Like it or not, SEO has become a brave new two-algo­rithm world in which search mar­keters must opti­mize for both Google’s algo­rithm, as well as human input, which is some­thing the search engine is start­ing to care about more. But that’s not to say SEOs and dig­i­tal mar­keters should shift gears com­plete­ly and sim­ply make pages for peo­ple instead of search engines. Rather, they should opti­mize for both, says Moz Co-Founder Rand Fishkin. Here’s why.

As Google launched new algo­rithms to fight manip­u­la­tive links and con­tent – and used fear and uncer­tain­ty about penal­iza­tion to keep sites in line – over the last three years, its actions have erased a decade of old-school SEO prac­tices, Fishkin said.

But it isn’t nec­es­sar­i­ly bad news from a con­sumer per­spec­tive. Google has also fig­ured out searcher intent and start­ed exam­in­ing lan­guage instead of just words to pro­vide bet­ter results and they’ve also fig­ured out sce­nar­ios when con­sumers want recent results, such as in a search for “dig­i­tal mar­ket­ing con­fer­ences.”

Mean­while, Google’s search qual­i­ty team has also under­gone an evo­lu­tion, Fishkin said. That includes the incor­po­ra­tion of machine learn­ing to not only pre­dict ad click-through rates (CTRs), but also organ­ic results. As machine learn­ing takes over more of Google’s algo­rithm, the under­pin­nings of the rank­ings change, Fishkin said.

What’s more, with a machine learn­ing sys­tem in search – in which poten­tial rank­ing fac­tors and train­ing data, such as what con­sti­tutes good and bad search results, are used to cre­ate a learn­ing process and then the best-fit algo­rithm – it’s some­times hard to fig­ure out why some­thing ranks the way it does. That’s even more pro­nounced with deep learn­ing, which takes machine learn­ing a step fur­ther in that it’s essen­tial­ly an algo­rithm that builds its own algo­rithm.

What Does Deep Learning Mean For SEO?

For one thing, it means Google won’t know why some­thing ranks the way it does or whether a vari­able is in the algo­rithm. That’s because query suc­cess met­rics – such as long- to short-click ratio, user engage­ment across the domain and sharing/amplification rate ver­sus oth­er results – will be all that mat­ter to machines.

“We’ll be opti­miz­ing less for rank­ing inputs and opti­miz­ing more for searcher out­puts,” Fishkin said.

That means the near future is real­ly about opti­miz­ing for two algo­rithms, Fishkin said.

“The best SEOs have always opti­mized for where we’re going,” he said. “Today I think we know bet­ter than ever where we’re going.”

That means find­ing bal­ance between clas­sic rank­ing inputs like key­word tar­get­ing, qual­i­ty, and unique­ness and searcher out­puts like rel­a­tive CTR and short- ver­sus long-click.

So how should search mar­keters do that? Here’s Fishkin’s advice.

Tip 1: Optimize More For Clicks

Search mar­keters should opti­mize the title, meta descrip­tion, and URL a lit­tle for key­words and then a lot for clicks.

“If you rank num­ber three, but have a high­er than aver­age CTR for that posi­tion, you might get moved up,” Fishkin said.

Because Google often tests new results briefly on page one, Fishkin said it may also be worth repeat­ed pub­li­ca­tion on a top­ic to earn high CTR.

In addi­tion, Fishkin said dri­ving up CTR through brand­ed search­es may give an extra boost as the per­cent­age of peo­ple who do brand­ed search influ­ence how a result ranks for non-brand­ed search.

Also, in a cat­e­go­ry like car insur­ance, brand spend on TV influ­ences the search­es being per­formed by con­sumers, which sort of sec­on­dar­i­ly influ­ences their suc­cess met­rics, which is why Fishkin said he thinks Triva­go will start creep­ing up in trav­el search­es.

Tip 2: Compel Site Visitors To Stay Awhile

Fishkin said pogo-stick­ing (when users have to go to a bunch of dif­fer­ent sites to find what they’re look­ing for because sites rank high­ly, but don’t sat­is­fy user queries) and long clicks (when users per­form a search, click on a result, and remain on that site for a long peri­od of time) togeth­er may deter­mine where a brand ranks and for how long. So SEOs should seri­ous­ly con­sid­er con­tent that ful­fills both the searcher’s con­scious and uncon­scious needs, as well as ensure speed, in order to com­pel vis­i­tors to go deep­er into a site.

One exam­ple is the New York Times, which had an inter­ac­tive graph that asked users to draw their best guess about how income pre­dicts a child’s col­lege chances because read­ers would nat­u­ral­ly spend a long time on that page draw­ing said graph.

Tip 3: Be As Comprehensive As Possible

Google is look­ing for con­tent sig­nals that a page will ful­fill all of a searcher’s needs and machine learn­ing mod­els may note the pres­ence of cer­tain words, phras­es, and top­ics pre­dict more suc­cess­ful search­es.

In oth­er words, in a search for New York City, a page that men­tions each of the five bor­oughs may rank high­er than a page that does not.

Tools like Alche­myAPI and Mon­keyLearn can help here, Fishkin said.

Tip 4: Create Content That Inspires Loyalty

Pages that get lots of social activ­i­ty and engage­ment, but few links, seem to over­per­form even for high­ly com­pet­i­tive key­words like “pho­tos of dogs.”

Fishkin said we see this kind of behav­ior when a URL becomes hot in social, which could mean Google is using oth­er met­rics to get data that mim­ics social shares, such as click­stream, engage­ment, and brand­ed queries.

Per Fishkin, Google almost cer­tain­ly clas­si­fies search results pages dif­fer­ent­ly and opti­mizes to dif­fer­ent goals, such as for a med­ical query that might be best served by a page that does not have many social shares.

In addi­tion, Google prob­a­bly wants to see shares that result in loy­al­ty and return vis­its, Fishkin said. Know­ing what the audi­ence and their influ­encers share is essen­tial, as is know­ing what makes them return or pre­vents them from doing so.

“We don’t need bet­ter con­tent. We need 10X con­tent,” Fishkin said.

Tip 5: Solve The User’s Entire Task, Not Just A Query

Google wants searchers to accom­plish their tasks faster.

So if Google sees that many users who per­form sim­i­lar types of queries end up on the same site, it might use click­stream data to rank that site high­er, even if that site doesn’t have tra­di­tion­al rank­ing sig­nals.

A page that sim­ply answers the ini­tial query may not be enough because Google wants to send users to web­sites that resolve their mis­sion, Fishkin said.

What do you think about opti­miz­ing for both Google’s algo­rithm and human input?