When assessing a student project made with an AI app builder, look for understanding, decisions and testing. A polished interface or a fast build tells you little about those things. Templates and AI assistance can be legitimate parts of the assignment. The useful question is whether the student can explain the result, evaluate its limits and make a purposeful change.
Set the learning goal before judging the app
Decide what the project is meant to demonstrate. In a programming class, students may need to explain data handling or write a particular function themselves. In a history lesson, the goal might be selecting reliable sources and organizing them for an audience. A working app is evidence of a finished product; it does not, by itself, show that either learning goal was met.
Make the rules for templates and AI assistance explicit before students start. Specify which tasks may be delegated, what students must do independently and how they should acknowledge help. Using a template should prompt a discussion of what the student adapted and why, rather than an automatic penalty.
Access belongs in the brief, too. Check that the required features and revision allowance are available to everyone. If a paid Lovable plan is being considered, Lovable discounts from HostAdvice can be one place to check for offers, with eligibility and current terms confirmed before purchase. Provide an equivalent route for students who cannot pay. Subscription spending should not become a grading criterion.
Ask students to explain a choice, then change a feature
Take a club sign-up app. Ask the student why they limited each club's places, what information the form collects and what happens when a club is full. Follow one answer into the app: “Show me where that limit is set.” Look for an explanation that connects a choice to a user need or a concept taught in class.
Then request a small change within the student's taught skills. They might reduce the capacity, add a waiting-list message or make a field optional. Before they edit, ask what should change and what should stay the same. Afterward, have them demonstrate the result and check a related feature.
If AI use is permitted, a successful prompt alone is not the whole demonstration. Ask the student to explain what the generated change did and whether it matched their request. If the lesson requires independent coding, assess that part separately under the agreed conditions.
This approach has a precedent in TEQSA's assessment guidance for higher education, which includes interactive code reviews among its examples of assuring learning. Adapt the task to the student's level and provide agreed accommodations. A hesitant explanation or a failed edit identifies something to investigate or teach; neither is a verdict on misconduct.
Examine the testing and the help behind the result
Ask for a test the student actually ran, the expected result and what happened. For the sign-up app, try an empty required field, a duplicate entry and a club at capacity. “It worked” is less useful than an account of which input they used, what they observed and why that result was acceptable.
A test can also reveal a limitation without making the whole project a failure. A form may reject a blank field in the browser while lacking checks on the server. MDN's form validation guide explains why browser checks alone are insufficient. At the appropriate level, ask students to distinguish a classroom prototype from an app ready for real users. Use fictional personal details during testing.
Request a brief account of AI assistance: the tool used, the task it helped with and an output the student checked, changed or rejected. A relevant prompt or version-history excerpt can support that account, but a long log is not proof of understanding. Messy drafts are not proof either. Use the records alongside the explanation and demonstration.
Keep feedback tied to the learning goal: “You explained the capacity rule clearly, but you have not tested duplicate entries.” The student's next task is to run that test and explain what the app does.