Deliver relevant results from voice search without frustrating your shoppers

Jun 27th 2023 ai

Voice and natural language input has grown steadily over the past decade. You may recall getting your first smart speaker  and being enamored with the possibilities. Then voice usage on the phone became more popular with Google Assistant and Siri’s increasing usefulness. Now, with large language model powered chat interfaces like ChatGPT, people are once again energized by natural language input.

Voice is such a great user interface because it is so versatile. It works at home with smart speakers and on the go in the car or with a phone. People can use it with hands full of groceries or if they don’t know how to read. It can be used for a chat-style back and forth conversation or for simple requests of information.

However, voice can also be frustrating for reasons closely related to why it is versatile. People can be more conversational with it and they expect it to just work without hiccups.

That’s why it’s important to get voice right.

Voice Search

As we’ve written many times on this blog before, voice and search are a perfect match. People want information and they want it fast.

Which leads to the first frustration searchers have when it comes to voice search: your site or app not having it.

Many people don’t integrate voice input into their search because they think it will be difficult, or because they think they will need to have a smart assistant level of interaction. But that’s not the case. There’s no need to be able to have a conversation with your users. They aren’t expecting to say, “Hello e-commerce site, I would like…”

No, it is enough to have a voice input take what the user says and put it directly into the search box. This way, users can input via one mode (voice) and view the output via another (the screen). This is useful, because users may have a lot of different reasons for wanting to avoid typing with their hands. Maybe only one hand is free because the other is holding a toddler’s hand. Or perhaps one has trouble typing properly on a phone’s small keyboard. Indeed, when we’ve done user research, many people say that they just simply dislike typing on a phone.

Other research supports as well:

• Web browsing is still the main feature being used on mobile devices, but voice searching is in second place (Perficient) 
• Half of American drivers have used voice search while jetting around (Voicebot.ai)
• 71% of consumers who wear devices say they’ll be utilizing voice search in the future (99 Firms)

This last statistic is worth highlighting: phones aren’t the only devices that people are using to search. They’re also utilizing their earbuds or even watches. And, of course, their cars.

However, mobile is the primary interaction point that people have with voice, and you need to support voice search there.

Voice Support

Thankfully, over the past few years, supporting voice input has become easier. Some of that is thanks to the platform and browser providers, and some of that is thanks to companies like Algolia.

For example, Algolia has multiple libraries which help to integrate voice search. There are libraries for voice on Android, voice on iOS, and for voice on mobile web using a voice overlay. This last one is interesting because not all browsers support speech to text. Notably, Edge and Firefox do not support it. In this case, the overlay will simply not show the input.

If you do want to support voice input in all browsers, you’ll need to turn to a third-party provider, such as AssemblyAI or Google Cloud Platform.

Now that you support voice input, you’ll need to support the way that people search with voice.

How People Search Via Voice

Generally, the best way to support people searching with their voice is by delivering a good search for everyone. Get the basics down with fast, relevant results and you’ve done most of your job already.

When people are searching through voice, they might be more conversational and use longer queries overall. This is where query understanding and AI-driven intelligent search is useful. Query understanding, such as query categorization, can help narrow or boost results, which makes exact matching of query to record less important.

Meanwhile, AI-driven intelligent search is incredibly powerful for increasing the recall of results. As queries get longer, they bring more information about the searcher’s intent. At the same time, they make it more likely that relevant records won’t match, because they don’t match all of the keywords in the query. (You can set all of the keywords to be optional, but this will increase recall too much, and should only be used when necessary, such as when there are no results returned at all.)

To increase the likelihood that relevant results are in the final result set, you should use search that works on concepts as well as keywords. Algolia’s NeuralSearch does exactly this. It preserves keyword matching in order to be very precise, while also blending in vector-retrieved results to increase recall. This way, whether shoppers search for “drink warmer” or “keep tea hot at desk” they will still get that ceramic drink warmer they want to buy.

Voice search is long past the days of a novelty or a “we’ll do it next quarter” initiative. Users expect voice search, and you owe it to them to do it in a way that benefits their user experience.

Contact our team to learn how we can help you leverage the power, efficiency and accuracy of AI-assisted voice search.