Rand looks at Moz's 2015 analysis of ranking factors in Google's search engine and compares opinion data, correlation numbers, and experiments to give a picture of how modern SEO fits together.
1. Rand Fishkin, Wizard of Moz | @randfish | rand@moz.com
Search Ranking Factors 2015
What data, opinions, and testing have revealed
about how Google’s rankings operate.
3. A look at Google’s algorithm
in 2015 according to
150 professional SEOs
4. We usedto show graphics
like this to illustratethe
relativeimportanceof
differentareasof
optimizationto Google’s
algorithm.
2013
5. But a pie chart suggests that
you can only get so much
value from any given set of
features.
In reality, factors like higher link
authority on your domain have as
almost unlimited ability to positively
influence rankings
6. Thus, we’vegot a new wayto illustrate
how rankingfactors fit together:
14. 1) Professional SEOs feel that, on average, the algo
is flattening, and the days of a single factor having an
overwhelming impact are fading.
Takeaways:
15. 2)After years of dominating the algo, links, while still
powerful, don’t feel like an overwhelming ranking
force to SEOs.
Takeaways:
16. 3) Engagement data is on the rise. If growth rate
continues, by our next survey, it may be in the top
two features.
Takeaways:
18. Becausecorrelationtells us
somethingelse of great value:
Correlation DOESN’T tell us why one
page ranks higher than another.
It DOES tell us what features higher-
ranking pages tend to have over their
lower ranking peers.
19. Do correlationcoefficients in the 0.1 – 0.4 range
(typical for single factors in searchengine studies)
mean anything?
Debunk statements
about what’s NOT
causal in rankings
3 UsefulApplications:
Show relative
potential
influence
ID factors for
more testing /
investigation
22. Coefficients canalso be used to show relativecorrelation:
The best SEOs use multiple
repetitions of keywords in their titles. I
guarantee it works better than some
fancy LDA model.
23. On average, content that better fits an LDA
topic model dramatically outperforms KW
repetition in the title
Coefficients canalso be used to show relativecorrelation:
24. Correlationnumberscan leadus to interestingtheoriesthat we
can then validatethrough other means:
Could it be that partial match anchor text now has equal or
greater ranking influence than exact match?
25. Correlationnumberscan leadus to interestingtheoriesthat we
can then validatethrough other means:
Let’s go run some experiments to see
if this is true y’all!
32. Traffic & Engagement
For the first time, we measured usage data. While traffic looks strongly
correlated, engagement metrics have weaker numbers.
Trafficandengagement
metricsvia
33. Keyword Use & On-Page Optimization
As we get more sophisticated in our text-modeling abilities, we’re seeing higher
correlations (though still low relative to links & social shares)
34. For the first time, we also broke correlationsdownby
categoryof keywords/SERPs
35. Health websites that
link out more tend to
rank higher.
Dining sites see almost no
correlation between linking
out & ranking.
36. It tended be more
present in higher
ranking sites for these
verticals
Anchor text had a
smaller relationship w/
high rankings in these
verticals
37. Those meager restaurant
websites? Looks like Google
doesn’t mind much.
Buzzfeed & Upworthy are
always showing how
lengthier articles perform
better for them.
38. Twitter & Facebook have
very similar relative
correlations, which fits w/
Google’s statements that
they don’t directly use
either.
In some verticals, social
sharing is much less
connected to ranking
positions than others
39. 1) Correlations with links have remained relatively
similar, suggesting that perhaps links haven’t faded in
influence as much as some in our industry have
suggested.
Takeaways:
40. 2) We need more sophisticated on-page analysis
tools. With the right algorithms/ software, we may
find real opportunities to improve rankings through
content.
Takeaways:
41. 3) Correlation is even more useful (and interesting)
on subsets of SERPs than on an entire corpus. In
the future, calculating correlations for the SERPs
you/your company care about may become
standard.
Takeaways:
42. 3 Examples of What Correlation &
Experimentation Can Do:
#1: Help us
validate what
Google says
#2: Verify
theories about
what’s in
Google’s algo
#3: Lead us to
better tactical
approaches
57. We knowthat linkscan still
overwhelm otherranking
signals.
Via RishiLakhanionRefugeeks
Pointing a few anchor-text
links at this blocked-by-robots
page on Matt’s blog made it
rank (even in 2015).
58. We knowaboutloadsof linkelementsthat influencerank-boosting
ability:
1) Anchor Text
2) PageRank
3) Relevance
4) Domain
Authority
5) Location on
the Page
6) Internal vs.
External
7) Quality of Other
Links on Page/Site
8) Editorial Weight
9) Engagement w/ Linking
& Linked Pages
10) Follow vs. Nofollow
59. 11) Source Depth
12) Text vs. Img
13) Link Age
14) Topical
Authority of
Source
16) Spam Signals
17) Speed/Acceleration
of New Link Sources
18) Author Authority
19) 1st Link to Target on
Page vs Duplicate Links
10) Prior Links to Target
from Source Domain15) Javascript vs. HTML
We knowaboutloadsof linkelementsthat influencerank-boosting
ability:
60. This stuff mattered a lot whenwe did manuallink
buildingto move rankings
But today,many of us justlet
content buildlinksfor us, right?
61. Moz & Buzzfeed joined forces for a
report looking at 1 million pieces of
content.
DataviaBuzzsumo&Moz’sJointStudy
63. This is a powerlaw
distribution– the top
contentgetsthe
overwhelmingmajority
of linksand shares.
64. The reality of social amplification and earning
links is…
0.028? That’s too close to 0 to infer any
consistent, direct influence.
65. For the most heavily shared content, there’s a little bit more of
a correlation, but it’s small enough that relying on social
shares to earn your links is probably folly.
We tried segmenting the samples:
66. This data showswhy I can’tendorseeitherof these common
maxims in SEO/contentmarketing:
Create good, unique content and Google will
figure out the rest.
The best way to earn links is to create great
content.
67. In the past,I presenteda conceptthat,basedon this data,
now appearsto be fundamentallyflawed:
69. Publish
Amplify
Grow network Rank for slightly
more competitive
terms & phrases
Get links Grow authority
Earn search
traffic
This doesn’t just happen. Link
building – outreach, embeds,
nudges, etc – are still essential.
70. 1) Social shares by themselves almost never
lead directly to the quantities of links necessary to
rank well.
Takeaways:
71. 2) Content that performs extraordinarily well on
social networks and ranks well in search engines
may not be benefitting solely from links.
Takeaways: