by Juan Robin II
This post was originally published on the STAT blog.
We all know that featured snippets provide easy-to-read, authoritative answers and that digital assistants love to say them out loud when asked questions.
This means that featured snippets have an impact on voice search — bad snippets, or no snippets at all, and digital assistants struggle. By that logic: Create a lot of awesome snippets and win the voice search race. Right?
Right, but there’s actually a far more interesting angle to examine — one that will help you nab more snippets and optimize for voice search at the same time. In order to explore this, we need to make like Doctor Who and go back in time.
From typing to talking
Back when dinosaurs roamed the earth and queries were typed into search engines via keyboards, people adapted to search engines by adjusting how they performed queries. We pulled out unnecessary words and phrases, like “the,” “of,” and, well, “and,” which created truncated requests — robotic-sounding searches for a robotic search engine.
Of course, as search engines have evolved, so too has their ability to understand natural language patterns and the intent behind queries. Google’s 2013 Hummingbird update helped pave the way for such evolution. This algorithm rejigging allowed Google’s search engine to better understand the whole of a query, moving it away from keyword matching to conversation having.
This is good news if you’re a human person: We have a harder time changing the way we speak than the way we write. It’s even greater news for digital assistants, because voice search only works if search engines can interpret human speech and engage in chitchat.
Digital assistants and machine learning
By looking at how digital assistants do their voice search thing (what we say versus what they search), we can see just how far machine learning has come with natural language processing and how far it still has to go (robots, they’re just like us!). We can also get a sense of the kinds of queries we need to be tracking if voice search is on the SEO agenda.
For example, when we asked our Google Assistant, “What are the best headphones for $100,” it queried [best headphones for $100]. We followed that by asking, “What about wireless,” and it searched [best wireless headphones for $100]. And then we remembered that we’re in Canada, so we followed that with, “I meant $100 Canadian,” and it performed a search for [best wireless headphones for $100 Canadian].
We can learn two things from this successful tête-à-tête: Not only does our Google Assistant manage to construct mostly full-sentence queries out of our mostly full-sentence asks, but it’s able to accurately link together topical queries. Despite us dropping our subject altogether by the end, Google Assistant still knows what we’re talking about.
Of course, we’re not above pointing out the fumbles. In the string of: “How to bake a Bundt cake,” “What kind of pan does it take,” and then “How much do those cost,” the actual query Google Assistant searched for the last question was [how much does bundt cake cost].
Just after we finished praising our Assistant for being able to maintain the same subject all the way through our inquiry, we needed it to be able to switch tracks. And it couldn’t. It associated the “those” with our initial Bundt cake subject instead of the most recent noun mentioned (Bundt cake pans).
In another important line of questioning about Bundt cake-baking, “How long will it take” produced the query [how long does it take to take a Bundt cake], while “How long does that take” produced [how long does a Bundt cake take to bake].
They’re the same ask, but our Google Assistant had a harder time parsing which definition of “take” our first sentence was using, spitting out a rather awkward query. Unless we really did want to know how long it’s going to take us to run off with someone’s freshly baked Bundt cake? (Don’t judge us.)
Since Google is likely paying out the wazoo to up the machine learning ante, we expect there to be less awkward failures over time. Which is a good thing, because when we asked about Bundt cake ingredients (“Does it take butter”) we found ourselves looking at a SERP for [how do I bake a butter].
Not that that doesn’t sound delicious.
Snippets are appearing for different kinds of queries
So, what are we to make of all of this? That we’re essentially in the midst of a natural language renaissance. And that voice search is helping spearhead the charge.
As for what this means for snippets specifically? They’re going to have to show up for human speak-type queries. And wouldn’t you know it, Google is already moving forward with this strategy, and not simply creating more snippets for the same types of queries. We’ve even got proof.
Over the last two years, we’ve seen an increase in the number of words in a query that surfaces a featured snippet. Long-tail queries may be a nuisance and a half, but snippet-having queries are getting longer by the minute.
When we bucket and weight the terms found in those long-tail queries by TF-IDF, we get further proof of voice search’s sway over snippets. The term “how” appears more than any other word and is followed closely by “does,” “to,” “much,” “what,” and “is” — all words that typically compose full sentences and are easier to remove from our typed searches than our spoken ones.
This means that if we want to snag more snippets and help searchers using digital assistants, we need to build out long-tail, natural-sounding keyword lists to track and optimize for.
Format your snippet content to match
When it’s finally time to optimize, one of the best ways to get your content into the ears of a searcher is through the right snippet formatting, which is a lesson we can learn from Google.
Taking our TF-IDF-weighted terms, we found that the words “best” and “how to” brought in the most list snippets of the bunch. We certainly don’t have to think too hard about why Google decided they benefit from list formatting — it provides a quick comparative snapshot or a handy step-by-step.
From this, we may be inclined to format all of our “best” and “how to” keyword content into lists. But, as you can see in the chart above, paragraphs and tables are still appearing here, and we could be leaving snippets on the table by ignoring them. If we have time, we’ll dig into which keywords those formats are a better fit for and why.
You could be the Wonder Woman of meta descriptions, but if you aren’t optimizing for the right kind of snippets, then your content’s going to have a harder time getting heard. Building out a voice search-friendly keyword list to track is the first step to lassoing those snippets.
Want to learn how you can do that in STAT? Say hello and request a tailored demo.
Need more snippets in your life? We dug into Google’s double-snippet SERPs for you — double the snippets, double the fun.