AI and the future of ecom
AI agents are gradually replacing human buyers. CRO now splits into two distinct games.
According to BCG, 81% of US consumers have already used an LLM for new-product discovery. AI shopping agents like ChatGPT Shopping and Google Gemini handled close to 25% of high-consideration purchases in Q1 2026, pushing orders from this segment up 15× year-on-year.
Those numbers confirm: a meaningful chunk of selling pages will soon be “read” by AI, not by people.
That’s why I’m writing this.
This isn’t some far-off prediction either. In the past 6 months: OpenAI pushed ChatGPT into product discovery and merchant apps. Google launched a Universal Commerce Protocol for AI agents transacting across marketplaces. Amazon shipped tools letting their agents shop on other retailers’ sites.
A hard-to-predict future for CRO
On May 12, HBR published research from Jafar Sabbah (Bayes Business School) and Oguz A. Acar (King’s Business School). They ran thousands of simulated shopping rounds across 4 leading AI models × 4 popular product categories, testing the CRO tactics every DTC brand uses daily.
The conclusion: the more capable the reasoning AI model, the more skeptical it is of overt persuasion tricks.
Put simply: as AI gets smarter over the next 6-12 months, the traditional CRO playbook gets less effective, not more.
A lot of people still believe tactics like scarcity, urgency, and countdown timers are the core levers of a high-converting Landing Page. The tools I built were also founded on that belief.
Now research is showing that a meaningful chunk of the future “buyers” won’t react to any of those tactics.
From here on, position in an AI feed will matter the way ranking #1 on Google did in 2010.
Psychological tactics don’t work on AI
First, AI agents don’t have FOMO. Scarcity works on human buyers because it triggers the feeling of loss, which lights up the amygdala. AI agents don’t have an amygdala. No feeling of loss.
Second, and this is the deepest point I think: AI agents have wide context windows. For an agent comparing and pulling data from 20 landing pages simultaneously, a countdown timer on one page won’t create urgency. To the AI, that timer is just one variable in a 20-element dataset.
Finally, advanced reasoning models are explicitly trained to detect manipulation. “Only 3 left!” in copy usually gets flagged by the AI as a manipulation signal (dark pattern), not as added information.
The summary: AI agents aren’t persuaded. AI agents compare. The brand that wins this “fight” is the brand with the most complete data and the clearest comparison signals — not the brand with the most persuasive copy.
The entry point is shifting
Agent behavior is only part of the story. The deeper shift, the one DTC brands should worry about more, is how traffic finds the store.
The old funnel everyone knows:
Facebook ad → landing page → checkout
Three years from now, it gradually becomes:
Agent → discover & negotiate → dynamic commerce → checkout
Or:
Chat → generated storefront → generated offer → payment
When OpenAI, Google, TikTok, and Meta become the entry point for shopping, traffic stops flowing into homepages. PDP, navigation, and the “above the fold” of a landing page will get weaker.
What gets stronger? Generated offer, generated trust layer, generated comparison, conversion intelligence. Things rendered per visitor, per context, at runtime.
One storefront per visitor, per context. No more “one page for everyone.”
CRO splits into two games
CRO isn’t dead. CRO splits into two distinct games, requiring two different tool stacks and two different content mindsets.
Game 1 is still CRO for humans. Emotion, story, psychology, social proof, urgency, scarcity.
Most traffic today is still human, and a brand that abandons Game 1 = losing 80% of current revenue. Good news for skilled copywriters, designers, and CRO consultants: you still have work, and your work is now even more complex.
Game 2 is CRO for machines: reliable signals, structured data, schema markup for AI crawlers, authentic reviews at volume and depth, position in the AI feed, API response speed.
This game is entirely new. Good news: every brand is at the starting line together.
The brand that wins over the next 5 years won’t be the brand best at one game. It’ll be the brand playing both.
Within these two games, things like static themes, pure drag-and-drop builders, template marketplaces, landing page clones, manual CRO testing… are slowly “dying.”
Realtime generated storefronts (storefronts built on the fly), per-visitor/per-agent personalized funnels (funnels personalized for each visitor / agent), autonomous experimentation, memory-based commerce, intent-driven product discovery, AI-native merchandising… will live well and keep getting stronger.
Three things to act on now
The first is authentic reviews. Of every signal we tested, this is the only one that consistently moves AI agents in a positive direction.
Brands need a system to collect real reviews — high volume, with depth — not just 5 stars + “great product.” If 12 months from now your competitor has 2,000 reviews with depth and you have 200 shallow ones, the AI agent will commit to them before even considering you.
There’s no “persuasion” here. Just signal.
The second: schema markup matters like SEO did in 2010.
AI agents read structured data before they read marketing text. Product schema, FAQ schema, Review schema, and so on.
I see many DTC brands skipping schema because “we’re already in Google’s index” — that’s 2020s thinking. Today, schema markup is how you talk to AI crawlers. And AI crawlers are now a traffic source on par with Google.
Finally, A/B testing for AI agents is an entirely new stack.
Optimizely, VWO, and Convert were built to test human behavior: click, scroll, time on page. Nothing in that stack measures “does the AI agent pick this LP when comparing 20 others.”
We need new test infrastructure that simulates how an AI agent evaluates — and this is a huge tooling gap.
The three above (reviews, schema, A/B test) are entry-level. The long-term moat isn’t those three.
Every LLM is available. Every image generation model is available. AI page builders will commoditize too.
What doesn’t commoditize: your conversion data across years, the winning patterns you’ve accumulated from thousands of tests, the behavioral loops you’ve measured on real stores, and your team’s ecommerce intuition.
A DTC brand that starts recording winning patterns systematically today will have a real moat in three years. A brand only using AI to generate pages without accumulating data will still be at the starting line three years from now.
Wrapping up
The truth: before AI, page builders were genuinely hard for regular people to use. Even at GemPages, only around 5% of users could actually design a beautiful page.
AI changes that. With AI, anyone can design a landing page and optimize conversion. GemPages pioneered Image-to-Layout: upload a screenshot, AI turns it into an editable page. The launch video from two years ago has nearly 1 million views, and hundreds of thousands of sections have been generated by AI.
But AI is also a massive challenge. If we don’t keep evolving, GemPages gets replaced. That’s why we’re all-in on GemAI — CRO Brain: an AI layer that connects GemPages, GemX, and Gemians into a CRO brain that learns and self-optimizes for every store.
My note: if you’re building a Shopify app, ask yourself “if an AI agent can do my job, what’s left for me to exist for?” Build that future before it builds you.
— Chris