Michigan
ChatGPT
Where this started
Michigan built the AI. Students could not tell the four tools apart.
Michigan ChatGPT is the university's own AI workspace. It holds four tools: U-M GPT, Maizey, Go Blue and the GPT Toolkit. The homepage asked you to pick one of the four before it told you what any of them did.
I used this platform as a student before I ever redesigned it. What surprised me when I asked other students about it is that nobody was intimidated by the AI. They were stuck on the names, because Maizey, Go Blue and GPT Toolkit do not tell you what is behind them. Students could not pick a starting point, and when that takes too long they open ChatGPT in another tab instead.
How do I get someone to the right tool before they give up and open another tab?
Project Location
Ann Arbor, Michigan
Nothing on the page told you which tool to pick
- 01
Say what each tool actually does, without turning the homepage into a wall of text.
- 02
Pull four different navigation patterns into one that behaves the same way everywhere.
- 03
Serve the person who has never been here and the person who comes every day, on one screen.
I asked students on campus, and none of their problems were about the AI
I ran surveys and interviewed students who use campus technology every week. Their answers came back to the same four problems: they could not say what any one of the tools was for; they were not sure how to get from one tool to another; they opened ChatGPT instead, because guessing there was easier; nothing about the interface looked like the sites they already use. All four are about getting to the tool rather than using it, so I focused the redesign on the wording and the first screen.
- The words were the barrier, not the technology. People who use AI every day still could not tell these four tools apart.
- Familiar beats clever. When a page behaves like the pages someone already trusts, they stop bracing and start using it.
- When navigation changes between tools, people assume they have left the site and quit exploring.
I used to think research meant finding out what people want. Everyone I asked wanted "something simpler", which is not something I can build. The useful part was watching where people stopped and hesitated. People describe what they already did accurately, and they guess badly about what they would do next, so I pay attention to what they have already done and ignore what they say they would do.
If this is my first time here, I need to:
- Work out in one screen that these are the AI tools Michigan built
- See what each one is for, and which one fits what I came to do
- Move around without feeling like I am one click from breaking something
If I am here every week, I need to:
- Get straight into the tool I use, without reading anything first
- See what has changed since last time
- Reach login, documentation, or an actual human in one click
I designed for the student who had never heard of Maizey
I went through the old site and wrote down every place where I had to stop and think. Most of them were decisions the page asked me to make before it had explained anything. Three ideas shaped what I did next. Jakob's Law says people expect a new site to work like the sites they already use, so I did not invent any new patterns. Hick's Law says every extra option slows the choice down, so the homepage offers four things instead of fourteen. Then I reordered the page so the first thing you see is what the tools do, rather than a menu of their names.
Naming the laws is the easy part. To be transparent, I had already decided the homepage should hold four tools, and I used Hick's Law to justify that decision afterwards. I think that is fine as long as you know you are doing it, and a problem as soon as you forget. I use design principles to explain a decision to other people after I have already made it.
Eight ideas, drawn quickly, most of them cut
I sketched eight directions before I committed to one, including finishing your sentence for you, a video analyzer, translation, and exporting a chat as a PDF. I drew them quickly and badly on purpose. The ideas I kept were the plain ones: how much fits on the screen before you scroll, and whether you can tell what a tool does from its card alone.
I keep drawings rough for as long as I can. A rough sketch makes it easier for other people to tell me I am wrong, and the rougher it is the more directly they say it. As soon as a screen looks finished, people soften their feedback and it stops being useful. Staying rough for longer is the easiest way I have found to get honest feedback.
Every tool on one screen, so nobody has to guess
The homepage opens with one line explaining what the platform is, then four cards, one per tool, each described in words a first year student would use. You can read all of it in a few seconds, which is about as long as anyone was going to spend before giving up.
The same navigation inside every tool
The sidebar groups tools, resources and support into categories that stay in the same place no matter which tool you are using. The tap targets are bigger, the labels are readable, and the same icon means the same thing everywhere. People who come back every week go straight to where a thing was last time, so nothing moves between one tool and the next.
Answers written for people at this university
Responses are tuned for academic, research and campus questions, so the answers are about this university rather than generic ones. It also helps people who have never written a prompt before. You should not have to be good at talking to AI to get something useful out of it.
Reading it in the language you think in
Translation mode lets multilingual students work in whichever language they are fastest in and switch back when they need the official campus wording. A lot of students here think in another language first, and making them do financial aid paperwork in their second language costs them time nobody else has to spend.
For when you cannot look at the screen
Audio feedback reads responses and confirmations out loud. It matters most to students with vision loss, and it also helps anyone walking to class, cooking, or finished with reading for the day. I keep finding that accessibility work turns out to be convenience work for someone else as well.
Layout
The page structure
Columns
4
Type
Stretch
Gutter
16
Margin
0
Width
Auto
Columns
12
Type
Stretch
Gutter
12
Margin
0
Width
Auto
How it turned out, and what I would change
- A homepage where all four tools are named in plain language and sit in one row
- Cards you can scan in a few seconds, instead of a paragraph you have to read
- Michigan branding that is unmistakable without taking over the screen
- Bigger targets, clearer labels, and type you can read on a phone outside
- Answers tuned for campus questions, so a first time user does not need a good prompt
What I wanted was a platform people can understand in a few seconds, where the first visit and the fiftieth both start in the same obvious place.
What I did not expect to learn is how much trust comes from the interface being familiar. A student will only type a real question into a box owned by their school if the page looks like the sites they already use. Since this project I spend much longer deciding what to call things than I used to.
- A real documentation section, instead of links that go nowhere
- A homepage that remembers the tool you actually use
- A way in from Canvas and the rest of the software students already have open
These are tools my friends open between classes, and a few hundred words on a homepage decides whether they use them or give up.
What I took away is what it means to redesign something people already use. Nobody was waiting for a new homepage. Students already had a habit, which was opening ChatGPT in another tab, so I was competing with that habit rather than with the old design. It meant the homepage had to be faster to read than opening a new tab, not just clearer than the page it replaced.
This was the first time I owned a project from the first survey to the last screen, and I felt the difference. There was nobody to hand a decision to. I also felt exposed putting a redesign of my own university's product online, because the people who built the original work on the same campus. I tried to write this page in a way I would be comfortable saying to their face.
How I would change this today with AI
I accepted the premise that a student has to pick one of four tools before they can start. Today I would argue with that. One box that takes your question and sends it to the right tool would remove the choice completely, and that is a better answer than explaining four names more clearly. What I designed makes the menu understandable. Taking the menu away would be the stronger fix.
Two things would worry me. If it sent a question to the wrong tool, that would be worse than a menu, because the student cannot see what went wrong or how to correct it. And students still need to know which tool answered them, because the four handle university data differently, and that is not a detail you get to hide to make a homepage simpler.
