Rand's presentation on machine learning and deep learning in Google, Facebook, and beyond, and how engagement reputation will become key to every online marketing effort.
Fight Back Against Back: How Search Engines & Social Networks' AI Impacts Marketing
1. Rand Fishkin, Wizard of Moz | @randfish | rand@moz.com
Fight Back Against Back
Why the back button has become web marketing’s
greatest enemy (and how to defeat it)
8. In 2012, we only had to
worry about the path to
conversion and CRO on our
own sites.
9. We could let audiences
self-select out of these
phases after an initial visit
without fear of
repercussions
10. If putting a price here meant
80% of visitors left, no problem.
So long as the more qualified
ones (those we really wanted)
stayed, we were doing our jobs.
15. In 2012, Google Published a PaperAbout How
they Use ML to Predict Ad CTRs:
Via Google
16. 2012
“Our SmartASS system is a
machine learning system. It
learns whether our users
are interested in that ad,
and whether users are going
to click on them.”
17. By 2013, It Was
Something Google’s
Search Folks Talked
About Publicly
Via SELand
18. In October of 2015, they
finally revealed
RankBrain, anAI-
system input to the
search rankings
Via Bloomberg Business
19. As MLTakes Over More of Google’sAlgo, the
Underpinnings of the Rankings Change
Via Colossal
20. Google is PublicAbout How They Use MLin Image
Recognition & Classification
Potential ID Factors
(e.g. color, shapes,
gradients, perspective,
interlacing, alt tags,
surrounding text, etc)
Training Data
(i.e. human-labeled images)
Learning
Process
Best
Match
Algo
21. Google is PublicAbout How They Use MLin Image
Recognition & Classification
Via Jeff Dean’s Slides on Deep Learning; a Must Read for SEOs
22. Machine Learning in Search Could Work Like This:
Potential Ranking
Factors
(e.g. PageRank, TF*IDF,
Topic Modeling, QDF, Clicks,
Entity Association, etc.)
Training Data
(i.e. good & bad search
results)
Learning
Process
Best Fit
Algo
23. Training Data
(e.g. good search results)
This is a good SERP –
searchers rarely bounce, rarely
short-click, and rarely need to
enter other queries or go to
page 2.
24. Training Data
(e.g. bad search results!)
This is a bad SERP –
searchers bounce often,
click other results, rarely
long-click, and try other
queries. They’re definitely
not happy.
25. The Machines Learn to Emulate the Good Results & Try to Fix
orTweak the Bad Results
Potential Ranking
Factors
(e.g. PageRank, TF*IDF,
Topic Modeling, QDF, Clicks,
Entity Association, etc.)
Training Data
(i.e. good & bad search
results)
Learning
Process
Best Fit
Algo
26. Deep Learning is Even MoreAdvanced:
Dean says by using deep
learning, they don’t have to
tell the system what a cat is,
the machines learn,
unsupervised, for
themselves…
28. Googlers Don’t Feed in Ranking Factors… The Machines
Determine Those Themselves.
Potential Ranking
Factors
(e.g. PageRank, TF*IDF,
Topic Modeling, QDF, Clicks,
Entity Association, etc.)
Training Data
(i.e. good search results)
Learning
Process
Best Fit
Algo
29. Last October, Google Finally Went Public with
Their Use of ML-Based RankBrain
Via Bloomberg Business
38. Twitter’s new home
screen will work the
same way –
highlighting the “most
important” (aka “most
engaged-with”) tweets
from accounts you
follow
Via Mashable
54. Ask “WhatAreAll the Needs of These Searchers?” Then
ServeAs Many as Possible
You might be trying to sell desks, but
searchers are seeking answers to all of these
55. If the Competition Delivers Value to Searchers Who
Aren’t Buyers, But You Don’t…
58. Optimize the Title, Meta Description, & URL
a Little for Keywords, but a Lot for Clicks
If you rank #3, but have a higher-
than-average CTR for that
position, you might get moved up.
Via Philip Petrescu on Moz
59. Every Element Counts
Does the title match
what searchers want?
Does the URL seem
compelling?
Do searchers
recognize & want to
click your domain?
Is your result fresh?
Do searchers want a
newer result?
Does the description
create curiosity &
entice a click?
Do you get the
brand dropdown?
60. Given Google Often Tests New Results Briefly on Page One…
ItMayBeWorthRepeatedPublicationonaTopictoEarnthatHighCTR
Shoot! My post only made it to #15…
Perhaps I’ll try again in a few
months.
61. Driving Up CTR Through Branding Or Branded
Searches May GiveAn Extra Boost
62. #1 Ad Spender
#2 Ad Spender
#4 Ad Spender
#3 Ad Spender
#5 Ad Spender
63. With Google
Trends’ new, more
accurate, more
customizable
ranges, you can
actually watch the
effects of events
and ads on search
query volume
Fitbit was running ads on Sunday
NFL games that clearly show in
the search trends data.
65. Better Content > More Content
A lot of SEO used to be about establishing
authority through brute quantity, but Panda,
and now Rankbrain, are changing that.
66. Better Social Shares > More Social Shares
Via Rand’s Facebook Page
When I have a successful
post on Facebook, it boosts
Facebook’s likelihood to
show my posts in the
future…
67. Better Social Shares > More Social Shares
High engagement grows
my reach potential.
Low engagement shrinks
my reach potential.
70. Better Rankings >
More Rankings
A brand that consistently gets on page 1
but isn’t holding searchers’ interest or
develops a negative brand reputation in
SERPs may find those page 1 rankings
are hurting their ability to get #1
rankings!
72. Speed, speed, and more speed
Delivers an easy, enjoyable experience on every device
Compels visitors to engage, share, & return
Avoids features that dissuade or annoy visitors
Authoritative, comprehensive content that’s uniquely
valuable vs. what anyone else in your space provides
The Marketer’s User Experience Checklist
73. Uniquely Valuable Content
Via R2D3
Lots of articles try to explain machine learning, but
this one SHOWS how it works in a way anyone can
grasp.
81. Top-of-Funnel Content Can’t Be Used Solely to Filter
Out the Non-Customers
Trying to rank w/ content
that only serves one niche
of your search audience
may be a recipe for failure
82. FightingAgainst Back Means Serving a Broader
Audience
AngelList’s tool makes
salary comparison
easy, fast, and serves
a huge range of roles,
locations, and
markets
Via AngelList
83. Or, Getting More Precise with Your Search Query ->
Content Targeting
By targeting a less
competitive, lower volume
query, Compass can reach
the audience they’re
seeking
84. Either Way, Engagement Metrics on Content Must
Become KPIs
Improving Pages/Session and
lowering Bounce Rate should
probably play a “link-building-
like” role in your SEO arsenal
85. Our Content CTAs Deserve to Be Customized,
Tested, & Refined
(just like conversion-focused landing pages)
e.g. I bet I could make
a better CTA for the
comparison tool than
this (which looks far too
much like an ad IMO)
Via Talkpay (Comparably’s Blog)