AI for landing pages: built faster, tested better

Traffic first, then the page

Most landing pages don’t fail on design; they fail because nobody visits them. AI makes building so cheap that this trap gets bigger: in an hour you have a beautiful page nobody ever sees.

Before you build, answer three questions: where do visitors come from (ads, newsletter, search, social)? What did they just see before clicking? And what’s the one next step on the page?

Copy with AI: what works and what doesn’t

The biggest mistake with AI copy on landing pages is invented proof: “over 1,000 happy customers”, a quote from a customer who doesn’t exist, a study with no source. That’s not just embarrassing; under consumer protection law it’s misleading. Give the AI only facts you can back up and have it mark everything else as [check].

A structure that works

1. Headline: the outcome or the problem in the audience’s words.
2. Subhead: for whom, what, how fast.
3. One action: one button, in the first screen, on mobile too.
4. Benefits, not features: three to five points.
5. Proof: real numbers, logos with permission, real voices.
6. Objections: FAQ, price, process, privacy.
7. Repeat the action at the end.

Building: page builders, AI generators, code

Sources: durable.com, unbounce.com, framer.com. Prices exclude tax; check features and limits before you buy.

“Try” links are affiliate links (advertising): I earn a commission at no extra cost to you.

Personalisation: useful, not creepy

AI makes it easy to tailor landing pages by source: one variant per ad group, per sector, per search term. That works because the page continues what the visitor has just seen.

• By source: a headline that matches the ad or email, driven by UTM parameters.
• By sector: examples and proof from the visitor’s industry.
• Not by person: “Hi Anna from Munich” on an open page feels creepy and needs data you shouldn’t have for that.
• Mind consent: in the EU, personalisation via tracking cookies or device access that isn’t strictly necessary needs consent first.

My rule of thumb: trust is the most important asset. Personalisation should show you understand the situation, not that you know a lot about the person.

A/B testing with AI

AI helps on both sides: it writes variants, and it analyses results when you give it the numbers. Watch the volume: with a few dozen visitors per variant, a difference is usually chance. With little traffic, it’s worth more to settle big questions (offer, price, audience) than to test button colours.

Page speed: the underrated lever

Measured at the 75th percentile of page loads, split by mobile and desktop. Source: web.dev, Core Web Vitals.

• Images: modern formats (WebP, AVIF), the right sizes, don’t lazy-load the hero image.
• Scripts: every tracking and chat script costs time; load only what you need, and after consent.
• Fonts: self-host, few weights.
• Mobile first: most visitors from social media arrive on a phone.

What AI still gets wrong on landing pages

Imprint and GDPR basics for forms

• Imprint (Germany, § 5 DDG): name and address, legal form and representatives for companies, email and a fast contact route, register and number, VAT ID if you have one. Easy to find and directly reachable, so linked from every landing page aimed at Germany.
• Privacy notice at the form (Art. 13 GDPR): who processes, for what, on what basis, for how long, who receives the data. A link to the privacy policy right at the form.
• Newsletters and marketing: in Germany, marketing email needs prior express consent (§ 7 UWG). The usual proof is double opt-in: the person confirms via a link in an email.
• Be upfront: if you offer a guide in exchange for an email address, say clearly that emails will follow and what they’ll contain.
• Tracking: load pixels and analytics that aren’t strictly necessary only after consent (EU ePrivacy rules, in Germany § 25 TDDDG).
• Processors: sign an Art. 28 GDPR agreement with form, newsletter and builder vendors.
• AI images: under the EU AI Act, since August 2, 2026, disclose realistic AI images of people that could pass as real.

Checklist before go-live

• Traffic source defined, headline matches the ad or email
• One main action, visible in the first mobile screen
• Every number and quote sourced
• Imprint and privacy policy linked, notice at the form
• Newsletter with double opt-in and clear consent wording
• Tracking only after consent
• LCP under 2.5 seconds on mobile, no jumping content
• Checked at 390 and 1280 pixels wide, no sideways scrolling
• A test sign-up run through end to end

Metrics

Example: a landing page for a guide

Here’s the structure of a simple lead magnet page, say for a free guide in exchange for an email address. You draft the copy with AI; you supply the facts.

The consent wording is the most important sentence on the page: it says what the address will be used for, such as the guide and follow-up emails with tips and product recommendations, and that people can unsubscribe at any time. AI can draft it; you check it matches your actual emails.

After the click: confirmation and thank-you page

Many landing pages don’t lose leads on the page but afterwards. With double opt-in, the person has to open a confirmation email and click. Every step in between costs sign-ups.

• Confirmation page: right after submitting, say clearly what happens next: “Check your inbox and click the link”, with a note about the spam folder.
• Confirmation email: short subject line, one button, a sender name the person recognises.
• Thank-you page: delivers what was promised (download, access) and offers a next step.
• First email: soon, with the promised content, not a sales pitch.
• Test it: run the whole flow with a real address, including mailbox providers other than Gmail.

AI helps here with writing the pages and emails and with analysing where people drop off. Your newsletter tool supplies the numbers: sign-ups, confirmations, opens.

Images on landing pages

The hero image decides in seconds whether a page looks trustworthy. AI images are quick to make, but not automatically the right choice.

• No fake customers: an AI face next to a quote reads as a testimonial and is misleading.
• Labelling: under the EU AI Act, since August 2, 2026, disclose realistic AI images of people that could pass as real.
• Load time: the hero is often the largest element and drives LCP. Compress, use modern formats and the right sizes.
• Alt text: describes what’s shown, for screen readers and search engines.

More on prompts, rights and brand style in AI images for marketing.

12 prompts for landing pages

My offer: [description]. Audience: [audience]. State the offer in one sentence: for whom, what outcome, in what time, with no promises I can’t keep.

Write 10 headlines for this landing page [offer], two each in the patterns problem, outcome, question, number (only with a sourced number: [number]), contrarian. Max 60 characters.

This ad leads to the page: [ad copy]. Write the headline and subhead so the page picks up seamlessly where the ad left off.

Turn these features [list] into five benefits for [audience], max 12 words each. No superlatives.

What ten objections does [audience] have to [offer]? Write an honest FAQ answer to each, max 50 words. Where you need facts, write [check].

Outline a landing page for [offer] with sections, the purpose of each and copy length. One main action, mobile first.

Write the copy around a sign-up form for [guide/newsletter]: field labels, button, consent wording that says which emails will follow, and a note on the privacy policy and double opt-in.

For this test, write one headline variant that tests a clear hypothesis: [hypothesis]. Everything else stays the same. State the hypothesis in one sentence.

Variant A: [visitors, sign-ups], variant B: [visitors, sign-ups]. Is the difference reliable? Explain in plain words how confident the result is and what to test next.

Find every cliché, superlative and unsupported claim in this page copy: [text]. Suggest concrete, provable wording.

Check this landing page for missing required elements: imprint link, privacy notice at the form, consent wording, tracking only after consent, labelling of ads and AI images: [text/URL]. A first pass, not legal advice.

Here’s my page’s PageSpeed report: [report]. Rank the suggestions by impact on mobile LCP and effort, and explain the top three in plain words.

The most common mistakes

1. Building without a traffic plan. A beautiful page without visitors does nothing.
2. Invented proof. AI likes to invent numbers and quotes; that’s misleading.
3. Several goals on one page. Offer everything and you get nothing.
4. Forgetting required elements. Imprint, privacy notice and consent belong on every page with a form.
5. Tracking before consent. Load pixels only after agreement.
6. Checking only on desktop. The majority often arrives on a phone.