The way we search is changing and voice assistants are shifting how consumers search. When users initially started to use the internet they would create queries as if they were speaking to a person.
However, users soon learnt to remove any prepositions and conjunctions to group short words together in the effort of working as Google required of reaching the core of the keywords. At present, it is no secret that Google prioritises the user experience first, so, therefore, wants users to feel more comfortable typing in a natural, conversational way.
This guide will explain what voice search is, how it works and how we as digital marketers can optimise effectively for voice search.
What is voice search?
In short, voice search is the particular type of speech recognition technology that lets a user perform a query via a voice assistant or a device. The assistant then provides an accurate answer to fulfil the user’s intent. These can be used on a range of assistants, including Alexa, Google Home, Cortana, Siri and many more.
How does voice engine search work?
Voice search works by linking speech recognition to complex natural language processing (NLP) systems. These systems have to accurately identify and understand what the individual is asking and interpret how to effectively respond. NLP is an activity that is carried out by artificial intelligence in order to fully understand the text.
In 2012, voice search began to start using Deep Neural Networks (DNNs) to attempt to improve the overall speech recognition process. The combination of Google’s powerful search algorithms with sophisticated NLP technology has undoubtedly increased the accuracy of results.
A graph displaying the rise of Google’s machine learning accuracy from 2013-2017
Over the past 5+ years, there have been attempts to improve the accuracy of speech recognition. To increase its accuracy, a model that uses temporal classification and sequence discriminative training techniques are used.
Temporal classification is a specific type of training neural network output that tackles sequence problems where the timings is a key variable. Sequence discriminative training, on the other hand, looks to better match and improve the performance of speech recognition by analysing sequence constraints from different language models. These models are bespoke extensions of Recurrent Neural Networks that make searches even faster and more accurate in noise polluted environments.
What is Voice Search SEO?
With users testing out voice search assistants to make lower-risk purchases, the world of search marketing could be altering. Voice search has been a trending topic throughout 2019. To optimise for voice search, traditional SEO practices should be undertaken but should also be modified or include layered content strategies to meet the demand of how many users are now searching via voice? I will cover how you can effectively create this strategy later on in the article.
Voice search SEO, therefore, changes part of the way in which we optimise content on-site. The keywords that we need to be pushing to rank are inevitably more longer-tail with the conversational aspect of the user’s voice search. This is likely to increase the chances for visibility in featured snippets and PAA boxes. This is where voice assistants are most likely to pull their search results from, therefore increasing the chances of ranking for a voice search query.
The BERT Model
During October 2018, Google’s AI language researchers introduced BERT (Bidirectional Encoder Representations from Transformers) helping any individual to train a state-of-the-art answering system. Google has recently announced (25.10.19) that the BERT neural network-based technique for NLP is now being implemented within its search algorithm. This is to better understand the context of words within organic search, helping to match user queries with the most relevant results.
How Does The BERT Search Algorithm Update Work?
The BERT modelling system works by processing words in relation to all the other words within a sentence, as opposed to evaluating each word on an individual basis. By holistically analysing words within a sentence, this can be particularly helpful to understand the users intent behind the query.
Prior to this update, Google would provide featured snippet results that would not always answer the user’s specific query, instead simply lift the most related content from an article. By Google introducing the BERT model into their algorithms, this allows users to perform more detailed search queries and answers are detailed that are most concise and valuable to the user. This is particularly apparent through voice and displays on visual assistants where the responded search result is more conversational like they’re speaking to a real person.
Outlined below is how the BERT model will be altering rich snippet opportunities within the SERPs.
Rank brain and AI
With BERT being coined as the largest update in search since the introduction of RankBrain 5 years ago, these advancements in AI and NLP will largely help our voice assistants to understand language more like a human. From us having naturally flowing conversations with our assistants, they will be able to learn from our experiences to properly formulate a sentence, cross-reference with our scheduled diaries and serve preferences that are tailored to the person.
Local search and hyperlocality
A further aspect of voice search that we have seen being impacted by the increased popularity of voice search is local search. Voice search is used to tackle typing challenges as there is a desire for users to search ‘on-the-go’ which means that the local element is crucial. Results are more localised than ever now and the map pack can completely change by a user’s device moving just 100m.
Analysis into the search phrase “near me” over the past 5 years displays a strong increase as users have begun to adapt to the rise of assistants. With users having control over their voice queries and how they search, they are more likely to make their queries more clear, specific and of high intent.
The commercial intent of users is incredible high via voice assistants, by searching for queries such as:
- “Where can I buy stamps near me”
- “Can I buy”
- Crystal clear intent “can I buy” or “to buy”, e.g. “where can I buy stamps near me”
How To Optimise for Voice Results?
There are many different avenues to test out when optimising for voice search. When planning long-term strategies and to gain a first-mover advantage within the market before your competitors, we have listed some tactics to ensure that your SEO strategy is suitable for voice search.
Focus on longer-tailed keywords
Long-tailed keywords are more focused queries that the users when they are further down the conversion funnel. These queries usually have a lower search volume, are ‘easier’ to rank for which result in higher conversions when users are near to the point of purchase, as opposed to seed keywords. Therefore, create content that clearly and comprehensively answers users queries, which is imperative within voice search queries.
Primary keywords vs long-tail keywords
When it comes to on-page SEO, keyword research is arguably one of the largest factors to understand what your users are looking for and provide new opportunities. Primary keywords are considered to be short phrases that are around 3 words or less that are the main product terms.
Examples can include “houseplant” or “sushi restaurant”. Long-tail, on the other hand, are words or conversational phrases are a lot more specific to the users intent. These are more likely to be used within the interest or action stage of the conversion funnel where the user knows exactly about the product/service of purchase.
Conversational content strategy
When creating content strategies for voice search optimisation, it’s crucial to map out longer-term queries throughout various stages of the customer journey. By creating natural and conversational content, this will help to aid users more efficiently when they search for queries via voice search.
Outlined are example queries that different aspects of the conversion funnel bring.
Awareness – “what is the best home assistant?”
Interest – “can you call people on a home assistant?”
Desire – “is Google Home better than an Amazon Echo?”
Action – “how much does a Google Home cost?”
Varied content strategies
In order to rank for different terms and rich snippet opportunities, it’s recommended to create varied and layered content strategies to further increase the organic visibility within the SERPs. This can include informational, navigational, commercial and transactional content. Examples of each content intent are outlined below:
- Informational (how-to guides, thorough guides that provide an overview of a topic)
- Navigational (users looking for a specific landing page, increased targeting to increase the UX)
- Commercial (where a user is looking for more information before they purchase, e.g. product descriptions, FAQ pages, delivery options, location-based, customer service information)
- Transactional (most likely purchasing a product or service, videos, comparisons or product stories).
Most of all, the ultimate goal by producing varied types of content is to add value to the specific type of user at their point in their customer journey.
The optimal way to optimise landing page content for particular search intent is through E-A-T (expertise, authority and trustworthy). This will help you to create trustworthy content that is valuable to your user and they are not likely to have any further questions after reading this informational guide.
To understand how you’re able to utilise E-A-T to optimise your content fully, my colleague Liv has produced an in-depth explanation that can be seen here – /blog/16097/2-quick-wins-to-boost-your-contents-e-a-t/.
As the majority of voice searches are derived from mobile devices and the mobile-first indexing has been fully rolled out, it’s important that you have a mobile-friendly version of your site. Therefore, the mobile version of your site will be referred to as your primary version within Google’s index that it provides to users. An unresponsive website can impact the overall user experience, as well as organic rankings within the SERPS.
Structured data optimised
Structured data, also known as Schema Mark-up is a configuration of HTML data that you can embed within your website’s code to help to become more discoverable by search engines. This can help to increase your visibility within relevant search results and promote the likeliness of gaining rich snippets, such as a People Also Ask result.
Consider how users are speaking to their voice assistants in more of a questions-like basis. With informed keyword research, this is a perfect opportunity to answer these queries within your detailed and authoritative content across the site. See how structured data has changed during 2019 within this structured data slideshow.
As voice search users are more likely to be using their smartphones to carry out these queries whilst being on-to-go, it’s crucial that the load time suffices to keep users from creating a further query and not interacting with your site. In fact, the loading speed of a voice search result is a huge 3.8x faster than your average website.
If you are not ranking well within the organic SERPs at current due to slow site speed issues, this would be considered one of the first points of call to increase your likeliness of ranking for voice.
Focus on local business strategies
As we’ve mentioned, voice searches are more likely to be used on the go than not. As an extension of this, it has been reported that in 2019, 58% of users have primarily used voice search to find local business information. In order to optimise your business for local organic search, you can carry out the following activities:
- Fully optimise local pages on your site with location-specific keywords
- Set-up and stay active on your Google My Business profiles
- Location, opening hours, high-resolution images and contact information
- Review and respond to local reviews to ensure you are providing your users with the best possible experience – also sending key trustworthiness signals towards Google
- Include local business Schema markup to help Google understand where you’re located
The application of voice search is happening now and impacting many websites with the introduction of more accurate algorithm updates. With this in mind, voice search is definitely not expected to be disappearing anytime soon.
It’s fascinating how users are now feeling more comfortable typing and speaking in a more natural and conversational way towards our search engines and getting an accurate response in return. Language understanding still remains to be an ongoing challenge, however, with the introduction of artificial intelligence algorithms (such as BERT) it will be interesting to see how they continue to learn from its users to deliver the most useful content, ultimately helping the user.
If there’s one takeaway from this voice search optimisation guide, it would be that voice assistant and smart display screens are bringing conversational AI and touch together in a knowledgeable way. Because of this, websites should be creating informational and concise answers to optimise for this shift in organic search.
If you’d like to find out more information regarding voice search SEO, contact our team of technical SEO specialists, today.