Some links are affiliate links — we may earn a commission at no cost to you. It never changes our picks. Disclosure

7 Mistakes That Cost Me Hours Building Sites with AI (So They Won't Cost You)

By Shawn HamiltonUpdated June 2026⏱ 12 min read

The short version

Across building several live sites and an app as a non-technical beginner, I made the same kinds of mistakes everyone does — and each one cost me real hours. Here are the seven biggest, exactly what went wrong, and the simple fix for each. Read this and skip the pain I didn't.

The seven mistakes

  1. Pointing at the wrong nameservers
  2. Pushing code I hadn't tested
  3. Trusting browser cache (seeing old versions)
  4. Believing the AI when it was confidently wrong
  5. Putting fake numbers on the site
  6. Exposing secrets that should stay hidden
  7. Panicking instead of waiting

I'm not a developer. Over a few weeks I built and launched several websites and a mobile app using AI tools as my guide. The building was the fun part. The mistakes were where I actually learned — and most of them cost me hours I'll never get back. So here they are, honestly, so they cost you minutes instead. Every one of these genuinely happened to me.

1 Pointing at the wrong nameservers

This is the one that nearly broke me. On one launch, I carefully entered every DNS record correctly — and the site still showed a "parking page" for hours. At one point it even displayed a message suggesting the domain had expired, which sent me into a small panic. I'd just paid for it.

The domain hadn't expired at all. The problem was a layer above the records: my domain's nameservers were pointing at the registrar's parking service instead of its real DNS service. So all my perfectly-entered records were sitting in a directory that nothing was consulting. They were being completely ignored.

The fix: If your DNS records are correct but the site won't appear, check your nameservers before touching anything else. They must point at the directory your records actually live in. I switched them to the proper DNS nameservers and the site appeared within minutes. (More on this in my domains & DNS guide.)

2 Pushing code I hadn't tested

On my word-game project, I got excited, made changes, and pushed them straight to the live site without checking them first. The deploy failed. So I tweaked and pushed again. It failed again. Four times in a row. Each failure cost time and a little more confidence, and the cause was always the same: I was publishing changes I hadn't actually run myself first. One of them referenced a file I'd forgotten to include.

The fix: Always test your changes before you publish them. For simple sites, open the file and click around. For anything with a build step, run that build locally and confirm it passes first. "I'm sure it's fine" is how you get four failed deploys in a row. A deploy should never fail twice for the same reason.

3 Trusting the browser cache (and seeing ghosts)

This one made me feel like I was losing my mind. I'd deploy a fix, the platform would confirm it was live, I'd reload the page — and the old behaviour would still be there. I'd assume my fix hadn't worked and start "fixing" something that was already fixed, making things worse.

The culprit was browser cache: my browser was showing me a saved older copy of the page instead of fetching the new one. The fix was live; I just couldn't see it.

The fix: After any deploy, do a "hard refresh" (Ctrl+Shift+R on Windows) or open the page in a private/incognito window. That forces your browser to fetch the genuinely current version. Before you conclude a fix didn't work, rule out cache first — it fooled me repeatedly.

4 Believing the AI when it was confidently wrong

I love AI tools — they made all of this possible. But I learned the hard way that they can be confidently, completely wrong. The worst saga was getting my word game to accept only real words. The AI's solutions sounded authoritative each time, yet one approach accepted obvious gibberish, another rejected perfectly real words, and it took round after round to get right. A separate tool once kept inserting the wrong year into dates because it was guessing rather than checking.

The pattern: the AI's confidence had no relationship to its correctness. A wrong answer was delivered just as assuredly as a right one.

The fix: Treat AI output as a strong first draft, never as gospel. Verify anything important by testing it yourself or checking a reliable source. When facts matter, use a tool that cites its sources (see Perplexity vs ChatGPT). The AI is a brilliant assistant, not an infallible authority — keep your own judgment switched on.

5 Putting fake numbers on the site

When my sites were brand new, the design templates came with impressive-looking placeholder stats — "thousands of users," big visitor counts. On more than one site I left them in at first, thinking they made me look established. That was a mistake on two fronts: it's dishonest, and ad networks and search engines actively dislike sites making claims that obviously aren't true. Visitors can sense it too.

The fix: Never publish numbers that aren't real. Strip out placeholder stats before you go live. Honest framing ("independently tested," "hand-picked") builds more trust than invented metrics — and keeps you out of trouble when you later apply for advertising programs that review your site.

6 Nearly exposing secrets that should stay hidden

Some of my projects needed secret keys — credentials that let my site talk to a paid service. The dangerous beginner instinct is to drop those keys straight into the website's code, where anyone viewing the page could find them. On the projects that needed them, I learned to keep keys on the server side only, hidden from the browser entirely, so visitors never see them.

The fix: Never put secret keys, passwords, or tokens into anything a visitor's browser can read. Keep them server-side, or in the secure "environment variables" your host provides. And keep secrets out of any code you share publicly — it's easy to leak one without realising. If you ever do expose one, treat it as compromised and replace it immediately.

7 Panicking instead of waiting

My most expensive mistake wasn't technical — it was emotional. Several times, something looked broken when it was actually just slow. A new domain not showing up (DNS still spreading worldwide). A change not appearing (cache). My instinct was to assume disaster and start frantically changing things — which sometimes broke what was quietly working fine.

The irony: one time my gut told me something was genuinely wrong (a setup taking far too long), and I was right — that was the nameserver issue. The skill is telling the two apart, and you only get there by slowing down.

The fix: When something looks broken, pause before you "fix" it. Ask: could this just be propagation or cache? Verify from the outside (a DNS-checker site, an incognito window, your phone on mobile data) before changing anything. Most "it's broken!" moments are really "it's not finished updating yet." Calm and methodical beats fast and frantic every single time.

The bottom line

None of these mistakes meant I'd failed — they're the actual curriculum of learning to build things. But every one cost me time you don't have to spend. If you remember just three: check your nameservers when DNS misbehaves, never publish untested changes, and don't trust the AI (or your own panic) without verifying. Do that, and you'll skip most of the hours I lost. Breaking things is part of building. Breaking them the same way I did is optional.

Frequently asked questions

Is it normal to make this many mistakes building a website?
Completely. Breaking and fixing things is how you learn — even experienced developers hit these. The goal isn't zero mistakes; it's not repeating the avoidable ones.
Can I really build a site with AI if I'm not technical?
Yes — I did, several times, with no background. The mistakes above are the main friction points, and now you know them in advance.
What's the most important lesson here?
Don't panic. Most "broken" moments are propagation or cache, not real failures. Slow down, verify from the outside, and change things only once you understand what's actually wrong.
Should I stop trusting AI tools after the wrong answers?
No — they're genuinely powerful and made all my projects possible. Just verify important outputs rather than taking them on faith. Confidence isn't correctness.
📘 In this series: How I Built 7 Web Projects with AI · Domains, DNS & Hosting Explained
SH

Shawn Hamilton

Shawn is a non-technical builder who taught himself to ship websites and apps using AI tools — launching several live projects, from a word game to a finance portal, in just a few weeks. He writes about what actually works (and what broke) for complete beginners.