The first AI draft looks coherent. It has a large headline, friendly colours and an eye-catching button. That is precisely why a second question is worth asking: why is this particular action at the top? A convincing screen can make an arbitrary priority seem so self-evident that nobody discusses it any more.
On 18 March 2026, Google Labs described an expanded Stitch approach to AI-assisted UI design on a shared canvas. This is an older tool announcement, not a current news item from October. For this practical article, we use it as the starting point for a design exercise of our own: three directions that differ in content, all built on the same confirmed facts.
A variant needs a different purpose
Our fictional neighbourhood library has six pieces of content: book search, opening hours, events, registration, location and information for a first visit. The first direction prioritises search. The second starts with the next event date. The third helps people who have never been there.
These differences go further than a change of corner rounding or accent colour. Each direction answers a different assumption about why people visit. Once the assumption is spoken aloud, it can be questioned. Perhaps a library with many regulars should be weighted differently from a new institution that first has to help people find their way.
Keep the facts fixed while you do this. A draft must not seem more convincing simply because it invents longer opening hours or leaves out a difficult condition of participation. Design organises the information; it does not get a licence to make it fit.
A brief that really asks for three directions
You can complete the following template, which is our own, with your confirmed content. It describes a work brief, not a Stitch output that has actually been produced.
Develop three different entry points for the same library page. A prioritises the book search, B the next event date, C orientation before the first visit. Use only the facts provided. Keep the opening hours and conditions of participation unchanged. For each variant, explain which task is supported first and what moves further back as a result.
The last question matters. A priority costs space and attention elsewhere. Anyone who only asks about the advantages of a variant will easily get three positive justifications. Making the downside visible helps to take the decision seriously.
First ask for the order and the relationships between pieces of information, not for additional decoration in the image. A simple schematic often already shows whether the directions differ in substance. Colours and illustration can then support the chosen intent.
A shared task makes the comparison fairer
Before judging, choose a specific task: find a suitable event date and work out how you can take part. Look at the same task in all three versions. Where is the date? How is the condition made visible? Where does the highlighted action lead?
Write down observable difficulties. “Registration was only recognisable after rereading” is something you can work on. “Looks old-fashioned” first needs a more precise explanation. This is not about banning taste. It simply should not be the only justification for a decision about information.
An editorial review of this kind is not yet a usability study. If people later actually try out the variants, the tasks, environment and observations have to be documented. This article does not pretend that any test sessions took place or that there are any results.
The choice becomes a precise hand-over
After the comparison, note the preferred entry point, its main task and the details that are still required. Add any open questions. Perhaps the events direction works well for a campaign page, while the regular home page should prioritise search. A reasoned choice may be nuanced.
The hand-over text is more valuable than an isolated screenshot. It explains which relationship between content and interaction should be preserved. A developer who takes over next, or another AI, then has to guess less from the appearance.
A UI draft is still not a complete application. Data connections, behaviour and accessibility need work of their own. The strength of the exercise lies before that point: it makes alternatives visible while changing the basic direction is still easy. The first attractive draft then remains one possibility among several, instead of quietly becoming the only conceivable path.
Sources and status
Research and documentation as of 25 October 2026. Introducing vibe design with Stitch.
