How User Research Shaped MathWorks’ New AI Quick Answers Feature
Today’s guest blogger is David Marshall. David is a User Experience Researcher supporting MathWorks Online Products, including the Community, Help Center, and search. In this blog he describes his recent experience supporting the new AI-generated Quick Answers feature, recently made available in our Help Center search.
Mathworks user experience (UX) group
Long before the popular design tool Figma was a figment of people’s imagination, when the internet was still called the “information superhighway”, MathWorks had a User Experience Group. We’ve grown since then - to over 100 people around the world, striving to understand our customers and design great experiences. We specialize in research and design, and work closely with developers to ensure that our products aren’t just useful, but are also easy to use.
So what does this work actually look like? You may have read Mike Croucher’s blog post about the new generative AI feature in Help Center search, or recently experienced it for yourself. In this post, I’ll share the UX journey that brought Quick Answers from concept to reality. It's a great case study of how an idea is tested with users and refined before becoming a fully-baked feature that is ready to ship.
Case Study: Adding generative AI to Help Center Search
The project kicked off with a desire to explore how AI could improve the experience of searching MathWorks Documentation - but we started with more questions than answers. This is an exciting moment for a researcher: it’s our business to ask useful questions, then go about answering them. So we started to chip away, 1 at a time, knowing that each would lead to new insights, discoveries, and of course questions to explore. Below are the questions we asked along the way, as well as what we learned.
Question 1: Why do people come to the MathWorks website?
Part of a UX researcher’s role is to keep the customer’s perspective front and center, as it’s often different from how people think about and use our products internally. When it comes to Documentation, previous research tells us that user goals fall into two broad categories:
- Find information you know exists. Maybe you’re looking up the syntax for a function you use regularly, or just want to double-check the required inputs and outputs. You know what you need; you just need to find it quickly.
- Learn how to do something new. Maybe you’re wondering if MATLAB has a function for your specific problem, or you want to learn the best way to tackle a new challenge.
For this project we had a slightly more specific question that could apply to both scenarios above:
Question 2: What do people search for?
With so many searches happening every day, it’s easy to see what people are searching for. When we analyzed recent search terms, 3 categories emerged:
- Function name searches: plot, randi, find, linspace
- Concept searches: simulink, machine learning, fourier transform, image processing with AI
- Natural-language queries: how to denoise an image with AI
Turns out most queries are short—just a word or two, often a function name. That’s a tough challenge for any AI! To understand what people really want from these different types of searches—and how AI could help—we turned to the best resource we have: our users themselves.
Question 3: For each type of search, what does an ideal response look like?
To answer this question we spoke directly to customers. We asked about recent experiences with our Documentation and watched them retrace their steps using a prototype that included AI responses alongside traditional search results. Several themes emerged regarding a helpful AI response: be brief and direct, include example code to see functions in context, and help people discover new features through AI-powered suggestions.
We also learned that each type of search does indeed benefit from a different kind of response. When someone searches for a function name they want details like inputs, outputs, related functions, and example code. When searching for a concept – think “Fourier transform” or “image processing with AI” – they want a list of relevant functions or tools. And for natural-language questions like “how to denoise an image with AI,” they prefer a customized explanation with example code—ideally drawing from multiple sources at once.
Question 4: Is the new feature clear and easy to use?
Figuring out how to make a useful generative AI tool was just the first part of this project. We then had to add it to the website in a way that felt intuitive, without getting in the way of existing functionality. That meant testing – and iterating – on the details. Here are a few highlights from the design journey:
We explored both top-of-page and sidebar layouts. People liked both, but many felt that putting AI answers at the top signaled importance and made them easier to notice. So we went with top placement.
The search box was another challenge. Our first version was small, but users wanted space for longer, detailed questions—especially since AI encourages more descriptive queries. We made it bigger and more inviting.
Headings were the final tweak. “Generative AI Search” sounded technical and wasn’t clear to everyone. Switching to “AI-generated response for: [search phrase]” helped people immediately connect the AI response to their query.
Outcome & Impact
Now when you search the MathWorks Help Center, you’ll see Quick Answers right at the top. Since launch there’s been a steady increase in the length of search queries—suggesting that users are asking more direct, specific questions to get the answers they need. Early feedback is positive, and we’re excited to keep improving the experience.
Try It Out—and Help Shape the Future
We’d love for you to try Quick Answers the next time you search the Help Center. Have thoughts or suggestions? Leave them in the comments below.
Even better—sign up for user research and you can be a part of making MATLAB, Simulink, and the MathWorks website even better. Participants are paid cash for their time, but the swag is a perennial favorite. In addition to informing the project at hand, there is always an opportunity to share additional feedback directly with the people who build the tools you use every day.
Thanks for reading, and happy searching!
댓글
댓글을 남기려면 링크 를 클릭하여 MathWorks 계정에 로그인하거나 계정을 새로 만드십시오.