I asked an AI app builder for a "simple" landing page. Here's what it actually built.

MODEL Editors — Sat, 17 Oct 2026 21:30:08 +0000

We run MODEL, a small weekly magazine that tests new AI tools. Our rule is simple: if we didn't test it ourselves, we label it "not tested." Here's one real test, with the good parts and the problems.

The test

Tool: Bolt.new, free plan, $0 spent, tested Oct 15, 2026.

Prompt, word for word:

a simple landing page for a weekly AI tools newsletter with an email signup box

What happened

Speed: about 2.5 to 3 minutes from hitting enter to a finished first version.
Did it work? Yes. We typed a test email into the signup box and got "Subscribed! You're on the list!"
What it built: a lot more than "simple." It made a hero section, stats bar, features grid, sample issue, FAQ, a second call to action and a footer. It also set up a database table for subscribers (Bolt Database, built on Supabase) and pulled in a stock photo.
Errors: none.

The catch

It made up social proof. The page says "12,000+ readers." That number came from nowhere. If you ship it as-is, you're lying to your visitors.
It promised something it didn't build. The copy promises a welcome email, but nothing sends one.
"Simple" wasn't simple. The extra sections and the database use up free-plan tokens, and the free plan doesn't show you an exact token count.

Our verdict

Great for getting a working, good-looking signup page in a few minutes. But read every line of the copy before you publish, because AI builders fill empty space with confident fake numbers.

Tip: add "don't invent stats, testimonials or features that aren't built" to your prompt.

What we couldn't test

We also tried Synthesia's free plan, but signup stopped at email verification, so we didn't make a video. We're calling it not tested rather than guessing.

We write up tests like this every week. Here's how we test, and what we label when we can't: https://readmodelmag.com/how-we-test