Posted by purna_v
I was conned into my love of cooking by my husband.
Never having set foot in the kitchen until the grand old age of 22, my husband (then boyfriend) — a former chef — said he’d teach me some simple recipes. I somewhat enjoyed the process but very much enjoyed the lavish praise he’d bestow upon me when eating whatever I whipped up.
Highly encouraged that I seemingly had an innate culinary genius, I looked to grow my repertoire of recipes. As a novice, I found recipe books inspiring but confusing. For example, a recipe that called for cooked chicken made me wonder how on Earth I was meant to cook the chicken to get cooked chicken.
Luckily, I discovered the life-changing power of fully illustrated, step-by-step recipes.
Empowered by the clear direction they provided, I conquered cuisine after cuisine and have since turned into a confident cook. It took me only a few months to realize all that praise was simply a ruse to have me do most of the cooking. But by then I was hooked.
When it comes to voice search, I’ve talked and written a lot about the subject over the past year. Each time, the question I get asked is “What’s the best way to start?”
Today I’ll share with you an easy-to-follow, step-by-step guide to empower you to create your own voice search test. It’s sure to become one of your favorite recipes in coming months as conversational interfaces continue their rapid adoption rate.
Testing voice search? But it’s not monetized.
That’s correct. It’s not monetized as of yet. However, the usage rates have been growing exponentially. Already search engines are reporting that:
- One out of ten searches are voice (per Baidu)
- Twenty percent of all mobile Android searches are voice (Google)
- Usage spans all age ranges, as we discovered at Cortana (which is owned by Microsoft, my employer):
With Cortana being integrated into Windows 10, what we’re seeing is that age range demographics are now comparable to what eMarketer is reporting for overall smartphone usage. What this means: Using digital assistants is becoming more and more common. It’s no longer an edge case.
More importantly, voice searches done on the search engines can often have PPC ads in the resultant SERPs — as you’ll see in my examples below.
Why a PPC test?
It’s easier to get started by testing voice search via PPC since you can get more detailed reporting across multiple levels.
I would recommend taking a teeny-tiny budget — even $50 is often good enough — and putting it toward a voice search test. (Don’t fret, SEOs, I do have some tips in here for you as well.)
Before we start…
Here’s a quick reminder of how voice searches differ from text searches:
- Voice has longer queries
- Natural language means more question phrases
- Natural language reveals intent clearly
- Voice search has high local value
- And greatly impacts third-party listings
You can read about it in more detail in my previous Moz article on the subject.
Let’s get cooking!
Step 1: See what, if any, voice activity exists for you currently
Goal: Find out what voice-related activity exists by identifying Assumed Voice Queries.
Estimated time needed: 30 min
Tools needed: Search Query Reports (SQRs) and Excel
A good place to start is by identifying how your audience is currently using voice to interact with you. In order to do this, we’ll need to look for what we can term “assumed voice queries.”
Sidebar: What are Assumed Voice Queries?
Since the search engines do not currently provide separate detailed reporting on voice queries, we can instead use the core characteristics of these queries to identify them. The subtle difference between keyboard search and voice search is “whom” people think they are interacting with.
In the case of keyboard search, the search box clearly ties to a machine. Searchers input logical keywords they think will give them the best search results. They generally leave out filler words, such as “the,” “of,” “a,” and “and.” They also tend not to use question words; for example, “bicycle store,” rather than “what is a bicycle store?”
But when a searcher uses voice search, he is not using a keyboard. It’s more like he’s talking to an actual human. You wouldn’t say to a person “bicycle store.” You might say: “Hey Cortana, what is the best place to buy a bicycle near me?”
The key difference between text and voice search is that voice queries will be full thoughts, structured the way people speak, i.e. long-tailed queries in natural language. Voice searches tend to be approximately 4.2 words or longer on average, according to research from both Google and Microsoft Cortana.
Thus, assumed voice queries would be keywords that fit in with these types of queries: longer and looking like natural language.
Caveat: This isn’t going to be 100% accurate, of course, but it’s a good place to start for now.
Even just eight months ago, things were fairly black and white. Some clients would have assumed voice queries while others didn’t. Lately, however, I’m seeing that most clients I look at have some element of assumed voice queries, indicative of how the market is growing.
Okay, back to step 1
a.) Start by downloading your search term report from within your Bing Ads or Google AdWords account. This is also commonly referred to as the search query report. You want to run this for at least the past 30 or 60 days (depending on volume). If you don’t have a PPC account, you can pull your search term report from Google Search Console or Bing Webmaster Tools.
b.) Open it up in Excel, so we can get sorting.
c.) Sort the columns to just the essentials. I usually keep only the search term, as well as the impression columns. For larger accounts, you may prefer to leave on the campaign and ad group name columns as well.
d.) Sort by query length to isolate the search queries that are 5+ keywords in length — I’m going with 5 here simply to increase the odds that these would be assumed voice queries. A simple Excel formula — taught to me by my colleague John Gagnon—- can help count the number of words:
Replace A1 with the actual cell number of your search term, and then drag that formula down the sheet. Here it becomes C2 instead of A1:
e.) Calculate and sort, first by query length and then by impressions to find the assumed voice search queries with the most impressions. The result? You’ll get your final list — success!
Step 2: Extrapolate, theme, sort
Goal: Find additional keywords that could be missing and organize the list based on intent.
Estimated time needed: 45 min
Tools needed: Keyword tools of choice and Excel
Now that you can see the assumed voice queries, you’ll have handy insights into your customer’s motivation. You know what your audience is searching for, and also important, what they are not searching for.
Next, we need to build upon this list of keywords to find high-value potential queries we should add to our list. There are several helpful tools for this, such as Keyword Explorer and Answer the Public.
a.) Go to the keyword research tool of your choice. In this example, I’ve used SEMRush. Notice how they provide data on organic and paid search for our subject area of “buy (a) bicycle.”
b.) Next, let’s see what exists in question form. For any given subject area, the customer could have myriad questions along the spectrum of motivation. This output comes from a query on Answer the Public for “buy a bicycle,” showing the what, when, where, why, and how questions that actually express motivational intent:
c.) These questions can now be sorted by degree of intent.
- Is the searcher asking a fact-based question, looking for background information?
- Are they farther along the process, looking at varieties of the product?
- Are they approaching making a purchase decision, doing comparison shopping?
- Are they ready to buy?
Knowing the stage of the process the customer is in can help tailor relevant suggestions, since we can identify core themes and sort by intent. My brilliant colleague Julie Dilleman likes to prepare a chart such as this one, to more effectively visualize the groupings:
d.) Use a research tool such as Bing Ads Intelligence or your demographic reports in Google Analytics to answer core questions related to these keywords, such as:
- What’s the searcher age and gender breakdown for these queries?
- Which device is dominating?
- Which locations are most popular?
These insights are eminently actionable in terms of bid modifications, as well as in guiding us to create effective ad copy.
Step 3: Start optimizing campaigns
Goal: Review competitive landscape and plan campaign optimizations.
Estimated time needed: 75 min
Tools needed: PPC account, NAP listings, Schema markup
To get the lay of the land, we need to look at what shows up for these searches in the voice platforms with visual interfaces — i.e., the Search Engine Results Pages (SERPs) and Digital Personal Assistants — to see what type of results show up. Does the search provide map listings and reviews? Where are they pulling the data from? Are ads showing?
a.) Run searches across multiple platforms. In my example, I am using Siri, Google app and Cortana on my desktop.
Near me-type searches:
These all had map listings in common — Apple maps, Google maps, and Bing maps, respectively.
Siri got it wrong and led me to a store, while both Google and Bing Ads provided me with SERPs to answer my question.
Quick answer-type queries:
While Siri pulled up multiple results from a Bing search, both Google and Cortana found what they considered to be the most helpful answer and read them aloud to me while also providing the option for looking at additional results.
b.) Optimize your NAPs. Make sure you have listings that have an accurate name, address, phone number, and open hours on the top business listings such as Apple Maps, Google My Business, and Bing Places for Business.
c.) Ensure you have proper Schema markup on your site. The more information you can provide to the search engines, the more effectively they can rank and display your site. Be sure to add in:
- Contact info
d.) Optimize your PPC campaigns.
- Choose a small handful of voice search queries from your list across different intents.
- Add to new ad groups under existing campaigns. This helps you to take advantage of historical quality score benefits.
- Adjust bid modifiers based on your research on age, gender, and device.
- Adjust bids based on intent. For example, the following keywords demonstrate completely different levels of purchase intent:
- Do I need a hybrid or mountain bike? – More research-based.
- Who invented the bicycle? – Zero purchase intent. Add this as a negative keyword.
- When does bike store XYZ open today? – High likelihood to purchase. Bid up.
Step 4: Be the best answer
Goal: Serve the right message at the right time in the right place.
Estimated time needed: 60 min
Tools needed: Creativity and Excel
Make sure you have the relevant ad for the query. Relevance is critical — the results must be useful or they won’t be used.
Do you have the right extensions to tailor toward the motivational intent noted above and the consumer’s ultimate goal? Make it easy for customers to get what they want without confusion.
Voice searches cover a variety of different intents, so it’s important to ensure the ad in your test will align well with the intent of the query. Let’s consider this example:
If the search query is “what’s the best waterproof digital camera under $500?” then your ad should only talk about digital cameras that are waterproof and around the $500 range. Doing this helps make it more seamless for the customer since the selections steps along the way are much reduced.
A few additional tips and ideas:
a.) Voice searches seem to frequently trigger product listing ads (PLAs) from the search engines, which makes sense since the images make them easier to sort through:
If you can but haven’t already done so, look at setting up Shopping Campaigns within your PPC accounts, even just for your top-selling products.
b.) For results when the SERPs come up, be sure to use ad extensions to provide additional information to your target audience. Consider location, contact, conversion, and app information that is relevant. They make it easy for customers to find the info they need.
c.) Check citations and reviews to ensure you’re showing up at your best. If reviews are unfavorable, consider implementing reputation management efforts.
d.) Work to earn more featured snippets, since the search engines often will read them out as the top answer. Dr. Pete has some excellent tips in this Moz article.
e.) Your helpful content will come to excellent use with voice search — share it as a PPC ad for the higher-funnel assumed voice queries to help your test.
f.) Video has been getting much attention — and rightly so! Given the increased engagement it can provide, as well as its ability to stand out in the SERPs, consider offering video content (as extensions or regular content) for relevant assumed voice queries.
Step 5: Analyze. Rinse. Repeat.
Goal: Review performance and determine next steps.
Estimated time needed: 60 min
Tools needed: Analytics and Excel
Here’s where the power of PPC can shine. We can review reporting across multiple dimensions to gauge how the test is performing.
Quick note: It may take several weeks to gather enough data to run meaningful reports. Remember that voice search volume is small, though significant.
a.) First, determine the right KPIs. For example,
- Lower-funnel content will, of course, have the most conversion-specific goals that we’re used to.
- Research-type queries will need to be measured by micro-conversions and different KPIs such as form fills, video views, and leads generated.
b.) Pull the right reports. Helpful reports include:
- The keyword performance report will show you the impressions, clicks, CTR, quality score, conversions, and much more about each individual keyword within your campaigns. Use the keyword report to find out which keywords are triggering your ads, generating clicks, and leading to conversions. You can also identify keywords that are not performing well to determine whether you want to delete them.
- Ad performance reports show you the impressions, clicks, spend, and conversions for each ad. Use this report to help you determine which ads are leading to the most clicks and conversions, and which are not performing. Remember, having underperforming ads in your campaigns can pull down the quality of your campaign.
- Filter by device and by demographics. This combination telling us what devices are dominating and who is converting can help us to adjust bids and create more effective ad copy.
- Create a campaign report looking at your PLA performance. Do tweaks or major overhauls to close gaps versus your expectations.
c.) Determine where you can personalize further. AgilOne research indicates that “more than 70% of consumers expect a personalized experience with the brands they interact with.”
Carefully pairing the the most ad messaging with each assumed voice query is incredibly important here.
Step 1. See what, if any, voice activity exists for you currently.
Step 2. Extrapolate. Theme. Sort.
Step 3. Start optimizing campaigns.
Step 4: Be the best answer.
Step 5. Analyze. Rinse. Repeat.
Pretty do-able, right?
It’s relatively simple and definitely affordable. Spend four or five hours completing your own voice search test. It can open up worlds of opportunity for your business. It’s best to start testing now while there’s no fire under us and we can test things out in a low-risk environment — an ideal way to get a leg-up over the competition. Bon appétit!
Have you tried some other tests to address voice search queries? Please do share in the comments below.