5 landing page mistakes AI makes (and how to prompt around them)

You gave an AI builder a simple prompt: "Build me a landing page for my SaaS." A minute later, you had a page. It looked fine — clean sections, gradient hero, feature grid, a couple of testimonials, a big blue button that said "Get Started."

But something was off. Nobody was clicking. Nobody was signing up. And when you asked a friend to look at it, they said the words "professional but I have no idea what it does."

You're not alone. AI builders — Devaiy, Lovable, v0, Bolt, Cursor with a UI plugin — all generate landing pages, and they all tend to make the same five mistakes. Not because the underlying models are bad, but because "build me a landing page" is one of the most ambiguous prompts in existence. The model does its best with the information it has, and its best is generic.

The good news: each mistake has a specific fix, and once you know what to look for, you can fix them in-flight with a better prompt.

Mistake 1: A vague headline that doesn't say what the product does

Every AI-built landing page starts with a bold headline. Most of them read like this:

Transform your workflow with cutting-edge technology

Or:

The future of productivity is here

Or the classic:

Build something amazing

None of these tell a visitor what the product does, who it's for, or why they should stay on the page for another five seconds. They're headline-shaped, but they're empty.

Why AI defaults to this: vague headlines are safe. The model doesn't know if you sell invoicing software or a dog-walking marketplace, so it picks language that could apply to either. Grand and generic beats specific and wrong when the model is guessing.

What good looks like:

Send invoices in 30 seconds. Get paid twice as fast.

That headline names the outcome (get paid) and the mechanism (send invoices in 30 seconds). A visitor knows in one line what the product is and why they might care.

The prompt fix: don't just say "landing page for my SaaS." Include the specific outcome and the specific mechanism. Try:

"Write a landing page for an invoicing SaaS aimed at freelance designers. The headline should promise 'get paid faster' as the outcome and reference the 30-second invoicing flow as the mechanism."

The more specific the setup, the sharper the copy.

Mistake 2: A feature list instead of benefit-driven copy

The second thing every AI builds is a feature grid. Three or four columns, one icon per column, one line of copy each:

• ✓ Cloud-based
• ✓ 24/7 support
• ✓ Fully scalable
• ✓ Enterprise-ready

None of that tells anyone what they'll actually get. "Cloud-based" is a description of infrastructure, not a promise. "24/7 support" is table stakes. "Scalable" is a word engineers use when they don't want to say a specific number.

Why AI defaults to this: bullet lists are structured, symmetrical, and easy to generate. The model has seen thousands of SaaS pages that look exactly like this, so it produces one that fits the pattern.

What good looks like — same features, benefit-framed:

• ✓ Works anywhere with a browser (no install, no VPN)
• ✓ We answer support tickets in under 2 hours, every day
• ✓ 10,000 users? 1 million? Same price, same performance
• ✓ SOC 2 Type II — your legal team will nod once and move on

Each line answers "so what?" for the reader. Same features, but the reader can see themselves in the benefit.

The prompt fix:

"For every feature, tell me what problem it solves and for whom. Skip generic tech words like 'cloud-based' or 'scalable' — instead, describe what the user actually experiences."

Mistake 3: Generic CTAs — "Get Started," "Learn More," "Sign Up"

Every AI-built page ends its hero with the same button: Get Started or Learn More or Sign Up. Sometimes all three, stacked on top of each other, each in a slightly different color.

The problem isn't that these words are wrong. It's that they don't tell the reader what happens if they click. "Get started" doing what? "Sign up" and then what?

Why AI defaults to this: these are the most common CTA phrases across the entire training data. They're the safe, statistically-average choice.

What good looks like: CTAs that describe the next 30 seconds.

• Create my first invoice — free (instead of "Get Started")
• See how it works in 90 seconds (instead of "Learn More")
• Try it with your own data (instead of "Sign Up")

Each one tells the reader exactly what they'll experience if they click. That specificity removes friction — the reader isn't wondering "what am I committing to?" — they can see the outcome before they act.

The prompt fix:

"Every CTA should describe what happens in the next 30 seconds after the user clicks. Not 'Sign Up' — 'Create your first invoice, free, no credit card.' Rewrite every button on the page this way."

Mistake 4: An abstract hero image or generic gradient

The hero section usually gets an image. In an AI-built page, that image is one of three things: a purple-to-blue gradient with no visual anchor, a stock photo of a happy person in front of a laptop, or an "AI-generated" abstract shape that means nothing.

Why AI defaults to this: picking a real image requires knowing what the product actually looks like. The model doesn't have your product screenshots, so it defaults to visually safe abstractions.

What good looks like: a screenshot of your product doing the thing your headline promises. If the headline says "send invoices in 30 seconds," the hero image should be a screenshot of the invoice form. Not a hero shot of a laptop from Unsplash. The actual product.

If the product doesn't exist yet or doesn't have a UI worth showing, use a real photograph of a real customer using it — or use a customer quote in place of the image. Anything specific beats anything abstract.

The prompt fix:

"Use a product screenshot as the hero image, not a gradient or an abstract graphic. If I haven't given you a screenshot, leave the hero section with a placeholder that says 'insert product screenshot here' — don't fill it with generic imagery."

Mistake 5: Placeholder social proof that reads like it was invented

Every AI-generated landing page has a testimonials section. Three quotes. Three portraits. Three job titles. Something like:

"This product transformed our business." — Sarah K., CEO of Acme Corp

Nobody talks like that. And Sarah K. of Acme Corp doesn't exist. Every visitor who has ever read a landing page can smell the fabrication instantly.

Why AI defaults to this: the model fills gaps with plausible-looking placeholder content. It doesn't know your real customers, so it invents testimonial-shaped words to keep the layout balanced.

What good looks like: either real testimonials with real names, roles, and links to real profiles — or nothing at all. An empty testimonials section is more credible than a fake one.

If you can't yet get real quotes, use real numbers instead:

• 12,000 invoices sent this month
• Trusted by 340 freelancers in 22 countries
• Featured in Indie Hackers, ProductHunt (#3 of the day), Hacker News

Numbers are harder to fake and easier to verify. When you have real testimonials, replace the numbers with quotes. Not the other way around.

The prompt fix:

"If I haven't given you real customer testimonials, leave the social proof section empty for me to fill in. Do not invent quotes, names, companies, or job titles. Placeholder testimonials look worse than no testimonials."

This applies even on platforms like Devaiy

AI-builder platforms make generating a landing page fast, but they don't magically make it convert. Devaiy, Lovable, v0, and every AI builder relies on your prompt to know what to build. A vague prompt gets a vague landing page — regardless of how good the underlying model is.

The good news: platforms that support iterative conversations (like Devaiy) let you fix these mistakes in-flight. Don't accept the first version. Ask the AI to rewrite the headline with a specific outcome. Ask for benefit-driven copy instead of feature bullets. Ask for real numbers instead of placeholder testimonials. Push back.

The tool is only as good as the direction you give it. Even the best model can't read your mind about who your customer is or what makes your product different — you have to tell it. And once you know these five mistakes, telling it becomes half the job.

The 5-mistake landing page checklist

Before you launch or share any AI-built landing page, check it against this list:

1. Headline — does it name a specific outcome and a specific mechanism, or is it generic marketing prose?
2. Copy — is every feature framed as a benefit for a specific user, or is it a list of tech words?
3. CTAs — does every button describe what happens in the next 30 seconds, or is it "Get Started"?
4. Hero image — is it a real product screenshot, or a decorative gradient?
5. Social proof — is it real names and real numbers, or fabricated testimonials?

If you commit any of these mistakes, the fix is a better prompt — not a better model. Try rewriting one section with a more specific ask and compare the result. That's the whole game.