It's hardly surprising that Google Home is an extension of Google's search ecosystem. Home is attempting to answer more and more questions, drawing those answers from search results. There's an increasingly clear connection between Featured Snippets in search and voice answers.
For example, let's say a hedgehog wanders into your house and you naturally find yourself wondering what you should feed it. You might search for "What do hedgehogs eat?" On desktop, you'd see a Featured Snippet like the following:
Given that you're trying to wrangle a strange hedgehog, searching on your desktop may not be practical, so you ask Google Home: "Ok, Google — What do hedgehogs eat?" and hear the following:
Google Home leads with the attribution to Ark Wildlife (since a voice answer has no direct link), and then repeats a short version of the desktop snippet. The connection between the two answers is, I hope, obvious.
Anecdotally, this is a pattern we see often on Google Home, but how consistent is it? How does Google handle Featured Snippets in other formats (including lists and tables)? Are some questions answered wildly differently by Google Home compared to desktop search?
Methodology (10K --> 1K)
To find out the answer to these questions, I needed to start with a fairly large set of searches that were likely to generate answers in the form of Featured Snippets. My colleague Russ Jones pulled a set of roughly 10,000 popular searches beginning with question words (Who, What, Where, Why, When, How) from a third-party "clickstream" source (actual web activity from a very large set of users).
I ran those searches on desktop (automagically, of course) and found that just over half (53%) had Featured Snippets. As we've seen in other data sets, Google is clearly getting serious about direct answers.
The overall set of popular questions was dominated by "What?" and "How?" phrases:
Given the prevalence of "How to?" questions, I've broken them out in this chart. The purple bars show how many of these searches generated Featured Snippets. "How to?" questions were very likely to display a Featured Snippet, with other types of questions displaying them less than half of the time.
Of the roughly 5,300 searches in the full data set that had Featured Snippets, those snippets broke down into four types, as follows:
Text snippets — paragraph-based answers like the one at the top of this post — accounted for roughly two-thirds of all of the Featured Snippets in our original data set. List snippets accounted for just under one-third — these are bullet lists, like this one for "How to draw a dinosaur?":
Step 1 – Draw a small oval. Step 5 – Dinosaur! It's as simple as that.
Table snippets made up less than 2% of the Featured Snippets in our starting data set. These snippets contain a small amount of tabular data, like this search for "What generation am I?":
If you throw your money recklessly at your avocado toast habit instead of buying a house, you're probably a millennial (sorry, content marketing joke).
Finally, video snippets are a special class of Featured Snippet with a large video thumbnail and direct link (dominated by YouTube). Here's one for "Who is the spiciest memelord?":
I'm honestly not sure what commentary I can add to that result. Since there's currently no way for a video to appear on Google Home, we excluded video snippets from the rest of the study.
Google has also been testing some hybrid Featured Snippets. In some cases, for example, they attempt to extract a specific answer from the text, such as this answer for "When was 1984 written?" (Hint: the answer is not 1984):
For the purposes of this study, we treated these hybrids as text snippets. Given the concise answer at the top, these hybrids are well-suited to voice results.
From the 5.3K questions with snippets, I selected 1,000, excluding video but purposely including a disproportionate number of list and table types (to better see if and how those translated into voice).
Why only 1,000? Because, unlike desktop searches, there's no easy way to do this. Over the course of a couple of days, I had to run all of these voice searches manually on Google Home. It's possible that I went temporarily insane. At one point, I saw a spider on my Google Home staring back at me. Fearing that I was hallucinating, I took a picture and posted it on Twitter:
I was assured that the spider was, in point of fact, not a figment of my imagination. I'm still not sure about the half-hour when the spider sang me selections from the Hamilton soundtrack.
From snippets to voice answers
So, how many of the 1,000 searches yielded voice answers? The short answer is: 71%. Diving deeper, it turns out that this percentage is strongly dependent on the type of snippet:
Text snippets in our 1K data set yielded voice answers 87% of the time. List snippets dropped to just under half, and table snippets only generated voice answers one-third of the time. This makes sense — long lists and most tables are simply harder to translate into voice.
In the case of tables, some of these results were from different sites or in a different format. In other words, the search generated a Featured Snippet and a voice answer, but the voice answer was of a different type (text, for example) and attributed to a different source. Only 20% of Featured Snippets in table format generated voice answers that came from the same source.
From a search marketing standpoint, text snippets are going to generate a voice answer almost 9 out of 10 times. Optimizing for text/paragraph snippets is a good starting point for ranking on voice search and should generally be a win-win across devices.
Special: Knowledge Graph
What about the Featured Snippets that didn't generate voice answers? It turns out there was quite a variety of exceptions in play. One exception was answers that came directly from the Knowledge Graph on Google Home, without any attribution. For example, the question "What is the nuclear option?" produces this Featured Snippet (for me, at least) on desktop:
On Google Home, though, I get an unattributed answer that seems to come from the Knowledge Graph:
It's unclear why Google has chosen one over the other for voice in this particular case. Across the 1,000 keyword set, there were about 30 keywords where something similar happened.
Special: Device help
Google Home seems to translate some searches as device-specific help. For example, "How to change your name?" returns desktop results about legally changing your name as an individual. On Google Home, I get the following:
Other searches from our list that triggered device help include:
- How to contact Google?
- How to send a fax online?
- What are you up to?
Special: Easter eggs
Google Home has some Easter eggs that seem unique to voice search. One of my personal favorites — the question "What is best in life?" — generates the following:
Here's a list of the other Easter eggs in our 1,000 phrase data set:
- How many letters are in the alphabet?
- What are your strengths?
- What came first, the chicken or the egg?
- What generation am I?
- What is the meaning of life?
- What would you do for a Klondike bar?
- Where do babies come from?
- Where in the world is Carmen Sandiego?
- Where is my iPhone?
- Where is Waldo?
- Who is your daddy?
Easter eggs are a bit less predictable than device help. Generally speaking, though, both are rare and shouldn't dissuade you from trying to rank for Featured Snippets and voice answers.
Special: General confusion
In a handful of cases, Google simply didn't understand the question or couldn't answer the exact question. For example, I could not get Google to understand the question "What does MAGA mean?" The answer I got back (maybe it's my Midwestern accent?) was:
On second thought, maybe that's not entirely inaccurate.
One interesting case is when Google decides to answer a slightly different question. On desktop, if you search for "How to become a vampire?", you might see the following Featured Snippet:
On Google Home, I'm asked to clarify my intent:
I suspect both of these cases will improve over time, as voice recognition continues to advance and Google becomes better at surfacing answers.
Special: Recipe results
Back in April, Google launched a new set of recipe functions across search and Google Home. Many "How to?" questions related to cooking now generate something like this (the question I asked was "How to bake chicken breast?"):
You can opt to find a recipe on Google search and send it to your Google Home, or Google can simply pick a recipe for you. Either way, it will guide you through step-by-step instructions.
Special: Health conditions
A half-dozen or so health questions, from general questions to diseases, generated results like the following. This one is for the question "Why do we sneeze?":
This has no clear connection to desktop search results, and I'm not clear if it's a signal for future, expanded functionality. It seems to be of limited use right now.
Special: WikiHow
A handful of "How to?" questions triggered an unusual response. For example, if I ask Google Home "How to write a press release?" I get back:
If I say "yes," I'm taken directly to a wikiHow assistant that uses a different voice. The wikiHow answers are much longer than text-based Featured Snippets.
How should we adapt?
Voice search and voice appliances (including Google Assistant and Google Home) are evolving quickly right now, and it's hard to know where any of this will be in the next couple of years. From a search marketing standpoint, I don't think it makes sense to drop everything to invest in voice, but I do think we've reached a point where some forward momentum is prudent.
First, I highly recommend simply being aware of how your industry and your major keywords/questions "appear" on Google Home (or Google Assistant on your mobile device). Look at the recipe situation above — for 99%+ of the people reading this article, that's a novelty. If you're in the recipe space, though, it's game-changing, and it's likely a sign of more to come.
Second, I feel strongly that Featured Snippets are a win-win right now. Almost 90% of the text-only Featured Snippets we tracked yielded a voice answer. These snippets are also prominent on desktop and mobile searches. Featured Snippets are a great starting point for understanding the voice ecosystem and establishing your foothold.
I HAVE BEEN WAITING FOR THIS BLOG POST SO LONG HOLY CRAP
YAY MY NAME IS IN THIS POST
At the risk of encouraging the already out-of-control menace that you are, that data set was pretty much perfect for what I was trying to do.
I wonder if the video results will end up being ones where Google Home sends the response to your TV... I'd test my theory on my Android tv, but I can't cuz Google Home isn't available in Canada yet *throws things at Google*
Yeah, given the announcement at I/O, I think that's going to be more and more likely. I expect Google to pursue inter-device communication heavily in the next couple of years.
Great blog Peter! I have noticed when i type in questions and the answers come up 9/10 that featured snippet is also number one on Google. I also noticed this on voice search. I am wondering what factors make you become the answer? Do you have to have certain meta keywords or descriptions? Does your title need to be the question you are trying to answer? Is the SEO basically the same as trying to rank businesses for there keywords?
Also Google have brought out that if you are in the map pack your rankings will decrease for those keywords you are in the map pack for. Do you have any idea whether or not they are going to start doing this to the people who are the answers?
Cory
The selection factors are already pretty complex (it's basically it's own mini-algorithm), but we know two things at this point:
(1) You have to rank on page 1. There are some small signs that Google might expand into page 2, but we're not sure on that. Within page 1, there's definitely a trend toward the upper positions. You're going to have the best chances if you're in the top 5.
(2) You need to show relevance to the question Google thinks is being asked (which, unlike this study, is sometimes implied, not explicit). We recommend a brief summary at the top of the page and then expanding the content. If your title/headers clearly target the question, and you've got a concise answer, you'll have a better chance. Of course, the trade-off is that you want a question/answer where the short version is going to draw interest to the longer answer. In some cases, people will just see the answer and be done, which isn't great for CTR.
The recommendations i've made to clients (for mainly text snippets) is ensure it is between 40-50 words, includes the query (which it should do naturally), synonyms of that query and most importantly we guided the client to create a text snippet that would mimic the response of a human to that question.
When we were next face-to-face with the client we actually got sales representatives in and asked them the question/query and jotted down their short answers. It seems to have worked for quite a few queries, we are still testing/implementing a few but the majority have actually stuck for a few weeks.
I highly recommend analysing what is holding the featured snippet now to see what they have, where it is positioned, the keywords highlighted by Google and as you say, the meta information, although from what i've seen the meta data has not made the impact we expected on appearance in the featured snippet.
Great information. One of the most interesting post here.
Next search: How to protect my home from spiders. I guess it may have some funny results.
Thanks for sharing, but you skipped the most important part. What ever happened to the spider?
The spider, after a brief game of hide-and-seek behind my monitor, was carefully removed and relocated to a nice plant next to the front porch.
I bet that was a Google crawler that came to life, came out of the Google Home. It was crawling the words right out of your mouth.
I am still wondering about the difference between the feature snippet to total question ratio between "how" questions and "how to" question in your "question phrases by question type" graph. How questions only have about a 50% corresponding feature snippet while how to questions have about a 90% corresponding feature snippet. It begs for a follow on article about "how do" "how is" "how come" and all the other variations of how questions and what is their population of feature snippets. Once the how family is explored, they obvious question is - why. Why does Google see fit to populate key snippets so vastly differently. Do they consider "how to" questions more important or is it simply easier for the algorithms to match snippets to how to questions? Inquiring minds want to know,
The other hows were all over the map. Quantity-oriented how questions, like "How many...", "How far...", etc. often had Knowledge Graph results or Featured Snippets. A lot of "How did..." questions were about how a celebrity died, and many of those have Knowledge Graph results.
I suspect (but my evidence is limited) that this is in part due to the nature of content on the web, and not just the Featured Snippet algorithm. There is a *lot* of content built around "How to..." questions and a lot of money spent to monetize those questions (home improvement, etc.). I have some reason to believe that Google displays Featured Snippets in part based on a quality threshold -- it's not just if they think you're asking a question, but there has to be a sufficiently relevant answer in the organic results. So, there's a relationship between the kind of questions people create content for and the kind Google answers.
Featured snippet are said to be the derivative of normal rich snippets, according to a recent study by Stone temple.
Ideally, whichever result is eligible to get featured snippet, should also be eligible for voice result. If not, all content marketers are in trouble.
How can someone optimize the content to make it eligible for featured snippet and voice search result both?
Unique and great, In search creating more devices and educating them about technology is required first before expanding search voice to the next level.
Uffff difficult to carry all this out, you are a phenomenon. I try to stay with some things that I liked and implemented them in my projects. Thanks Dr. Peter !!!
Insightful blog Pete and thanks for putting in the long hours putting your mental stability to the test!
I think it's been asked before, but do you ever see schema becoming the dominant source of data for search engines when providing voice answers (or text answers for that matter)?
For example with recipes, taking the response from structured data seems to me a much cleaner source than from the text on the page which often may be littered with the idiosyncrasies of the author. In a recipe for avocado on toast, reading a joke about mortgages for millennials is great for UX of someone reading the recipe on desktop, but not so great for the poor 20-something in their rented kitchen trying to avoid avocado hand.
It may also lead to easier interaction.
"OK Google, go back a step" - easy to imagine when working through a difficult recipe.
Or even Google Home prompting you:
"Are you ready for the next step?"
Apologies if this has been discussed in detail elsewhere, I did search.
Me and my team are really on the ball when it comes to featured snippets from text search in our industry, but we're still not there with schema on our content and just want to get others' thoughts on how schema may impact featured snippets in the future.
My mental stability was never in any real danger, given how little of it I had going into this post.
This may be an unpopular opinion, but no, I don't see schema being the dominant source of data. The simple problem is that Google needs access to as much of the available data as possible, and only a relatively small % of people will have the know-how to implement any given form of structured data. I think this is what killed authorship mark-up, to some degree.
That said, I do suspect that structured data will be very important in some niches and for certain Google partners. Look at the new fact check, for example -- that's entirely implemented with structured data (but only for approved sites). Recipes and other niche data likewise are much more likely to use structured data.
I don't think we're going to see structured data heavily in play for general answers, though.
Nice research, thanks for going all googley for us. The variance in results between desktop and Home is familiar - local vs map... maybe they'll converge once these features comes out of beta in a couple of years ;)
I also have the same question as Cory, I'm very much exited to get into the answer box for my blog posts, but failing to figure out how exactly Google rank a blog/article to be eligible to appear on answer box. There is a clear indication on Google support that we can't manually add any snippet or something on our blog post to be eligible for showing up on answer box, then how can we do it! any Guesses??
Replied to Cory... we don't know all the details, but we do know a couple of things that have worked.
Great blog post!! It's a really good information provide here..now days lots of people going to Voice base search important and most;y find in mobile.. These information through get ranking and increase the sales
I'm planning to buy Google Home as well. The voice search is really great. Especially, how it asks you a question about your intent, that's cool.
Nice work! - thanks for the data share
These are incredible tips. Thanks for sharing!
The great Dr. Peter !!
I love reading your post, I always learn many things that I like to carry out.
incredible post, very useful and packed full of information base don hard results.
Fact is: Google search is going to be huge so utilisting information as in this post is absolutely key.
Love the chart on Question Phrases By Question Type! I think this is going to be more and more and integral part of SEO and featured snippets will get more complex as Google and other search engines use data from voice searches. Great insight and examples Dr. Meyers!
Very Informative article, I also use Google voice search most of the cases & it gives exact search results for my query.
Awesome guide. I really enjoyed this one.
This may be an unpopular opinion, but no, I don't see schema being the dominant source of data. The simple problem is that Google needs access to as much of the available data as possible, and only a relatively small % of people will have the know-how to implement any given form of structured data. I think this is what killed authorship mark-up, to some degree.
That said, I do suspect that structured data will be very important in some niches and for certain Google partners. Look at the new fact check, for example -- that's entirely implemented with structured data (but only for approved sites). Recipes and other niche data likewise are much more likely to use structured data.
Thanks :-)