July 30, 2016 | Sarah Danks
Larry Kim’s MnSummit Presentation
I can’t believe it’s been a month(+) already since the MnSearch Summit!
Once again, the MnSearch board attracted all kinds of awesome talent for the 360+ attendees to this year’s Summit — including a guy we look up to in the pay-per-click world, Larry Kim.
While Larry didn’t talk about PPC this year, he did give us a lot of food for thought about RankBrain, Google’s machine-learning, artificial intelligence system used in its search results algorithm.
Sounds a bit “Terminator,” doesn’t it? Well, that’s probably the reason Larry decided on that very theme for his presentation!
Without further ado, here’s a recap of Larry’s session.
3 random facts about Larry:
- Founder of Wordstream, Inc., and MobileMonkey
- Studied electrical engineering for 5 years at Waterloo
- Built a suite of long-tail keyword research tools
Suffice it to say Larry’s a very smart dude. I can’t lie — if Larry says it, I have a tendency to believe it.
Here’s what he had to tell the MnSummit attendees about RankBrain.
RankBrain & Machine Learning-Enabled Algorithms
There are five SEO strategies you’ll need to survive SEO judgment day…
…but what’s “Judgment Day?” It’s when Google’s core search algorithms will all be based on machine learning. Algorithms for spam detection, image search video search, query interpretation, core ranking, etc. — all will be machine-learning enabled.
Millions of websites would be impacted.
What’s machine learning?
- Checks to see if search results met user expectations
- Makes adjustments to query and rankings based on outcomes
Sounds creepy. That’s way in the future, right?
Where does Google currently use machine learning?
- Google cloud vision API
- Google cloud speech API
- Google cloud translate API
- YouTube’s redesigned app uses machine learning to recommend better videos
- AdWords entirely based on machine learning for ad targeting
- Facebook News Feed & Ad Targeting
- Twitter ads also use machine learning
…machine learning systems are pretty much everywhere.
What about the core algorithm changes? Validation signals? WHO CARES — get in! Come with me if you want your website rankings to live.
So what IS RankBrain?
Via Google: “RankBrain used on long tail queries / query interpretation…that changes rank.”
But how does it really work?
A searcher enters a long-tail query; RankBrain makes a guess and re-writes the query into something it recognizes (query interpretation). Then it decides: Did the result satisfy the user?
No ➞ Next time I see a query like this, I’ll try some other page
Yes ➞ Next time I see queries like this, I’ll put this page at the top.
Searching for RankBrain
Is this possible evidence of RankBrain???
Paid Search vs Organic Search
Paid Search: Top spots (1-2): long tail and head terms have similar CTR
Organic: Top spots long tail and head terms have large differences in CTR
Lower Spots (3-7): in paid search, long tail and head terms have large differences in CTR; in organic, they’re similar CTR:
Totally begs the question: Which came first — the CTR or the ranking???
Larry’s RankBrain Theory:
High click-through rate, long-tail rankings get *additional* bonus ranking boost.
Takeaway: Machine learning (in general) makes stuff increasingly winner-take-all.
If that’s true, what should SEOs do in order to prepare for RankBrain and other machine-learning enabled algorithms?
They’re certainly not going to tell you everything…
…and new machine-learning algorithms are tougher than their predecessors.
Luckily, Larry’s got some “weapons” for you to get through this…
5 SEO Weapons to Survive SEO Judgment Day
Larry’s #1 SEO Weapon: Mobile Donkey/Unicorn Detector
First off, you need to know what your organic CTR is.
To find it: Download your Search Analytics query data in Google Search Console.
And then you need to know what a good click-through rate is for your website…
…since SERPs differ greatly by industry, so you need to use your own data to plot the average CTR for YOUR industry!
What’s with the donkeys and unicorns, anyway?
You need to learn the difference between donkeys and unicorns, that’s what.
Other than the obvious magical properties, unicorns perform up to 6x better than donkeys.
So. You know the difference. You can discount the donkeys.
But now — how to convert a unicorn?
Larry’s #2 SEO Weapon: Donkey to Unicorn Converter
Who knew you COULD convert a regular boring donkey into a unicorn? Larry says you can, but there are some rules.
Don’t get stuck in a rut of creating the same ol’ same ol’ types of title tags…
Because, boring. Also, how’re you going to stand out in a sea of sameness?
It’s like Dynamic Keyword Insertion (DKI) for SEO…
…but lots of people use DKI in PPC ads.
Why shouldn’t you use DKI in headlines?
Take a look:
So, keyword headlines using dynamic keyword insertion do okay, but not great. And since we’re striving for unicorns, you can’t settle for “not great.”
Ok. So how SHOULD you write your headlines, then?
Turn to Larry’s Unicorn Research Lab. Larry says you need to rely on people’s emotions.
Leverage Emotional Triggers
Step 1: Pick an emotional trigger
The top 9 emotions that make people click like crazy:
(Can’t lie — I had to fudge and do some research on those last 3 since Larry’s slide was warped and I couldn’t see them…)
Step 2: Write headline copy from the perspective of one of these personas:
1. Bearer of bad news
4. Feel-good friend
Step 3: Use this title template commonly employed by “viral” articles:
Format — list post, quiz, infographic
Content type — images, quotes, pictures, facts
Emotional hook — emotional word or superlative
Topic — love, cats, dogs, fitness, health, Donald Trump
Because the same emotions that make people want to share things also cause high click-through rate.
Since unicorn headlines are the top 10%, that makes the odds of finding one, 1 in 10.
So test 10 different headlines to find your unicorn.
But beware: changing punctuation/bolding/capitalization with the same wording does not a new headline make!
You need 10 *different* headlines!!!
But after writing your 10 headlines, how do you spot the unicorn?
Finding a Winning Headline
Use AdWords to test your headline variations. That’s right — you can create an entire ad group just to test these out and find your winner(s).
Be sure to bid to position (i.e., adjust bids for an average position of the top spot).
Check out the average cost per click (CPC) around the world:
As Larry says, it’s okay to target cheaper markets for your headline testing.
Use broad match keyword match type.
Summary: Larry’s Donkey to Unicorn Converter
- Identify donkeys (bottom 10%)
- Write 10 different — DIFFERENT! — headlines
- Test 10 ads on AdWords
- Replace old headline with test winner
BUT — a note of caution:
Don’t use spam sites like CrowdSearch.me: “Rank your websites using organic crowd search from thousands of real people!”
Click-through rate has become Google’s number 1 factor in ranking websites (they say).
Inside AdWords — “Booting the bots: New botnet protections across our ads systems”
SO…we’ve got RankBrain by the balls!
But wait — they have a back-up system…
…so we’ll need bigger guns.
Larry’s #3 SEO Weapon: Engagement Rate Unicorn/Donkey Detector
Also known as Bounce Rate vs Organic position.
According to Larry, it’s a pass-or-fail type of test between the two KPI — if your bounce rate is low enough, you pass. But, you fail if bounce rate gets too high.
Larry says we know Google measures dwell time, i.e., person performs a query search, gets SERP results, clicks on a result…
…and Google pays attention to how long that person takes to come BACK to the SERP (or not) and clicks on another result.
You used to be able to “hides sites to find more of what you want.” Google would show a “block all from this site” option that would occur right in the SERP. While they don’t do that anymore, it’s thought that they’re still monitoring the dwell time in order to influence rankings.
Time on Site vs Organic position — what’s the correlation?
If people spend a “good enough” amount of time on your site, you pass…
…if, on the other hand, visitors come to your site and Google deems they’ve not spent an adequate amount of time, you fail.
Key takeaway: “Task completion rates” used to validate CTR.
Engagement Rate Unicorn/Donkey Detector
What’s a good conversion rate?
Something to know about donkeys vs unicorns:
Unicorns convert 5x BETTER than donkeys. Not 5%, Five TIMES.
So if the average conversion rate is 2.35%, those’re your donkeys. A unicorn, on the other hand — remember they’re the top 10% — will convert at 11.45%.
How do you turn conversion rate donkeys into unicorns?
Well, higher CTRs = higher conversion rates.
The great A/B testing fairy tale:
We changed the…
- font type
- button color
…and we got a 5% increase in conversions!
Whooptee doo. Those’re the small fry — and we obviously want large fry.
Especially since the “typical” A/B test results in early increases in conversions, but the honeymoon wears off and these gains don’t continue.
Change your offer to dramatically increase conversion rates:
- Start your free trail
- PPC analysis in 60 seconds or less
RankBrain terminated — now what?
Time travel to PAST: Before Users Do Searches
Larry’s #4 SEO Weapon: Facebook Ads & Google Display Network
We really liked this little tidbit about using Facebook Ads and the Google Display Network (GDN) to test!
How advertising really works:
- Step 1: Promote inspirational/memorable content about your brand to your target market
- Step 2: People see the ad, but don’t necessarily take action right away (but become biased – i.e., they’ve seen your brand and equate it with what they were looking for)
- Step 3: Later, when the need arises, people either a) do a branded search for you or b) do an un-branded search but see your result in the SERP and — since biased — are prone to clicking on/buying from you
…Larry, are you saying that paid advertising impacts organic CTR and conversion rates???
It’s been shown that Facebook ads DO impact clicks and conversions from paid/organic search.
Warning: Facebook Ads employ machine learning technology, so you can’t promote junk!
How DO you target on Facebook?
- Interests — people who have expressed interest in/liked pages concerning x
- Demographics — life events (e.g., anniversary within 30 days = jewelry)
- Behavior — travelers (behavior on Facebook suggests they travel frequently)
Google Display Network
How do you target on the GDN?
Custom affinity audiences. Simply type in the type of audience you’d like to reach with keywords/URLs. Clone existing customer lists.
Time travel to FUTURE: Ensure subsequent searches go your way.
Larry’s #5 SEO Weapon: Remarketing
Larry’s ad targeting stack (powered by time travel):
Delete your “bad neighborhoods” …
What does it all mean?
Old SEO tactics clearly don’t work anymore — cue Rand’s Whiteboard Friday of all the no-longer-effective SEO techniques — and they might even make things worse!
Don’t just wait for RankBrain and machine-learning algorithms.
Start working now:
- Better CTR and conversion rates = more clicks and conversions
- Improved CTR & CVR = better rankings
- Better rankings = even more clicks and conversions
VS the “death spiral of RankBrain and machine-learning algos:
- Terrible CTR & engagement rates = less clicks, leads, sales, etc.
- Crappy user engagement = terrible rankings
- Terrible rankings = even fewer clicks, sales, etc.
So, be a unicorn in a sea of donkeys!
Want to check out Larry’s entire MnSummit presentation? Get his slides here.
Don’t forget — if you want to read about all of this year’s presenters, you can check out my recaps from the 2016 MnSearch event.