How to test your AI-built app like a real customer (the live app, not the preview)

Short answer: open your published app in a private window on your phone, sign up with an email you've never used, do the one thing the app is for, and pay if it takes money. Then do it again after every publish. That's the test the builder's own checks don't do: they run on the preview, and your customers never see the preview.

When people pasted 28 of their own live apps into our free test (27 September to 3 October 2026), 26 of them had a problem a first customer would hit.

The preview is not what your customers get

Builders now test your app for you, inside the editor. Lovable's browser testing clicks through your app like a user, Replit Agent tests the app it's building, and Base44 has a testing agent. That's useful. But by their own docs, all three test the preview or a separate test copy, not your live published app (Lovable, Replit, Base44). The live app is different in four ways, and those four are where first customers get stuck:

• The address. Sign-in, email links and payment redirects have to trust your live domain. In the preview they trust the preview's address.
• The keys. Live payment keys, email sending and database rules are often set up only for testing.
• Real email. A sign-up email has to actually arrive, from your own domain, and not land in spam.
• Real phones. Small screens, slow networks, and the in-app browsers of Facebook and Instagram, where links shared in groups and messages open.

The 15-minute first-customer test

Do this on your live address, in a private (incognito) window, ideally on your phone.

1. Open the link the way a stranger would. Paste it into a message to yourself and tap it. Does the first screen say what the app is, or does it go straight to a login wall?
2. Sign up with a brand-new email. Does the confirmation email arrive within a minute? Does its link bring you back signed in on the live site, not on a preview or localhost address?
3. Sign out and sign back in. Then refresh any inner page. A "Page not found" on refresh is a hosting setting, not your code.
4. Do the main thing the app is for, start to finish: create the booking, save the entry, upload the photo. Refresh and check that it's still there.
5. Pay, if the app takes money. Use live mode with a real card and refund yourself. Test mode passing proves nothing about live.
6. Open a second account in another private window. Can it see anything of the first account's? It shouldn't. Of the 28 apps people pasted into our free test, 5 let anyone read data that should be private.
7. Read every page a customer can reach: footer links, privacy policy, terms. Look for unfinished content: template legal pages, placeholder text, a wrong language, prices that contradict each other.
8. Share the link in a chat app. Does it show your app's name and a picture, or a blank card?

Write down each problem as what you did, what you expected, what happened. That one sentence is what your builder's AI needs to fix it.

Turn each problem into a fix your builder can do

Paste this into your builder's chat for each problem, one at a time:

On my LIVE site (not the preview), this happens: [what you did] → [what happened]. I expected: [what you expected]. Find the cause first and tell me in two sentences, without changing code. Then fix only that cause, and tell me exactly how to check it on the live site.

Then publish, and repeat the step on the live site in a private window. "Fixed" means you saw it work on the live app, not that the AI said so.

Tools that test for you

• Your builder's own testing (Lovable browser testing, Replit Agent app testing, Base44 testing agent). Built in, included in your plan or paid in credits, and it can fix what it finds. It tests the preview or a test copy, and it's checking its own work.
• QA tools for teams (for example Momentic, testRigor, QA Wolf, Sofy). Powerful, and built for developer and QA teams that write, run and maintain test suites, usually inside a release pipeline. Momentic has a free tier and a $125/month plan; Sofy's AI testing agents start at $49/month; testRigor and QA Wolf's managed service don't publish a price (each from its own pricing page, October 2026).
• VibeFix. An AI agent that opens your live app on a phone and a computer, signs up with a real inbox, does the main thing and goes to checkout, then tells you what a first customer would hit. It works whatever you built the app with, needs no account and no access to your code, and the first test is free. It's an independent check: it didn't build the app, so it isn't grading its own homework.

Do it after every publish

Every publish can break something that worked: a new feature, a "fix", a setting that didn't carry over. A first-customer test is only true until the next publish.

Check yours now: paste your app's link at vibe-fixer.com. Test your live app free, get the exact fix for one problem to paste into your builder, also free, then re-test the live app to prove it worked.

Related: Works in the preview, breaks after deploy · Login broken after publishing · Sign-up emails never arrive · Who can fix my vibe-coded app?