Building a Website With AI: The Good, Bad, Very Ugly

Here’s what actually happens when everyone thinks they can build a website with AI: most of them ship something that looks finished and isn’t.

The Good is real. AI builders drop the time from weeks to hours and the cost from thousands to a $10 to $40/month plan, which is genuinely great for validating an idea.

The Bad is that “generated” gets mistaken for “done”: templated sameness, a strategy gap, and a site you outgrow fast.

The Very Ugly, the part vendors won’t tell you and most agencies only hint at, is that AI-generated code ships quantifiable security flaws, AI builders expose small businesses to real ADA accessibility lawsuits, and AI-built sites are often structurally invisible to the very AI search tools people now use to find businesses.

Four Eyes has been building websites since 1998. We’ve watched good sites go downhill because someone fed them to an AI builder with zero knowledge of design, code, or strategy. We’ve also watched people with that knowledge use the same tools to do incredible things.

The difference isn’t the tool. It’s the hands on it.

The short answer: what happens when everyone builds with AI

AI lowers the cost of starting a website and raises the cost of the mistakes you can’t see.

• Good, Bad, Ugly in one frame: AI is a legitimate head start for speed and cost, a trap when you confuse output with a finished product, and a genuine liability when the invisible parts (security, accessibility, machine-readability) go unaudited.
• “Generated” is not “done”: A generated homepage is a first draft that renders in a browser, not a launched business asset that’s been reviewed, secured, and made findable.
• Tool amplifies the operator: The same AI builder produces a sharp, fast, accessible site in skilled hands and a broken, exposed one in unskilled hands. The tool doesn’t decide which. You do.
• Skill = multiplier, ignorance = debt: A pro uses AI to execute decisions in minutes instead of hours. A novice uses it to skip the decisions entirely, and every skipped decision becomes technical debt.

The gap between the skilled and the unskilled didn’t shrink when AI arrived. It got wider, and more expensive to close.

The Good: what AI website builders genuinely get right

For speed, cost, and validating an idea fast, AI builders are a legitimately good starting point.

• Speed: AI tools can generate full page layouts in seconds, so work that took days or weeks now takes hours. That compression is real and it matters when you’re racing to test an idea.
• Cost: Free tiers and $10 to $40/month plans put a web presence within reach of nearly any budget, which used to mean a several-thousand-dollar build.
• Rapid iteration: For an MVP or a single campaign landing page, you can spin up three variations before lunch and throw two away. That’s a healthy use of the tool.
• Democratized access: A solo founder in Charlotte with no budget and no developer can now stand up a functional page and start collecting emails the same day.
• Multiplier in skilled hands: A designer who already knows layout, hierarchy, and conversion uses AI to skip the tedious parts and ship the polished version faster. Same tool, wildly better output.

If your goal is to prove a concept cheaply, an AI builder plus a smart local plan for affordable local SEO can get you further, faster, than it should. The trouble starts the moment you mistake that head start for the finish line.

The Bad: why “looks done” isn’t done

The generation step is the beginning of the work, not the end. Skipping strategy doesn’t remove the cost. It defers it.

• Templated sameness: AI-built sites often end up structurally identical to others on the same platform. Two businesses in the same industry can land on near-identical layouts, which means your “custom” site looks like your competitor’s.
• The “looks done” illusion: A rendered page with hero, three feature boxes, and a footer feels complete. Feeling complete and being complete are different things, and only one of them makes you money.
• Outgrowing the platform: Growing businesses need more service pages, location pages, integrations, forms, and tracking. Some AI builders make that harder, not easier, and rebuilding becomes the better option sooner than you’d think.
• Strategy gap: AI executes decisions. It should not make them. It doesn’t know your margins, your best customer, or which service actually pays your bills, so it defaults to generic.
• Technical debt at scale: Using AI to bypass planning generates debt fast. AI should execute structural decisions faster, not replace the thinking behind them.

Here’s a concrete version. A bookkeeper searches “AI website builder,” ships a clean-looking page in an afternoon, and feels done. Six months later she wants a page for “affordable bookkeeping for freelancers,” a separate page for small-business payroll, a working intake form, and conversion tracking. The builder fights her on all four. Now she’s paying to rebuild the thing she thought was finished. If she’d planned the structure first, a WordPress build in Charlotte would have grown with her instead of boxing her in. When a site becomes the reason you can’t grow, redesigning without losing your SEO becomes the project, and that’s a bigger job than the one she avoided.

Practical rule: Decide your page structure, services, and conversion goals before you generate anything. AI is a faster executor, never a substitute for the plan.

The same logic applies if you’re selling online. People underestimate the moving parts, and the real cost of an ecommerce website is where the “AI did it for $12” story falls apart hardest.

The Very Ugly #1: AI-generated code ships security flaws

A site that runs isn’t a site that’s safe, and AI won’t tell you what it left exposed. This is the downside no vendor puts on the pricing page.

• Generated code carries vulnerabilities: AI writes plausible code, not audited code. Studies of AI-generated output have repeatedly found a measurable share ships with real security flaws baked in.
• No one reviews what they can’t read: If you can’t read code, you can’t spot an unsanitized input or a leaky form handler. AI won’t flag it, and neither will you.
• Small sites are targets too: Automated bots don’t care that you’re a two-person shop. They scan for known weaknesses at scale, and a generic AI-built form is a soft target.
• “Works” is not “secure”: A page that loads and submits looks fine. Whether it exposes customer data or lets someone inject a script is invisible from the front end.

Independent research has found that a meaningful percentage of AI-generated code contains security vulnerabilities, and performance shortcuts often ride along with them. Fast and broken is still broken. AI website builders handle basic SEO reasonably but fall short on deeper technical control, and generated output often ships with generic, unaudited structures underneath the pretty surface.

The Very Ugly #2: accessibility lawsuits your AI builder won’t warn you about

An AI-built site can be a legal liability out of the box, and the builder’s terms of service won’t cover you when the demand letter arrives.

• ADA and WCAG apply to you: Courts have repeatedly treated business websites as places of public accommodation under the ADA, and WCAG is the standard used to judge compliance. “We’re small” is not a defense.
• AI output fails accessibility checks: Generated pages routinely ship missing alt text, poor color contrast, unlabeled form fields, and broken heading order. Every one of those is a documented barrier.
• Demand letters are rising: Website accessibility lawsuits and pre-suit demand letters have climbed for years, and they disproportionately hit small businesses that assumed they were too small to notice.
• Generic headings = defect + invisibility: AI’s habit of stacking generic heading structures both hurts AI-search readability and signals underlying accessibility defects. One flaw, two problems.
• The owner is liable, not the tool: When the letter comes, it names your business. The AI builder isn’t a party to the lawsuit. You are.

This is the same tradeoff we lay out in our take on the ethical use of ChatGPT. The tool is powerful and the responsibility for what it produces stays with the human who shipped it. Nobody at the AI company reviewed your contrast ratios.

Practical rule: Before any AI-built site goes live, run it through an accessibility checker and fix every flagged error. Accessibility is a legal requirement, not a nice-to-have.

The Very Ugly #3: you used AI to build a site AI search can’t read

You used AI to build a site, and AI search can’t read it. So it recommends your competitor instead.

• AI-built, AI-invisible: The same generic output that renders fine in a browser is exactly what an LLM struggles to parse into a confident answer.
• Generic headings hurt machine readability: Most AI builders manage basic SEO but produce flat, generic heading structures that make it hard for AI and LLM search to understand what your page is actually about.
• AI search needs structure, not just pixels: ChatGPT, Perplexity, and Google’s AI answers don’t “see” your design. They read your HTML, your headings, and your schema. Pretty means nothing to them.
• Missing schema and thin semantic HTML: No structured data, vague heading hierarchy, and generic markup give an LLM nothing clean to extract and cite.
• Unreadable is worse than blocked: Deciding whether to block AI crawlers is a strategy choice. Being unreadable to them by accident is just self-sabotage.

Concrete scenario. A customer opens ChatGPT and asks for “a reliable local roofer near Charlotte.” The model answers by pulling from sites it can read cleanly. If your AI-built page is a wall of generic H1s, no schema, and vague headings, you’re not in the answer. Your competitor with structured service pages and a real heading hierarchy is. That’s not a ranking penalty. That’s structural invisibility, and it’s the direct result of the tool you used to save time. Getting found now means the fundamentals of improving your local SEO rankings plus real technical structure, which is where technical SEO and AEO work earns its keep.

So who should actually use AI to build a site, and how

Use AI as a fast executor once you’ve made the strategic, security, and accessibility decisions a person still has to make. Here’s the honest breakdown of when it’s fine and when it’s a mistake.

When AI-only is fine, risky, or a mistake for building a website

• Fine for: An MVP, a startup site to test demand, or a single campaign landing page you’ll retire in a month. Low stakes, short lifespan, easy to throw away.
• Risky for: Revenue-critical sites, regulated industries (law, medical, finance), and growth-stage businesses that will need real integrations and structure. High stakes, long lifespan.
• Execute, don’t decide: Let AI draft copy, generate layout variations, and speed up the grunt work. Keep strategy, page architecture, and conversion logic in human hands.
• Audit before you trust: Every AI-built site needs three checks before launch: security review, accessibility scan, and structure and schema review. No exceptions.
• Skilled operator, incredible results: The people getting real value from AI builders already know what good looks like. They use AI to get there faster, then they audit and fix what it missed.

If you’re a contractor or builder, the stakes-versus-speed math is stark, because a bid request lost to a broken form is real money. Web design for Charlotte contractors and construction company web design both live in the “risky for AI-only” zone for exactly that reason. Selling online lands there too, which is why Charlotte ecommerce design needs a human on the architecture. Even your logo and brand fundamentals deserve intent, not a generated default.

Practical rule: AI is a green light for throwaway and test pages, a yellow light for anything you’ll keep, and a red light for anything that carries revenue, regulation, or your reputation, unless a skilled human audits it first.

The takeaway on building a website with AI

The payoff: AI didn’t make web expertise optional. It made the gap between the skilled and the unskilled wider and more expensive to close. “Generated” was never “done,” and the bill for that difference shows up later as security holes, accessibility demand letters, and invisibility to the AI search your customers now use.

So here’s the question. If your AI-built site got hit with an accessibility demand letter, or a customer asked ChatGPT for a business like yours and it named a competitor instead, would you know why?

Next step, do it this week: run your current site through a free accessibility checker (WAVE) and a page-speed audit, and write down every error each one flags. That list is your real to-do, the part AI told you was already finished. If the list scares you, that’s the point. We’ve been cleaning up messes like these in Charlotte since 1998. When you want a human read on it, our Charlotte web design team and a second opinion audit are here. Start with a quick note through our contact page.

Sources

• A measurable share of AI-generated code ships with real security vulnerabilities according to independent research.
• WCAG is the accepted technical standard used to judge website accessibility compliance under the ADA.