Google Bard UX Audit: A quick review to improve the AI interface
Since the launch of Google Bard in its experimental phase, many UX professionals have been eager to try out this new tool and evaluate its ability to provide a high-quality conversational experience.
I decided to analyze the usability and user interface of Google Bard to help other UX professionals and enthusiasts better understand the potential of this new tool.
I will discuss my experience with Google Bard, evaluating its ease of use, effectiveness and efficiency, and highlighting its main features and considerations when evaluating artificial intelligence-based conversational tools.
This analysis is a valuable resource for individuals seeking impartial feedback on the user-friendliness of Google’s Bard and hoping to gain knowledge about optimal UX techniques for AI-powered conversational devices.
Cognitive Learning
People learn through observation, experience, and interaction with their environment. When people are exposed to familiar and predictable stimuli, such as Google's "I'm Feeling Lucky" or "Google Search" button, they tend to develop automatic expectations and behaviors, making it easier to interact with the interface. That is, one button shows the result directly, and the second a list based on the searched term.
It turns out that the "Google It" button in Google Bard does not do the direct Google search, even though the label suggests it.
There are two clicks before the action of searching Google (the first one activates a segmented pre-search, and only after the second click is the user directed to Google). For this reason, the user's expected behaviour can be frustrating. For example, the user may expect this button to do a direct Google search but find that he has to perform two additional actions before seeing the results.
Confirmation bias
Only three options appear after the user clicks the "Google it" button. Besides limiting the search context, it is unclear to the user what criteria Google uses to include related topics.
Despite the learning language model’s (LLM) excellent data processing, it is essential to remember that these models are not perfect and are subject to errors. However, it is possible that first-time users confuse Bard with a traditional search engine and do not understand that its bias can influence the responses generated by the model.
Using only three limited topics in the search suggestion tool, Bard can lead to a biased selection of information and options for the user.
This is because confirmation bias is related to people’s tendency to search, interpret and remember information selectively, preferring information that confirms their previous beliefs and ignoring information that contradicts them.
Interrupted task flow
Also, in this context, if the user wants to search for the exact term used in Google Bard (in my example, "Create a list of top 50 companies based on the Forbes Global 2000") as the "Google it" button suggests, they cannot do so in the tool. They select the text and use control C + control V, though. The user is forced to switch tabs to complete the action, which is unpleasant for a tool in an experimental stage trying to maintain a smooth experience and consolidate itself in the AI industry.
Placeholder is hard for lay users to comprehend
The term "prompt" is not a standard or familiar term to all users, especially those unfamiliar with programming or computer language. This can lead to confusion or uncertainty about what is expected of the user when entering a "prompt".
Lack of clarity: Using the term “Enter your prompt here” in the Google Bard search field is not ideal regarding UX best practices because it is a generic and unclear description. The placeholder could be better for more clarity and guidance. It must be apparent to the user if it is a keyword, phrase, or complex question.
Affordance problem: As Alphabet Inc’s flagship product, Google Search has been the industry standard for years. Therefore, when designing Google Bard, a Large Language Model that is part of the same holding, it is crucial to consider that users may have certain expectations based on their experience with Google Search. To ensure a good user experience, it is essential to provide clear information on how the search system differs from Google Search, especially given that many other competitors, such as Bing, are already integrating the two into a single experience.
[UPDATE] New feature soon: Google Search’s AI features can distill information and offer insights that enable you to see the big picture more clearly. Seeing the next steps for Google’s AI-powered search engine will be exciting!
Lack of guidance: The description needs to guide how the user should enter the prompt. Google Bard uses a couple of hints on the first page to help contextualize the use, but the feature does not replace the placeholder function, which should be direct and instructive. A more transparent, more user-oriented description would be more appropriate.
While Google has provided documentation explaining the purpose of Bard, more is needed for users who land on the page directly. Some users may require additional context or guidance to understand the meaning and functionality of Bard. As such, it may be necessary for Google to consider implementing more user-friendly onboarding processes or providing clearer messaging to ensure that users can fully leverage the benefits of this innovative tool.
Inadequate response for lost connectivity
When a user faces connection problems, the interface is expected to provide an immediate and helpful answer to prevent frustration and potential abandonment of the platform.
When there is a connection loss using Google Bard, the interface can present a long wait time to display the error message. This can lead the user to believe that the application is stuck, is elaborating the response, or there is no response to their command. This lack of immediate response goes against the Law of Feedback, which states that the answer should be provided reasonably to keep the user informed of what is happening.
In the example below, it took almost 30 seconds to wait for a response back from the tool, which could not complete the task because there was no internet connection.
Providing clear and immediate feedback to users in response to their actions is essential. When the platform does not provide an adequate response to the lack of internet connection, the user can become confused and frustrated, which can damage the user experience and affect the perception of the tool.
To solve this problem, it is possible to implement techniques such as loading indicators and customized feedback messages for different types of errors.
Room for improvement: Sparkle resting
If the indicator is resting, why is it in color motion? 🤔
Sparkle’s animation may look nice, but the continuous movement of the colors may give the impression that the user is interacting with a dynamic field, despite the animated icon feature being used after the tool responds to the user (resting state).
We tend to stick to familiar patterns and forms of interaction that we already know and adapt to them quickly according to the Law of Similarity.
The fact that the animated element looks like a macOS Spinning Wait cursor can create more uncertainty; some users may think that the page or chatbot is still loading rather than being in a “resting state” (a hypothesis that can be tested).
Room for improvement: Sparkle Thinking
One UX law that addresses the use of typing animation is the Law of Continuity; in other words, the visual changes that occur in response to the user's actions are consistent and predictable.
In broad terms, the main concern would be the synchronization between the animation and the response generation rather than an automatic gif. This can help the user understand whether the response is processed (creating information) or not (error / no connection). This will ensure that the user experience is always positive and consistent.
The idea of using a typing animation is to ensure that the interface creates a sense of logical continuity. The typing animation gives the impression that the user interacts with the search field or form in real time, as in the example above.
Room for improvement: CSS property
Inappropriate use of the CSS property impacted the usability of interface elements, including the avatar and the scrollbar.
The problem in Google Bard's scrollbar occurs due to its location within an element with the "overflow" CSS property set to "hidden". This property is used to hide content that exceeds the boundaries of the main component but also has the side effect of suppressing the scrollbar when displayed.
Using the "overflow-x" property instead of "overflow" could solve this problem by controlling the visibility of the horizontal scrollbar separately from the vertical scrollbar to ensure that the scrollbar is displayed properly.
As for the avatar, a tweak in the CSS code ensures that the avatar is sized correctly and that there is enough space around it to ensure it is obvious.
Room for improvement: Sticky header
One of the improvements that could be made to Google Bard is implementing a fixed question at the top of the interface. This would give users a better understanding of the task and the ability to navigate the tool more efficiently, as the user's question would remain fixed at the top of the screen, even when scrolling through the page.
The navigation header or menu remains fixed at the top of the screen, even when the user scrolls down the page. This allows the user to easily access the navigation options without scrolling back to the top. In addition, the user can make faster and more effective decisions by providing a fixed question at the top of the interface, such as reading and editing it.
Room for improvement: Copy button
The fact that the user needs to access the three-dot icon to copy an answer in Google Bard can be considered a usability flaw.
When users use a menu option frequently, it can be removed from the three-dot menu and placed in a more prominent position in the interface. It is crucial to reduce the cognitive load on the user whenever possible.
In the case of Google Bard, the option to copy the answer is a necessary action that the user may want to perform regularly. Therefore, this option is recommended to be easily accessible via a clearly visible button.
Final thoughts
Due to the experimental phase of Google Bard, this is a valuable opportunity to provide feedback and contribute to the development of the tool.
By implementing changes that prioritize simplicity, clarity, and user-centric design, Google Bard can become an even more effective tool for users in their daily lives.
To create designs that fulfill users’ needs and surprise and delight them, it is crucial to strike a balance between sticking to established design principles and exploring new directions that challenge conventional boundaries.
Disclaimer: I'm just a UX Designer curious to explore emerging technologies. :) Although I have analyzed the usability of Google Bard in its beta phase, my observations are based on my own experiences and opinions. The perceptions concerned with usability tend to fluctuate based on factors such as the user’s identity and conditions during evaluations. Opinions are my own.
