I want to share a deeply frustrating experience I just had with v0 by Vercel. If you are planning to spend real money on credit top-ups for this platform, please read this first.
I’ve been working in web development for a long time. I know how to structure precise, sequential, and technical prompts. This wasn’t a case of “bad prompting.” Despite giving direct instructions, v0 entered a destructive loop that completely drained my paid credits (which I topped up multiple times), left me with a broken project, and straight-up lied about the output.
Here are the facts, completely verifiable by my logs and repository history:
-
The “Framework Migration” Lie (Astro vs Next.js): I explicitly and repeatedly ordered v0 to migrate my project to Astro (Single Page Application) to optimize SEO. In the chat, the model textually confirmed it had completed the task, even detailing how it integrated
@astrojs/sitemapand structured the new.astrofiles. When I checked the connected GitHub repository, the project was still 100% Next.js. The model completely fabricated the migration, did zero actual framework work, but happily charged me for the tokens used to write the lie. -
The Breaking-Code Loop: Instead of refining the existing frontend, v0 started rewriting massive chunks of code while introducing glaring bugs. It completely broke the mobile responsive layout (causing a horrific horizontal overflow) and completely emptied the “description” fields inside the tool detail modals. Why? Because it generated HTML attributes on the card component (e.g.,
data-desc) but then looked for a completely different variable name in the client-side script (data-description). It left me with an empty UI and stray<hr>lines floating on the screen. -
The “Brick Wall” Support Response: When I opened a ticket with Vercel Support—documenting the severe hallucinations where the model claims to commit code changes it never actually pushed—I received the classic canned corporate template: “It’s a generative AI tool, it can hallucinate, credits are non-refundable, please use the feedback button.”
Let’s be clear: There is a massive difference between an AI making a coding mistake that you can easily prompt to fix, and a system entering a repetitive billing loop. v0 forced me to top up my balance, burned through roughly €15-20 in minutes by endlessly rewriting static arrays or hallucinating structural changes, and then locked me out with an “out of credits” error exactly one second after pushing a broken UI to production.
If you are using v0 to scaffold a quick prototype from a screenshot, it’s a neat tool. But if you trust it to handle structural modifications, data refactoring, or production updates on an active repository, it will bleed your wallet dry and wreck your codebase.
Always double-check your actual GitHub commits riga per riga, because the v0 chat interface absolutely lies.
Thanks for your attention,
Marco Ludovico Tolomeis from Florence Tuscany Italy 
Founder & Webmaster
P.S. If you want to see the live results of this chaotic generation process and how the UI layout behaves after these automated loops, you can check the production site here: https://imagemoz.com (the main platform holding our photographic resources and AI directory project).