We checked 44 launched apps whose owners asked: 84% had a problem a first customer would hit

Between 27 September and 19 October 2026, an AI agent used 44 live, published web apps the way a first customer would. Each one was checked because a visitor pasted its link into VibeFix's free check. 37 of the 44 (84%) had at least one problem a first customer would hit, and half of them had a problem that stops the customer cold. Almost none were dramatic crashes. They were dead ends, and the owner never hears about them.

Find out in 2 minutes which of these is breaking your app. Test my app, free

What we found

• 37 of 44 apps (84%) had at least one problem a first customer would hit. 7 came through with nothing to report.

• 22 of 44 (50%) had a problem that stops the customer cold: sign-up or login fails, payment does not work, a page errors or never loads, or a main button does nothing.

• About 2 problems per app: 90 problems reported across the 44 apps (2.4 per app among the 37 that had any). One app can have several, and one check reported up to five.

• 5 of 44 (11%) let anyone read data that should be private, without signing in.

The problems, by how many of the 44 apps had them

• Broken on a phone, 12 apps (27%): screens that do not fit, buttons too small to tap, a page shrunk tiny.

• Unfinished or wrong content, 10 apps (23%): placeholder text, wrong language, labels that read "undefined", a status customers cannot understand.

• Sign-up, login or the verification email, 10 apps (23%): a new person cannot finish getting in.

• Errors, blank pages and dead links, 8 apps (18%): pages that error, open blank, or links that go nowhere.

• Private data or settings open, 8 apps (18%): open data, keys or admin pages, and tracking that runs without the consent the page promises.

• A button or form that does nothing or does not save, 7 apps (16%).

• Payment, prices or checkout, 5 apps (11%): payment cannot be completed, or plans and prices contradict each other.

• A confusing first screen, 4 apps (9%): popups or walls before the app.

• Slow pages, 2 apps (5%). Other, 2 apps (5%). Search visibility, 1 app (2%).

Why it happens

Builders test what they can reach, usually the preview. Every problem above was found on the published app, with its real address, real sign-up email, real payment settings and real phones. Those are the things a preview does not have.

Possible causes

• A setting the preview had and the live site does not, such as an address, a key or an email sender.

• A sign-in service that does not know the live address, so new people land on localhost or back on the login page.

• Content the builder left in as a placeholder, which looks finished in the editor.

• Database rules that are missing or too loose, so saving fails or data is open.

• A layout nobody tried on a small screen.

You can't see it yourself

You're signed in, your customers aren't. You know where everything is and your account already works. A stranger does not, and does not report what went wrong.

Find out in 2 minutes which of these is breaking your app. Test my app, free

How we checked, and what this is not

• When: the checks ran between 27 September and 19 October 2026 (UTC). The numbers were computed on 20 October 2026 from the saved results.

• Who is included: only apps whose link was pasted into VibeFix's own free check by a visitor, one result per app (the latest). We treat pasting the link as the owner asking for the check; we do not verify who owns each app. Nothing else is counted. Our own apps and test apps, and links that were not an app, are left out.

• What "launched app" means: a live, published web app that anyone can open at its public address.

• What the AI agent did: it opened each app on a phone and a computer as a stranger would, signed up, tried the main thing the app does, and went to checkout where there was one. It checks an app and never changes it.

• What counts as a problem: something a first customer would hit on the live app. For each one the agent wrote a one-line headline, and we sorted the 90 headlines into the categories above. A problem is counted as open data only when its headline says anyone can read it without signing in. One app counts once per category.

• Limits: this is not a random sample. People who paste a link may already suspect something is wrong, so the share with a problem could be higher than for apps in general. An agent can also miss problems, so these are minimums. We name no apps and no owners.

Related: My builder says the app is tested, but my customers hit problems · It works in the preview but breaks after I deploy

VibeFix is independent.