    0
timex unisex t2n651 “weekender    1
tissot    3
top    6
watch    10
watch strap    6
watch straps    5
watch straps watches    0
watches    10
webbing    0
weekender    5

data.xlsx

Note: the original keyword you posted only has 60 global search volume. Considering the obvious commercial intent, it's interesting, but I wouldn't personally target it directly (granted 1-3 good links to that specific page and you can get on page 1 of Google. If you were already in the watch business, okay sure. It's at a competition level where you can get page 1 rankings in about 30 days, which is obviously nice for a keyword with good commercial intent. Similar results are far more competitive. From a pure SEO perspective, the keyword "maratac nato" would be much better. The competition level is "do good onpage seo and get 1 good link" level for page 1. The volume is better and the CI is still there. The number 1 result only has one good link and two pretty bad links. So 3 good links should get you to number 1 assuming the onpage SEO is good.)

Share this post


Link to post
Share on other sites
Guest
On 4/20/2018 at 8:00 PM, mki said:

My theory on how Google figures out the topics is by analyzing the words on the pages that are in the index and then it compares that to click data that it already has. (I always assume it's not Aliens, AI, or that Google spent 1,000,000 hours building some insanely huge LSI keyword database which contains every keyword variation in every language. I have a degree in computer programming so I always think "What's an efficient way to do this without writing more code than I need to?" I'm pretty sure the programmers at Google think the same way. I'm sure it's more about a novel and efficient algorithm, rather than analyzing neural meshes (AI), which would probably require a supercomputer for every single search query...)

Back in 2003, Google bought Applied Semantics to form the core of their shift to LSI/A - they do use AI, along with a Vector Space Model, NLP, Term Document Matrices and even Stemming to come up with the results (not click data).

The reason I say "Not Click Data" is because it's like "Keyword Stuffing" in the old days - focusing on trends rather than actual context. LSI is all about context and how it relates to clearly define what a given web page is all about. 

Share this post


Link to post
Share on other sites
10 hours ago, BIG Mike said:

Back in 2003, Google Applied Semantics to form the core of their shift to LSI/A - they do use AI, along with a Vector Space Model, NLP, Term Document Matrices and even Stemming to come up with the results (not click data).

The reason I say "Not Click Data" is because it's like "Keyword Stuffing" in the old days - focusing on trends rather than actual context. LSI is all about context and how it relates to clearly define what a given web page is all about. 

 

VSM, sure.

NLP, maybe.

TDM, definitely, there's actually evidence of that.

Stemming, maybe.

The only machine learning I've personally observed is the click data / PLA algorithms, which I'm pretty sure it's called logistic regression. Which I personally don't consider it to be "AI." It honestly may not even be that complicated.

Remember: the algorithm can't be that complicated. Google crawls 160+ trillion pages, serves different results in every geographical region, and processes 3.5+ billion searches a day. Every search is data, every click is data, and every page probably has 1000+ individual pieces of data, if not more. Running a regression algorithm across that data set would be pretty intense... Not to mention everything else Google does. Also, it can't be that smart since people have cheated Google plenty of times. The first example that comes to mind is when Brain Dean ranked on "how to get high."

Share this post


Link to post
Share on other sites

It's not "smart".  That's a word that is really badly used when talking about AI.   Language ability can be built using a hierarchy of linking and blocking pathways.  Gates was looking for how to make an ap or program or whatever you want to call it that would actually be able to use appropriate language - instead of, for example, just hooking a noun to a verb so you would have sentences like "the chair ran".   The worst problem of building a co-occurrence recognition system is in the negation systems as far as I can see it.  I can't "do" the linkage systems, but I understand what they need to be able to do to get correct language in text.  It doesn't mean the computer "understands" anything but link commands. 

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.