Your AI Built Your Landing Page Around a Theory Debunked in 1976

You shipped the landing page in a weekend. The AI laid out the sections, suggested the headline, kept the copy tight. It looked good. It still does. And your conversion rate has not moved since launch.

That is not a design problem. It is a knowledge problem. The tools that built your page were trained on patterns. They were not trained on 128 years of controlled experiments about why people decide to buy, when they abandon, and what a single extra form field costs you in revenue. That research exists. It is specific, it is quantified, and almost none of it is in the training data your build tool used.

The framework your AI defaulted to was described in 1898 by an ad executive writing copy for the National Cash Register Company. He never called it a funnel. He never drew a diagram. He was explaining how to write a print ad. One hundred and twenty-eight years later, every AI code generator on the planet uses a version of his idea to lay out your landing page. The research that followed it includes a Nobel Prize-winning theory about loss aversion, an experiment that proved too many choices destroy conversion, a $60 million A/B test that helped elect a president, and a McKinsey study that proved the entire funnel model is empirically wrong. Researchers proved it wrong in 1976 and kept using it anyway. So does your AI.

We automated building websites. We did not automate understanding why visitors convert. One of those problems is solved. The other takes exactly as long as it always did.

Not the person you think. And not for the reason you think.

Lewis described what became the AIDA model (Attention, Interest, Desire, Action) across several columns in The Inland Printer between 1898 and 1903. His original version was actually Attention, Interest, Conviction. The "Desire" and "Action" stages came later. He was writing about how to compose a single print advertisement. Not a sales pipeline. Not a multi-touch journey. A print ad.

The funnel shape came 26 years later. In 1924, William W. Townsend published Bond Salesmanship and mapped AIDA to a narrowing diagram where prospects dropped off at each stage. The first visual conversion funnel was drawn by a bond salesman, not a marketer. E.K. Strong gave it academic backing in 1925 with The Psychology of Selling and Advertising at Princeton. By the 1930s, the funnel was a standard sales management tool for forecasting revenue. It was never designed for websites. It was designed for door-to-door salesmen and Wall Street traders.

Fifty years ago. The funnel survived anyway.

In 1976, Michael Ray proposed three orderings of the hierarchy-of-effects. The standard "think-feel-do" sequence was just one option. He documented a "do-feel-think" sequence for low-involvement purchases: people buy first, then form opinions to justify the purchase. Leon Festinger's cognitive dissonance research from 1957 had already shown that people rationalize decisions after making them, not before. The linear funnel assumed the opposite.

Three years later, Daniel Kahneman and Amos Tversky published Prospect Theory. Their finding that reshaped conversion: people feel losses roughly twice as intensely as equivalent gains. "Stop losing $200/month to manual invoicing" is psychologically more powerful than "save $200/month." This explains why guarantees, free trials, and return policies work at the bottom of the funnel. They reduce perceived loss, not increase perceived gain. In 2009, McKinsey studied 20,000 consumers across five industries and found that options actually increase during evaluation, not decrease. The narrowing funnel was empirically wrong.

Three conversion breakthroughs. Each came from studying human psychology, not code patterns.

In 1999, Sheena Iyengar set up a jam tasting display at a grocery store. A table with 24 varieties attracted more visitors. A table with 6 varieties sold ten times more jam. The conversion rate: 3% with 24 choices versus 30% with 6. Choice overload became one of the most cited findings in conversion optimization. Fewer options, higher conversion. Most vibe-coded pricing pages do the opposite.

Two years earlier, Amazon filed US Patent 5,960,411: one-click ordering. It eliminated an entire funnel step. The patent was so controversial that the Free Software Foundation organized a boycott and Barnes & Noble was sued for its "Express Lane" feature. One button, billions in incremental revenue. It remains the most impactful conversion optimization in history.

In 2008, Dan Siroker ran what is commonly called an A/B test on the Obama campaign's donation page. It was actually a multivariate test: 24 combinations of 6 media variants and 4 button text variants. The winning combination (a black-and-white family photo with "Learn More" instead of "Sign Up") increased email signups by 40% and was attributed to $60 million in additional donations. The test's success led Siroker to found Optimizely, which brought A/B testing to the masses.

It democratized funnel analysis. It also taught an entire generation to trust a broken visualization.

In 2005, Google acquired Urchin Software (originally sold at $895/license) and released Google Analytics for free. This was a seismic event. Any website owner could now set up goal funnels, see where visitors dropped off, and measure conversion rates. It created an entire generation of marketers who thought in terms of funnels. The problem: GA's funnel visualization was architecturally broken for years. If a user entered at step 3 (skipping steps 1 and 2), GA backfilled them into the earlier steps. The report showed more people at the top than actually visited those pages. Marketers made optimization decisions based on phantom data.

A better diagnostic framework already existed. In 2012, BJ Fogg at Stanford's Persuasive Technology Lab proposed B=MAP: Behavior requires Motivation, Ability, and a Prompt to converge simultaneously. If motivation is high but ability is low (the checkout form has 14 fields), no conversion. If ability is high but there is no prompt (no clear call-to-action), no conversion. Most vibe-coded pages are missing at least one of these three elements. The model is 14 years old. Most pages still use the funnel layout.

More than you think. And almost none of it was trained on conversion psychology.

On February 2, 2025, Andrej Karpathy posted a tweet that was viewed 4.5 million times. He described "a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." Collins Dictionary later named it a 2025 Word of the Year. The adoption that followed was staggering.

92% of US developers now use AI coding tools daily. 41% of all code globally is AI-generated. 25% of Y Combinator's Winter 2025 batch had codebases that were 95% or more AI-generated. And 63% of vibe coding users are not developers at all. They are founders, marketers, and designers building landing pages, SaaS products, and marketing sites by describing what they want in natural language. The speed is real. The scale is real. The problem is what the AI optimizes for: the most statistically common page, not the most psychologically effective one.

It depends entirely on who is directing it.

Crazy Egg tested 11 AI landing page analyzers and scored the best one at 5 out of 15 on quality assessment. Most scored 2 to 4. When they A/B tested the highest-scoring tool's recommendations, the result was a 16.4% decrease in conversion rate. The control converted at 34.36%. The AI-recommended variant converted at 28.74%. The analyzers suffered from what the researchers called "context blindness": they could not understand page purpose, audience, or intent. Only one tool even asked about the target audience.

But the same study found something else. When a CRO expert used AI to generate a detailed, conversion-optimized prompt specifying page structure, copy angles, and psychological triggers, the resulting page achieved a 44.83% relative lift compared to the human-designed version. The tool was the same. The knowledge directing it was different. A METR study added another layer: developers estimated they were 20% faster with AI tools while actually working 19% slower. A 40% perception gap. You think the page is done. The conversion layer is missing.

Because AI does not learn taste or visual hierarchy. It learns what repeats most often.

Designer Michal Malewicz described vibe-coded output as "functional, but soulless," comparing it to Ikea furniture: "will you remember any of them 30 minutes from now?" His observation: generic messaging like "seamless streamlined effortless automation" tells nobody what you sell. An arXiv paper (2603.13036) formally documented this convergence, finding that AI code generators produce "a globally average aesthetic" regardless of product, industry, or cultural context. The AI generates the most statistically probable page. The average landing page converts at 2.35%.

128 years of conversion psychology says differentiation drives decision-making. Townsend's original 1924 funnel was built on the premise that each stage must give the prospect a reason to advance. When every competitor's page has the same hero layout, the same feature cards, the same generic CTA, visitors cannot tell products apart. The funnel collapses not because of friction, but because of sameness. There is nothing to advance toward.

It would apply six principles that took 128 years to establish. Vibe coding applies zero of them by default.

A page informed by the research in this article would use loss framing because Kahneman proved losses hit twice as hard as gains. It would limit choices because Iyengar proved that 6 options outsell 24 by ten to one. It would reduce steps because Amazon proved that removing one click generated billions. It would test assumptions because Obama's campaign proved that intuition about what works is often wrong. It would ensure every section provides Motivation, Ability, and a Prompt because Fogg proved conversion requires all three simultaneously. And it would acknowledge that visitors do not move linearly because McKinsey proved the funnel bends back on itself.

The fix is not "stop vibe coding." The fix is adding a diagnostic step that applies the knowledge the AI was never trained on. Understanding why visitors leave requires testing the page against the psychology that 128 years of research produced. You do not need to read the papers. You need a tool that has already absorbed them.