Backlinko has done an extensive analysis of “voice search ranking factors” and identified 11 variables tied to appearing in Google Home results. The company examined 10,000 results delivered over the smart speaker.

What Backlinko found was consistent with what many others have been saying but there were also a few surprises. For example, the study discounts the impact of Schema to some degree and page authority.

Here’s a partial, paraphrased list of the ranking factors:

• PageSpeed is a significant factor; voice search results typically come from faster-loading pages.
• Google relies heavily on very authoritative domains for results, but pages not as much.
• Content that ranks well on the desktop tends to rank in voice search. This might be a correlation rather than causal however.
• Shema may not be a factor: 36 percent of pages voice search results came from pages using Schema.
• Roughly 41 percent of voice search results came from Featured Snippets.
• Voice search results are generally 29 words; however Google sources voice results from long-form content.
• HTTPS is critical.

Google has made page speed an explicit mobile ranking factor. Backlinko found that the page-load time for a voice result was almost 2X faster than traditional webpages. Not a surprise. What may be a surprise are the findings around Schema.

The company found that Schema was used on slightly more than a third of pages delivered over Google Home, somewhat more than in general results. Accordingly it discounted Schema as a voice search ranking factor:

Although voice search result pages tend to use Schema slightly more often than your average web page, the difference is not significant. Also, 63.6% of voice search results don’t use Schema at all.  Therefore, it’s unlikely that Schema has a direct impact on voice search rankings.

Below are Backlinko’s findings around Schema distribution in voice search results.

There are plenty of reasons to use Schema generally so this finding shouldn’t be seen as an argument against it. And many will question the validity of this finding. It may also be that Schema pages don’t appear more because they aren’t more prevalent and there are other important variables.

Among them, links matter for voice results as well. Domain authority was high but page authority was relatively low by comparison:

We discovered that the average Domain Rating of a voice search result was 76.8 . . . we found that the link authority of voice search result pages were significantly lower. In fact, the mean Page Rating of a voice search result was only 21.1.

Backlinko speculated that the voice algorithm was relying upon domain authority (over page authority) because that provided a higher level of confidence in the accuracy of results.

Long form content was also correlated with voice results. “Google voice search results predominately come from pages with a high word count,” the study asserts. In addition, “FAQ pages tend to perform particularly well in voice search.” Keywords were somewhat less important: “only 1.71% of voice search results use the exact keyword in their title tag.”

The company advises, “[D]on’t worry about creating individual pages that are each optimized around individual keywords. Instead, write in-depth content that can answer several different voice search queries on a single page.”

Finally, content that ranks well on the desktop appears to also rank well in voice results. This is logical. Nearly 75 percent of voice results on Google Home “came from a page ranking in the top 3 for that keyword.”

SEOs should review the post and do their own evaluations of the findings and recommendations.

While it’s not clear whether smart speakers will siphon off some query volume or merely be additive to the overall pie, at least two studies have shown that owners of Alexa and Google Home devices are spending somewhat less time with their smartphones.

Regardless, virtual assistants are emerging as an important consumer discovery tool and marketers need to take these platforms seriously and adapt accordingly. In Google’s case, the Assistant (which powers Google Home) is now available on 400 million devices.

A strong understanding of the economics of any business unit is absolutely critical to any digital marketing campaign managed against non-brand key performance indicators (KPIs) in a search engine marketing (SEM) campaign.

One would think any reasonably large and successful business would have a good handle on their unit economics, and that this knowledge will be shared down the chain of command to the mid and lower levels of the marketing team.

But time and time again, I have found this critical foundation is missing, miscommunicated, insufficient or is so outdated as to make it worse than worthless.  “Worthless” in this case means that bad data does no actual harm. “Worse than worthless” means the utilization of the wrong KPI goals resulting in media waste and — more importantly — missed revenue and profit opportunity.

In other words, the company’s health is at actual risk because the marketing team and the business team aren’t on the same page.  In some cases, members of the marketing team may be working at cross-purposes, using incompatible KPIs and metrics.

Silos standing in the way of marketing AI?

I had the privilege of attending a lunch recently with members of the Direct Marketing Club of New York, the Interactive Advertising Bureau (IAB), and other marketers, including MediaMath CEO Joe Zawadzki.

Although we attended the lunch to discuss the power of artificial intelligence (AI) and machine learning to solve marketing problems, the consensus was many brands aren’t ready to empower decision-making with AI.

Their organizations were often so siloed that the inertia of departments and organization charts were a big factor slowing down the adoption of AI and machine learning in terms of optimizing marketing spend. A common theme across the table was the winners and disrupters in many industries are the ones without any legacy departmental structure. These included Tesla, the Dollar Shave Club, Casper & Purple, and even Amazon.

How does this all relate to pay-per-click (PPC) search engine marketing campaigns? Well, the influence of legacy structures affects your ability to grow yourself as a marketer and business person and directly influences your ability to communicate — with rational questions — up the chain of command (while following protocol), even if those communications might decrease your own departmental budget.

Rational questions for marketers to ask

Let’s use some common KPIs as examples of how things are often done now in SEM campaigns and how one might apply smarter business unit economics by asking some rational questions in the following scenarios:

PPC account (typically retail) managed by return on advertising spend (ROAS) with last click attribution

Rational questions to ask include:

• Do we have any data that would predict which customers order more frequently?
• Do we have any data on which variables in the PPC campaign tend to attract “new to file” customers vs. returning customers?
• Do some categories of products have a significantly higher margin where a different ROAS KPI should be applied (so that the same dollar of spending generates more profit)?
• Do certain products and their associated keywords result in higher lifetime customer value (LTV)? (The same question applies to geography, time of day, mobile vs desktop, etc.)
• Should we deploy simple tests to understand the impact and attribution of other forms of paid and earned media (display, social, video/audio, etc.)? Any paid media that can be geo-targeted lends itself to such a test of incrementality. These kinds of tests often work better than attribution models that lack data points and are particularly important for search because often another marketing touch-point stimulates search behavior by the consumer. Therefore one can look not only at sales data but also at changes on brand search volume.  For example, if you double display spends in Albany, Denver, and Orlando, and both brand searches and sales happen to rise in those cities, you’ve isolated your interaction effect.  Do the same with any paid media you want to test.

PPC account managed based on last click attribution around cost per action (CPA) or lead gen

Not all businesses make the sale online, so marketing teams running campaigns used to generate leads and are given a cost-per-lead or CPA target to hit. Rational questions to ask include:

• Are all leads we see equal? This issue goes beyond the “Glengarry Glen Ross” level of good leads and bad leads, encompassing both lead conversion rate to a sale and also the value of the sale (immediate), along with lifetime customer value.  Part of that LTV discussion could revolve around churn rates (think cell phone plans and subscriptions of all types, including product and services subscriptions).
• Rather than the average cost per lead or CPA, can we be more nuanced in our optimization?

Fixed budget PPC account

Rational questions to ask include:

• What’s the marginal contribution to the business originating from extra search, social or display media? Despite nearly all online media being auctioned off in near real-time, many larger organizations like to fix their spend by channel or category. This isn’t often the right strategy. To have a conversation about where the media dollars should be allocated, estimate the marginal cost of the next KPI unit in search, social and display.  Yes, you’ll need to address all the intricacies of the interaction effects between other media and search, but PPC search is very inelastic (small changes in bidding often don’t result in position and volume change, whereas in display, depending on whether it’s retargeted or otherwise targeted, the ability to “buy your KPI” may be easier.
• Should we just add budget or take from another channel when we find opportunities? If you could buy dollar bills for $.90 (including the effort to do the transaction), would you cap your budget?  Of course not.  Neither should any business. Sometimes, when the KPI includes a LTV factor, there may be cash flow constraints that also have a time-value of money (it may take you a year to get your one dollar), but otherwise the more the better for “ninety cent dollar bills!”

There’s probably another column that could be written on the nuances of applying unit economics to marketing, but this should give you a solid start toward a good SEM campaign.

Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.

Google has been rolling out many new search features over the past few months related to images, featured snippets, and the knowledge graph. Today the search giant released another feature called “multifaceted featured snippets.”

Multifaceted featured snippets will be surfaced for queries that are sufficiently broad enough to allow for more than one interpretation of what was submitted. In these instances, the SERP returned will include more than one featured snippet, with the original query rewritten as the questions the algorithm assumes the user may have intended, and the results displayed in the multifaceted snippet will reflect those new questions.

From the announcement:

There are several types of nuanced queries where showing more comprehensive results could be helpful. We’re starting first with “multi-intent” queries, which are queries that have several potential intentions or purposes associated. The query “tooth pain after a filling,” for example, could be interpreted as “why does my tooth still hurt after a filling?” or “how long should a tooth hurt after a filling?”

For example:

Google Multifaceted Featured Snippet

Multifaceted Featured Snippets vs. Multi-Perspective Answers

Back in December, Bing began rolling out AI-powered multi-perspective answers as part of its “Intelligent Search” set of new features, which includes Intelligent Answers, Intelligent Image Search and Conversational Search. Multi-perspective answers are just one of the “Intelligent Answers” features that has been live since the rollout. These results surface two (or more) authoritative sources on a topic, and will typically include differing perspectives/answers to the query.

Bing leverages its deep recurrent neural network models to determine similarity and sentiment among authoritative sources, and extracts the multiple viewpoints related to a topic — providing the most relevant set of multi-perspective answers (covered in more detail here).

Bing – Multi-perspective Intelligent Answers

Google’s multifaceted featured snippets may appear not too dissimilar from Bing’s multi-perspective answers, in that they also provide multiple rich results for a single query, but they are instead based on the presumed multiple intentions of a query (resulting in both multiple queries and results) vs. multiple viewpoints resulting from a single query. With these types of broad queries, many interpretations of what the user is actually asking can exist.

Multifaceted snippets aim to provide a more comprehensive and actionable set of results for these multi-intent query scenarios. They differ from multi-perspective intelligent answers in that they presume a different question might be being asked altogether, and surface responses for each of the queries the algorithm assumes the user may have actually intended, as the screenshot below demonstrates:

Multifaceted Featured Snippet

Google plans to expand multifaceted featured snippets throughout 2018 to include other nuanced query types — beyond those that could have multiple intentions — and lists guidance-seeking queries as one example.

From the post:

“For example, guidance-seeking queries like “is it worth fixing my foundation?” have several components that could be important, such as cost, duration, methods and financing. We’ll continue to experiment with multifaceted featured snippets over this year to expand coverage.”

With both Google and Bing having fully adopted deep learning methods and using artificial neural networks to drive search advancements, we can expect to see a steady stream of changes in search results enhancements and improved information discovery.

As always, Google encourages users to submit feedback on these new search features as you encounter them in the SERPs. Read Google’s full announcement here.