In our test we instructed the chatbots exclusively via text: we deliberately avoided providing or editing code because the goal was to require no programming from the user. We also wanted to see how well AI solutions that communicate surprisingly well in Hungarian can handle this kind of task. It is worth noting that although these tools try to be helpful, some technical sense is still useful to achieve an acceptable result.
The task: a simple expense-tracking web app
We asked four platforms to produce a simple expense tracker web application: OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini and the vibe-coding-focused Lovable. The target functionality was straightforward: handle incomes and expenses, allow categorization, and be comfortable to use on mobile — ideally installable as a Progressive Web App (PWA).
Results and comparison
-
ChatGPT (OpenAI): the produced webapp proved largely unusable in our test. For example, it did not allow entering income and expense separately, so basic functionality was missing. Overall, the output was not acceptable as a ready-made solution.
-
Claude (Anthropic): delivered the most comprehensive and visually rich result. The web application it generated displayed financial data clearly, augmented spending bars with messages (for instance encouraging or warning notes), and used icons for spending categories. The user interface felt well thought-out and easy to use; we found no defects.
-
Lovable: Lovable's vibe-coding approach produced a cleaner, more minimalist UI, but nonetheless resulted in a practical and usable webapp. One convenience was that it displayed a live preview of the generated web application alongside the chat field. Within the daily free tier it even suggested a design reminiscent of Apple's Liquid Glass aesthetic.
-
Google Gemini: Gemini created a solution that met our requested functionality, but even after repeated prompts it only supplied code rather than a ready-to-run HTML page. Assembling and running that code is not complex, but it lacks the "turnkey" experience offered by some rivals.
PWA and mobile deployment
Regardless of platform-specific performance differences, our general finding is that a very simple spending-tracking web app can be produced with this text-first method. Each tool could also guide us through pinning the app to a mobile home screen as a Progressive Web App; PWAs preserve entered data and provide an app-like experience. The exact steps to create or install a PWA differ somewhat between Android and iOS, and the AI assistants sometimes needed multiple iterations to clarify a working method.
Who benefits from this approach?
This workflow is particularly attractive for users who want a quick, custom tool without web development skills. However, outcome quality varies significantly by platform: some AI systems generate immediately usable interfaces, others provide code snippets or less workable suggestions.
Conclusion
It is indeed possible to quickly produce an expense tracker and turn it into a PWA using AI chatbots through text prompts alone, but users should expect to apply a bit of judgement and occasional manual steps. In our test Claude and Lovable produced the most directly usable results; Gemini produced code that required user assembly; ChatGPT did not deliver an acceptable ready-made webapp.
