Why it matters?

Design plays an integral part in ensuring that people have a positive user experience across your product, the same applies when using an AI product.

For many people, your product could be the first place they encounter GenAI and understanding how to use it for their day-to-day work.

If we don't address these concerns directly through user education and design, we risk creating a bad first impression that will make it difficult for people to recognize and benefit from the value that GenAI can bring to their work.

There are specific considerations we need to account for when designing for work, like increased accountability for data and sensitive information.

They aim to act as a foundation and help teams solve for common risks involved when working on AI features for work

1.Anticipate Risk and Lead with human reviews:

AI is not perfect and will make mistakes. To ensure accuracy and prevent the spread of misinformation, we want to require human reviews before content is seen by a wide audience. Content should only be published when you intentionally press a publish or post button.

An example of this was a post summarising project. Posts were automatically summarizing other people's content inaccurately, which posed the risk of spreading misinformation.

Collaboration with AI is best for creative use-cases, whereas repetitive tasks call for automation. Regardless of approach, always put control in people’s hands.

🤔 Consider

Allowing people to review any content before it gets published to a wider audience. Allow the freedom to experiment with AI features without penalising them.

🚩Avoid

Automatically publishing content on behalf of moderators and producing anxiety for people around losing progress or not giving the ability to edit outputs.

✅ Instead

Give people an easy way to publish content, undo mistakes, or change directions whenever they want. Allow people to have clear ways to give feedback or get help.

2. Keep users informed and build trust:

People should feel that GenAI’s presence is useful, not invasive. Education helps people understand this.

Using Modals, Text, Help Articles for Education

When users encounter a Gen AI feature for the first time, it can be helpful to have a modal pop-up that explains what the feature is and how it works.

However, it is not necessary to display the modal every time the user uses the feature. Use the modal sparingly.

Add references when providing AI outputs:

Users need to be able to determine the correctness and quality of generative AI responses. They often need to rely on other tools and sources to verify responses. When designing generative AI experiences, consider how you might be able to help the user in this verification process.

🤔 Consider

Providing onboarding or educational content for users who encounter a Gen AI experience for the first time. This content can help them understand how the feature works, how it improves their efficiency, and what guardrails are in place.

🚩Avoid

Jamming too much educational information in top-level experiences.

✅ Instead

Instead lean on modals, bottom sheets, “Learn more” links, and the like to give space for explanations around how AI works.

3. Lead with the problem:

When brainstorming generative AI products/tools/features, think about the role you want it to play for the user. Here's three categories of use cases for generative AI:

⚡️ Super Speed

Helps the user perform tasks they already know how to do, but faster. For example, automatically adding chapters to a video.

❤️ Help Me

Helps the user to do something that they want to do (user already has the intent) that they aren't sure of exactly how to do it (e.g. write a summary of the post i'm creating).

💡 Inspire & Unblock

Helps the user get started (combating blank post/page syndrome with a rough draft) or to explore possible directions (by quickly generating multiple concepts/options).

🤔 Consider

Linking to existing workflows people already carry out in the work context and try to make them more efficient.

🚩Avoid

Avoid simply pasting in a new Gen AI experience just because it's a fun, new technology. Make sure it solves existing problems..

✅ Instead

Create solutions within the existing JTBD framework that fall into the categories of "super speed," "help me," and "inspire and unblock."

4. Create familiarity

A Gen AI experience should look cohesive across product tools and clearly communicate that AI is being used to the user.

By following the same patterns and guidelines, we can help achieve familiarity and reduce the learning curve and build confidence in using our Gen AI products for work.

🤔 Consider

Demystify new capabilities and technologies through familiar UX experiences using existing design system components and styles.

🚩Avoid
A user experience that requires the user to learn a completely new interaction. Avoid creating custom icons, components, and interactions for users to learn. This adds an extra learning curve on top of learning how to use the AI experience, creating additional friction.

✅ Instead

Leverage an existing behaviour, or bring in familiar patterns / metaphors to make the experiences easier to grasp.

5. Don’t oversell AI

Gen AI is a new feature that everyone wants to try and implement in their product; however, it should be part of a unified experience that doesn't compromise other existing features users use.

🤔 Consider

Using existing components. Gen AI should feel like an embedded cohesive experience.

🚩Avoid
Styling and interactions that make the Gen-AI experience stick out from the rest of the product. Don't have different colour palettes, over the top interactions and animations, this can compromise the existing experiences of features of your product.

✅ Instead

Use existing styling and guidelines. Reach out the design system team if there's something new you want to implement.

What's next?

I'm excited to embark on the design journey for the next iteration of Gen AI for work products and to keep pushing the way customers can work, last year was the first year Gen AI took off, we're really just at the start. I'm looking to evolve these principles as we learn more in the space.