Skip to main content

More than 39 million people across the U.S. own a voice-activated smart speaker.

NPR’s last study on smart speakers revealed that 16% of Americans own a voice-enabled device. Gartner predicts that 75% of all U.S. households will have smart speakers by 2020. As new devices, capabilities, and cheaper price points drive exponential growth, comScore predicts that 50% of all searches will be completed by voice by 2020.

Confronted with evolving search technologies and behaviors, brands and agencies need to adapt. To remain relevant and continue to reach consumers, optimizing websites for voice search queries will become absolutely crucial.

The shift to voice search creates new questions around search strategy: How do you achieve the coveted position of the voice search result? Should your strategy differ depending on device? Do the results from Google Assistant better match SERPs versus other smart speakers because Google is a major search engine?

Chloe Yeung, our SEO expert, performed some research to help get answers. This study looks into voice search results across three different voice agents and advises on how to best optimize content for voice search.

Voice Search

Important findings about voice search


  • In order to better match a user’s intent, voice search agents will append certain words to your search or change it to a related query.
  • Siri’s image search is powered by Bing.
  • Alexa favors “top rated” products and businesses over products and businesses with high organic rankings.
  • Google Assistant favors content coded with <ol> ordered lists over pages organized in header and <p> paragraph format regardless of organic rank and featured snippet status.
  • Google Assistant’s localized results most often match the local 3-pack result.
  • For news results, Google Assistant utilizes Accelerated Mobile Pages (AMPs). The “speakable” structured markup attribute can help you achieve a voice search result over a page that doesn’t use it.



Table of contents


How to optimize for voice search

Study methodology

Voice search usage: What we learned

Query testing

Voice search analytics

Final thoughts  



How to optimize for voice search


  • Keep your local SEO up to date with correct information. Select appropriate business categories.
  • Publishers who have new content published at least daily can build an Alexa Flash Briefing skill that users can add to their news readings.
  • Create AMPs for your content and utilize the “speakable” attribute in your structured markup to create “metadata” for voice. Remember that it will be read aloud, so it should be more conversational than website metadata.
  • Use descriptive alt tags and titles for your images. Optimize images for Google and Bing.
  • Monitor and respond to reviews about your product or business.
  • Remember that many voice search devices don’t have screens. Use descriptive language and complete sentences in your content.
  • Utilize ordered and unordered lists for how-tos and tutorials.



Study methodology


For this study, Hero Digital ran a survey on voice search usage, receiving over 100 respondents in the United States.

The survey consisted of 17 questions covering demographics and a variety of topics related to voice search. Of the respondents, 71% used voice search. Males and females were split pretty evenly, and respondents came from a range of ages, although 67% fell between 25-34.

In addition to the survey, our team tested about 50 queries spanning a range of topics and types of queries, utilizing both short and long tail language.

We tested these across Siri, Alexa, and Google Assistant. We completed the same searches on both Google and Bing, desktop and mobile, to compare to our voice search answers.  





Voice search usage: What we learned


Voice AI

Siri and Alexa were the most widely used devices, followed by Google Home. This is likely because our respondents were skewed toward early adoption, with 81% of respondents using voice search for more than 1 year and 41% using for more than 2 years. Google Home was released in December 2017, missing out on the early adopter market. Only 2% of our respondents used Cortana. Of respondents who use multiple devices, most either preferred Alexa or Google Assistant, stating that they were better than Siri at understanding queries and providing relevant answers.


Of respondents who use voice search, 93% use it at home, 11% at work, 53% in a car, and 22% outdoors or in a public place. Car and public place users are most likely utilizing phone and tablet devices Siri and Cortana to complete their searches.


Our respondents typically use voice search agents for basic tasks like checking the weather and news as well as playing music. 54% said they use it to find local businesses, 90% to learn information and 80% to complete an action such as booking reservations and setting reminders. Only 4% use their voice search agents to purchase items and 3% for home automation. As more skills and integrations are released, these queries and actions will become more advanced.

SERP landscape

According to Rank Ranger’s mobile SERP feature tracking tool as pulled on May 7, 2018, for mobile Google queries, Featured Snippets have been showing for 7.9% of queries, down 3.5% month over month. On the other hand, Direct Answers have been increasing their real estate, now showing for 13.5% of mobile queries, up 4% month over month.

Featured Snippets typically shows a paragraph or bulleted list that best matches your search, while Direct Answers provide a direct answer to your question. For instance, “how old is Betty White” provides a direct answer that says she’s 96 years old.

There’s no doubt that the shift to more direct and simplified answers is partially due to the growing popularity of voice devices. In our survey, 18% of respondents were more dissatisfied than satisfied with voice search. The biggest dissatisfaction responses cited involved issues around receiving insufficient answers and devices not understanding them well enough. Some people said they disliked how searches yielded a listing of results rather than answering the question outright. The shift towards more direct answers is a step in the right direction for more useful voice search results.  



Query testing


User behavior

To distinguish differences in language and wording in queries, we asked our survey respondents what they would type to search for a restaurant as well as what they would say to complete a voice search.

Unsurprisingly, people used complete sentences significantly more when searching via voice than when typing. There were more non-location modifiers such as “best” and “top rated” seen in voice searches (16% compared to 11% for typed searches), further showing that optimizing for long-tail keywords is important for voice search. Most interestingly, a handful of people utilized very conversational language with voice agents including queries like “I’m hungry” and “What should I eat?”

We tested some of the most common searches across devices:


Local search

Queries tested:

  • “Thai restaurants in Philadelphia”
  • “Thai restaurants near me”
  • “I’m hungry”
  • “What should I eat?”
“Thai restaurants in Philadelphia”

Siri: “One option I found is Baan Thai on N American St in Philadelphia, which averages 4 stars and is moderately priced.”

Alexa: “Here are a few top rated Thai restaurants in Philadelphia – JJ Thai Cuisine, House of Thai Cuisine, Cucina Zapata, Sawatdee”

Google Assistant: “I found a few Thai restaurants near Philadelphia. The first one is JJ Thai Cuisine at 2028 Chestnut Street in Philadelphia. The second one is Erawan Thai Cuisine at 123 S 23rd St in Philadelphia. The third one is Xiandu Thai Fusion at 1119 Walnut St in Philadelphia.”

Siri selects one listing rather than reading a few choices. Siri will ask if the selected location sounds good to the user and if the user says no, it will continue reading the next location in the list of 15 options. There didn’t seem to be any particular order matching SERPs or Yelp’s listings.

The query tested did not ask for top-rated restaurants, but Alexa listed all 4.5 star rated restaurants, causing these options to differ from SERP listings.

At the time of the first search, the Google local 3-pack showed Erawan and JJ Thai and both of these listings show in the top 10 organic listings. Xiandu was the fourth local listing, below Tamarind. So why did Google Assistant choose Xiandu above Tamarind? One potential reason is that Xiandu has structured markup while Tamarind does not. More likely though, is that this is just normal variation amongst the local 3-pack. When re-testing this query on mobile and desktop a few days later, Xiandu appeared as the third listing.


“Thai restaurants near me”

Siri provided 5 options arranged by proximity to our location while Alexa again provided “top rated” Thai restaurants 4 stars and above. Alexa’s results continue to favor local listings with higher ratings and more reviews. Google Assistant matched the local 3-pack results.

“I’m hungry”

The recognition of intent across all three voice search agents was pleasantly surprising. We tested this against typed searches on both Google and Bing.

Google attempted to recognize the intent of the search, showing a local 3-pack, but felt more confident in a YouTube video of a song entitled “I’m Hungry,” pushing the 3-pack below the fold.

Bing did not recognize the intent, just showing normal organic results for the query.

Siri responded with “I don’t want you feeling all peckish” and gave us 15 restaurant options of varying cuisine, all less than 1 mile from our location. Google Assistant matched the local 3-pack.

Alexa responded with a few options. The Alexa app showed that it searched for “food venue” based on the query. While this was the correct intent, the results included a grocery store and a Chevron gas station. These locations technically offer food products, but are not typical dining establishments.

This again reinforces the importance of choosing the most relevant categories for your local business.

“What should I eat?”

Google Assistant matched the local 3-pack for this query as well.

Siri gave us a joke answer, “Sunchokes and beans are great for you. Just not before getting on an elevator.”

Alexa presented us with options for recipes.

Users are able to place ads in the local 3-pack. It’s important to note that Google Assistant currently ignores paid local results and only presents the organic listings as options.



We received the best results with stock price searches because these provide Direct Answer results. All of the devices read the stock price and percentage change from previous close.



Because news is such a broad topic, this type of search offers more customizable results on both Alexa and Google Assistant via the apps.

Alexa always plays your custom “flash briefing.” It’s defaulted to play an NPR recording, but users are able to add and remove content that is most relevant to them.

For Google Assistant, unless you specify a topic, it will play the lineup of recordings from the news sources the user has set up in the app. 

Queries tested:

  • “Current news in America”
  • “What are the latest headlines?”
  • “News about Facebook”
“Current news in America”

Siri provided a news widget with 5 top stories for the headlines search, but had a difficult time understanding the “current news in America” search. It showed “America’s greatest hits: history” on iTunes, and then web results for what looked like a search for “America.”

“What are the latest headlines?”

For the latest headlines query, Google suggested ways to narrow future searches like “news about the Oscars” and then began playing the customized news feed. Google told us about Kanye West referring to Donald Trump as his brother in a tweet, and other details surrounding the incident. This was an interesting response, as it’s extremely specific, and is more pop culture-related than world news. Performing a typed Google search on both desktop and mobile for this query only showed organic listings with no story highlights or news carousels surrounding this story.

“News about Facebook”

When asked for “news about Facebook,” Alexa ignored the request for news surrounding a specific topic and instead read the flash briefing set up in the app.

Siri’s results did relate somewhat to Facebook, but were not completely relevant. For instance, it pulled in results that mentioned Facebook posts, but was not about the company itself.

Google search for "news about facebook" on desktop

Google Assistant read the story from The Verge, which shows up on both Google desktop and mobile carousels and is an AMP page.

It seems that Google Assistant favors the most recent articles, as The Verge’s post was only 2 hours old, compared to others in the carousels. However, we tested this query again less than an hour later and while there was a newer AMP post from Fast Company, Google still cited The Verge, followed by a blog called The Ringer, and another post from The Verge posted a day prior to our search.


Google search for "news about facebook" on mobile


Looking into each of the articles mentioned, all of them were AMP, but the difference is in the structured markup. Google announced “speakable” markup in July 2018. While it’s still in beta, there’s strong evidence that the “speakable” structured markup attribute is affecting which results get pulled into voice. Both Verge and Ringer posts utilized this attribute while Fast Company did not. In our testing, the content listing in the “value” fields were the pieces of content that Google Assistant read.

Key insight: Google Assistant will look to AMPs for news results. The “speakable” structured markup attribute can help you achieve a voice search result over a page that doesn’t use it.



Queries tested:

  • Cheap hotels in Philadelphia
  • Book me a room in Vegas
Cheap hotels in Philadelphia

Alexa ignored our request for “cheap” and again served us options that were “top rated” on Yelp.

Google Assistant’s results matched the local 3-pack.

Siri served us another listing of 15 options, seemingly organized from most expensive to least expensive.

Siri and Alexa voice searches compared to a typed Google search for book me a room in Vegas

Book me a room in Vegas

The voice agents were not as helpful with this requested action.

Google Assistant said “Sorry, I don’t know how to help with that” and Alexa did not respond at all.

Siri suggested a hotel called Santa Fe Station Hotel & Casino, about a half hour drive away from the strip.

None of the suggested hotels in the list were casino hotels on the strip that come to mind when you think about Las Vegas. Similarly, the first page of search results on Google showed discount booking websites including Expedia, Kayak, Travelocity and more.

On mobile, however, there is a People Also Search For carousel with a related search “best place to stay in vegas for the price” displaying bigger name locations like MGM Grand and Bellagio.



Queries tested:

  • How old is Betty White?
  • How much is Kim Kardashian worth?
  • Show me pictures of Ryan Gosling
  • When is Taylor Swift’s birthday?
  • How old is she? (specifically did not specify that “she” is referring to Taylor Swift to test successive questions)
How old is Betty White?

Again, this was a direct answer result so all voice agents provided the correct answer.

How much is Kim Kardashian worth?

All three voice search agents answered us directly with varying amounts. Google Assistant’s answer matched the featured snippet from the Google desktop and mobile searches. Alexa pulled an answer from Wikipedia. It’s unclear where Siri pulled its answer from, as it doesn’t state a source, but many articles from 2015-2017 reported this amount. However, Siri does include a link to the Wikipedia page in its knowledge graph.

Show me pictures of Ryan Gosling.

Siri showed us an image search widget powered by Bing. Apple switched Siri’s default search engine from Bing to Google in September 2017 for text-based searches, but has not yet switched images. 

Alexa and Google both required a connection to outside devices like Fire TV, Echo Show, or Chromecast, and could not show us pictures due to their lack of screen.

Multi-part query: When is Taylor Swift’s birthday? (after answer) How old is she?  

Surprisingly, Siri, Alexa, and Google Assistant all understood who we meant by “she” and answered with her age, something that Google’s typed search struggled with. Bing understood intent a bit better, showing a mix of results both related to and unrelated to Taylor.



Queries tested:

  • Best chocolate chip cookie recipe
  • How do I make a pizza?
  • Why does my cheesecake have cracks?
Google search for best chocolate chip cookie recipeBest chocolate chip cookie recipe

Below certain mobile results, there is an option to send the recipe to your Google Home. Google announced this functionality on their blog in May 2018. We clicked this button under a recipe and when we tested the query on our device, it read the recipe we sent to our Google Home. There are “supported cooking partners” that this button can appear for.

Alexa recommended “the top recipe” called outrageous chocolate chip cookies, which we did not find on the first page of results for Google or Bing. It does, however, have a higher Allrecipes rating than the recipe in the answer box. It’s unclear why Alexa recommended this specific recipe, but the Allrecipes website is thoroughly marked up using microdata.

Voice search optimization tips for recipes: It’s possible that Google will release an API in the future to allow you to collaborate as a supported cooking partner. Consider contacting Google to ask about becoming a partner.

Include as much information as you have available in your structured markup. Monitor and respond to reviews. If a user is having an issue, you could potentially provide a solution, leading to a happy customer and updated review. Side benefit: responding to reviews can increase paid search conversion rates.  


How do I make a pizza?

Google search for How do I make a pizza? recipeFor this query, we did not previously send a recipe to our Google Home. This search on Google produced an answer box, but Google Assistant’s response ( wasn’t showing on the first page of desktop or mobile.

When we searched for “how do I make a pizza? recipe” however, the result Google Assistant mentioned was ranked in the second position of organic results.

The King Arthur Flour recipe was still not the website in the answer box, but it did have strong structured markup, a higher rating, and a higher number of reviews. Most importantly, it utilizes <ol> ordered list markup for the recipe steps while the answer box website only uses headers and paragraph tags.

Key insight: In order to better match a user’s intent, voice search agents will append certain words to your search or alter it in a way that makes sense. This is shown through the pizza example above as well as the “I’m hungry” example. Include categorical keywords that your content falls under, like “recipe,” “tutorial,” “how-to,” etc. into your optimization plan.

Alexa again responded with “a top recipe” even though we did not request “best” or “top rated.” It was an Allrecipes recipe, not found on the first page of search results for Google or Bing but has over 2,000 reviews and a 4.7 rating.

Key insight: Alexa will give you listings with better ratings and more reviews over an answer box result, even if you don’t specify that you want the “best” product or location. As shown in the “Thai restaurants in Philly” and “cheap hotels in Philadelphia” queries, this is applicable to other types of queries including local listings.


Video and general information

Queries tested:

  • Videos on how to tie a windsor knot
  • How do you tie a windsor knot?
  • How do I change a tire?
  • Show me a funny cat video

Unsurprisingly, voice search agents without a screen struggled the most with requests for video, saying that they didn’t know how to help, or required connected devices.

Siri provided the most relevant answers, showing YouTube widgets with 9 video options.

How do you tie a windsor knot?

A desktop Google search presented an answer box result from (third organic listing) and a People Also Ask widget, followed by two YouTube organic listings and standard website organic listings filling up the rest of the first page.

Siri gave us the top 5 organic listings matching Google, but giving no special treatment to the answer box result, still showing it in the third position.

Alexa read a result from, which we didn’t find on the first three pages of Google or Bing. is a website that seems to be a question and answer aggregator, and this specific page referenced (the answer box result) as its source of information.

Google Assistant read an answer from, which was the fifth organic listing. Both the and sites were pretty well optimized, both showing diagrams, numbered listing of steps and videos on their pages, though neither of the sites use structured markup. owns the answer box result and a higher organic listing, and the site has an SSL certificate as well while does not. Google Assistant still favors over the seemingly superior, possibly because the wording of the steps is more clear and robust. Void any screen on this device, clarity is key.

Sample of wording:

To tie the Windsor Knot, select a necktie of your choice and stand in front of a mirror. Then simply follow the steps below:

1. Start with the wide end (“W”) of your necktie on the right, extending about 14 inches below the narrow end (“N”) on the left. Then cross the wide end over the narrow end.

2. Bring the wide end up through the loop between the collar and your tie.

3. Then bring the wide end back down.

4. Pull the wide end underneath the narrow end and to the right. The back side of the tie’s wide end should be visible.

1. Start with the wide end of the tie on the right and the small end on the left. The tip of the small end should rest slightly above your belly-button (this will vary depending on your height and the length & thickness of your tie). Only move the active (wide) end.

2. Wide end over the small end to the left.

3. Up into the neck loop from underneath.

4. Down to the left.

5. Around the back of the small end to the right.

How do I change a tire?

Siri and Alexa missed the mark with this query. Siri showed us a Wolfram Alpha widget that shows how many calories you burn while repairing an automobile. Alexa responded that it couldn’t help with that.

A Google desktop search shows an answer box result from (which also owns the first organic listing position), two YouTube videos, and standard website organic listings. Google Assistant gave an answer from, the fourth organic listing. On the front end, Bridgestone’s result is much more robust in terms of description than DMV’s. However, the difference for this example is in the code. Each step in Bridgestone’s process is marked up as an <h3> heading and supplemental paragraphs with <p> tags. On the other hand, DMV uses <ol> ordered list code. This structural format helps voice search agents delineate that this is a step by step process rather than paragraphs of text.  

Key insight: Google Assistant favors content coded with <ol> ordered lists, and will present pages organized this way over higher ranked and featured snippet pages organized in header and paragraph format.


Voice search analytics

As with any campaign, it’s best to track and measure the results of your work. However, the current setup of voice search analytics has flaws that limit the ability to accurately depict how much traffic is coming from voice.

All voice searches are currently being attributed as Direct traffic in Google Analytics instead of Organic. In addition, Google has been promising new voice search reporting features in Search Console, but we have yet to see them.

As a temporary workaround, utilize “queries” data from Google Search Console and paid search reporting. Because voice searches are typically long-tail and conversational, filter your data to only show queries that have more than 5 words or include words like “who, what, when, where, why, and how.” While not a true representation of voice search data, this method can provide you with estimated voice search traffic, clicks, and impressions until more accurate reporting features are released.  



Final thoughts

While these test searches were not run on a large scale and are not indicative of all voice search queries, here are the general takeaways we found from survey responses and individual device testing.

  • Generally, Siri will be more difficult to optimize for as there is no strong connection between its results and what appears in SERPs. For localized searches, Siri will give you listings that are closest to your specified location or your device. For informational queries that don’t produce a direct answer, Siri will give you top Google results excluding the answer box. Interestingly, Siri pulls web results from Google, but image results from Bing.
  • Alexa’s results revolve around the locations, pages, or products with the top rating. Reputation management is key for optimizing for Alexa’s voice search.
  • Google Assistant offers the most similar results to Google’s SERPs, especially with local searches. Structured markup and coded formatting are pertinent to winning the voice search result.

What’s next?

Adoption of voice search and voice assistant based transactions is on the rise, and as marketers think about ways to make sure their brands are “heard” in the consideration set, it’s important to go back to basics.

While the era of voice is ushered in, take a step back and think about how your users currently interact with your brand and how voice can reduce the friction for them. Take some time to do this – don’t just react. Have a brainstorming session and map out some experiences in the customer journey that could be enhanced by voice optimization.

For some, it will mean creating apps designed for Alexa. For others, it will mean making sure you go through the proper channels to make sure your location shows up for relevant queries.

The tactics might change slightly, but the strategy remains very similar. Make the experience relevant, frictionless, even enjoyable, and you can win in the voice space.



About Hero Digital

Hero Digital is a leading independent customer experience company born in California at the intersection of business, design, and technology. Our purpose is to bring moments of Truth & Beauty into people’s lives by creating meaningful customer experiences through consulting, design, engineering, digital marketing, and data science.

Hero Digital offers comprehensive SEO services including technical audits and implementation, content strategy, content creation and optimization, local SEO, international SEO, off-site asset optimization, and more. Learn more about Hero Digital’s marketing and SEO services or contact us to get started.



About the author

Chloe Yeung is Sr. Manager, SEO at Hero Digital. Chloe has led SEO engagements for clients like AmeriGas, American Standard, Sunrise Senior Living, and Universal Health Services. As a partner to these organizations, she helps develop and execute strategies to increase their organic presence.