Profile picture of Coren Feldman

8 min read

How AI can impact the user experience

Artificial intelligence collects and analyzes data in most interfaces. Here’s how we can make it more transparent and user-centered.

An illustration of a few web icons surrounding a phone that reads “You may also like”

Illustrations by {name}

Stay informed on all things design.

Thanks for submitting!

Shaping Design is created on Editor X, the advanced web design platform for professionals. Create your next project on Editor X. 

The AI in our imaginations and the AI in use in the real world are two very different things. While science fiction depicts artificial intelligence as a borderline sentient being, today’s AI is far clunkier and requires a lot more human intervention.


AI can be a useful tool to create great user experiences, but there are a lot of hurdles to overcome when adding artificial intelligence to a product. Making sure you’re taking the right precautions, involving the appropriate stakeholders, and thinking through edge cases can be the difference between a successful product and a cautionary tale.



When used correctly, AI makes interacting with digital products frictionless, removing barriers that separate users from programs by predicting user needs. This is achieved based on data, without the need for input from the user.


How we use AI to improve our products


As our digital footprints are getting larger, we’re producing more data than is possible for humans to parse. Due to that data influx and a significant increase in computing power, the use of AI to collect and analyze information in order to better tailor experiences to users has become more commonplace.


Many large companies are invested heavily in AI to increase engagement, retention, and sales. Digital assistants like Siri, Cortana, and Google Assistant expand the ways people can interact with their devices, increasing retention and engagement. Amazon’s recommendation engine, for example, is powered by an AI that has been partially credited for a 29% increase in the company’s sales.


AI is often used to help personalize experiences for users, learning more about them as they use the product. Google can anticipate, with growing accuracy, which search results will be relevant to a user based on their past search history, and predictive keyboards on iOS and Android can suggest words you’re likely to use even as you’re typing.


When used correctly, AI makes interacting with digital products frictionless, removing barriers that separate users from programs by predicting user needs. This is achieved based on data, without the need for input from the user. These insights can benefit both users and companies, as showing relevant data to users helps them engage with products more efficiently, and more accurate data and automation increases sales and other important metrics.



Why defining success metrics is important


When working with AI, a goal and parameters must be set in order to work towards a desired result. It may seem like just setting a goal, such as “increase sales,” without specifying parameters would be productive as long as it worked, but doing so might create other issues on your platform.


At the same time, running any kind of experiment requires defining the undesirable results just as much as it requires defining the desirable results.


If you were to hastily change items around, edit copy, and alter UI with the sole intention of increasing sales, you might find that success in one metric can be failure in others. For example, changing the buttons to neon green and using the copy “Get It” instead of “Purchase” might increase the likelihood that users on a product page will purchase the item.


However, this immediate benefit may not be tenable.


First, the vague copy and new color might not be on brand with your company’s visual and copy guidelines. Second, while the percentage of users converting might be higher, it may also decrease retention in the long run, as users might find the design and language updates off-putting.


This may seem extreme, but there are many high-profile companies that have had very problematic, public misfires with AI, such as Microsoft’s Tay chatbot, which used machine learning on Twitter to learn how to speak more naturally and instead learned conspiracy theories and racism. Amazon also had to shut down an AI they created in order to streamline their recruiting process by analyzing previous job applications because it showed sexist biases.


By not defining undesirable results, these companies accidentally created harmful experiences that did real damage to a lot of people. Had Microsoft hard-coded words or phrases to avoid and users that were better not to follow, they probably wouldn’t have had a brand-affiliated product that’s using hate speech to a significant following. Had Amazon been more careful about the dataset they gave their AI, or their own existing biases when hiring, they could have hired more qualified women and avoided a PR mishap.


AI might seem like something that lies squarely in the developer’s domain, but even when creating self-learning algorithms, it’s clear that UX designers are actually critical team members to involve in the process in order to create an experience that isn’t harmful to users or the brand.





Managing expectations for AI


Introducing users to technology–especially new technology–can be alienating. As UX designers, it’s our job to put them at ease and make AI feel like something that’s making life easier for them, on their terms. First and foremost, they need to be acquainted with the concept of AI and what it means for their interaction with your product.


  • Create transparency around use of data: Onboarding a user to an AI experience, ideally, should help clarify what information the AI uses to enhance their experience and how you protect that information. This may seem like an unnecessary step, but, with constant data breaches and abuses from large tech companies, people are becoming more wary of possible uses of their data. Being upfront about how and why you use their information can help alleviate that uneasiness and potentially even gain the user’s trust.

  • Introduce the role of the AI in the product: Before anyone starts using your product, it’s critical that they understand the scope of your AI’s abilities and how they will interact with it moving forward. Being acquainted with the type of actions they can perform with the AI will make it easier to persuade them on why they should use it and reduce the need to prompt them to use it later on in the process. Outlining your AI’s abilities will also get you and your users on the same page and help avoid a negative experience, such as where the user assumes the AI can do something it can’t.

  • Reinforce AI activity with indicators: When relevant, use indicators that will show the user when the AI is on or available. The best way to approach this may differ depending on how the user interacts with it. Siri plays a short sound and opens an overlay when opened, while Gmail’s predictive text shows up in light gray with a little indicator that the user can swipe, or if on desktop, the user can press tab to accept the suggestion.



A screenshot of an email with predictive text on Gmail
Gmail’s predictive text shows up in light gray with an indicator for users to accept the suggestion.


How to show AI to users


The way a user interacts with AI can vary greatly between different products. Some products use AI in the background to give the user the most relevant information it can find, like streaming sites’ recommendation engines, while others show AI more directly, like speech to text. Understanding the context of how your user interacts with your AI is critical to creating an engaging experience.



Background AI


When used in the background, AI usually gathers the user’s data in order to customize the experience to the information at hand. For example, Netflix uses AI to change items’ thumbnails to versions that the user will find most appealing.


Users understand that when they use a product, there’s a certain amount of information about themselves they are giving away in exchange, but there are limits to what people are comfortable with. If a user’s data is being pulled from sources they don’t expect, or if it’s being used in a way they hadn’t agreed to, they will feel that their privacy has been violated and lose trust in the product.


When possible, being transparent with users regarding when and how you use your AI and their data will put them at ease. If your AI is being used as a recommendation engine, make sure the users understand the context of the information they’re seeing. Instead of showing them personalized information without acknowledging the information source, preface it with “based on your likes” or “other people who liked this also liked…”



Foreground AI


Figuring out where and how user interaction with AI happens in your product will make or break it.


First, it’s imperative that all of the user’s interactions with your AI have a consistent tone. Between the presentation and the UX copy, there should be a clearly defined identity for this digital brand. Ideally, the user should understand what their interactions with the AI will feel like before they’ve even had one. Making sure the user feels familiar with it will also make them more likely to try it. No matter how powerful your AI is, if the user doesn’t know what they’re going to get out of an interaction with it, they won’t feel comfortable using it.


AI products are usually robust and have many commands and applications. Onboarding a user can’t always cover everything your AI can do, and we can’t expect a user to remember everything at once. Using contextual hints when appropriate will ensure that the user is aware of the different ways they can benefit from your AI, and can even help you promote actions you want the user to take.


A well-placed hint should be elegant and unintrusive–think the opposite of Clippy from Microsoft Word. Siri is a great example of how to show your user what they can do without it being intrusive. When Siri is triggered and there’s no data input, the voice UI will show a list of things you can do with it – a clear answer to someone who pulled up an AI and doesn’t know where to start.


Make sure your AI is only displayed when explicitly triggered by the user, or when absolutely relevant. Each time the AI is activated, users should be clear on the context for it appearing (such as an action they took), and it should help them achieve something they want. For instance, Gmail’s predictive text only displays when there’s a certain level of confidence it can help complete the user’s sentence. If it was constantly suggesting words as the user typed with wild inaccuracy, it would be perceived as a bad joke.



Siri's list of "Some things you can ask me" when not presented with data from users
When Siri is triggered and there’s no data input, the UI will present users with options of how to use the AI.


Your users come first


More often than not, introducing artificial intelligence into a product happens because it can achieve company goals and not because it will benefit the user. However, it’s almost always possible to implement features that will both satisfy the business’ objectives and improve the user’s experience.


Making sure UX designers are a core part of the team–as well as talking to real users to better understand their needs–will help you create an experience that users will feel good about. No matter what the reason is for implementing AI in your product, your users’ trust and experience are more important.


AI is a particularly difficult feature to navigate due to the sensitivity of the user’s information. There are few restrictions to the way companies can use information gathered from users, and many people have felt that their trust has been violated by products they thought they could trust.


If possible, letting the user opt in or out of certain data points will also help put them at ease. Understanding how their data is being used and how it will not be can both earn and keep your users’ trust.


The best implementations of AI help users with the actions they want to complete without forcing the developer’s agenda on them. As the digital landscape has grown, more companies have tried to take shortcuts to achieve metric growth in ways that have been harmful to their users. Using AI is not the way to get users to perform an action they’re not interested in performing - it’s a dark pattern. Artificial intelligence is only meaningful to a user as long as it’s in their interests. After all, an assistant that tries to push you in a direction you’re not interested in isn’t a very good assistant.


Finally, remember that any actions your user can perform with your AI need to be simple to communicate and remember. It’s easy to get carried away with artificial intelligence functionality, but like any feature, it has to be easily understood by the user. Keeping the features and messaging of the AI relatively straightforward can be the difference between users actually using it or feeling overwhelmed by it.