For thirty years, online retail looked the same: a search box, a grid of results, a page of specifications. That era is ending. AI is not adding a feature to the store, it is rebuilding the store around the shopper, and every part of the journey is shifting at once.
Consider what is changing. Search is moving from a list of ranked links to a single synthesized answer. The shelf space is moving outside of retail, where answer engines, not retailers, decide what gets seen, and new advertising formats are forming around them, from paid LLM-embedded ads to assistant-native placement. The economics of content are collapsing as its marginal cost falls toward zero. Shoppers are learning to delegate rather than navigate, expecting the AI to decide and to answer rather than help them browse. And underneath it all, brands risk losing their narrative to an intermediary that now owns the customer touchpoint.
Having spent building fine-tuned AI models for online retail, I keep watching e-commerce executives respond to all of this by making the same three mistakes. Each one is avoidable, and each one is expensive.
Mistake 1: GEO And AI Slop
The webpage is being unbundled. For decades your homepage was the front door, and brands worshipped Google to be found. Now large language models decide what gets recommended, and a new discipline, generative engine optimization, is replacing search engine optimization. I have argued that Amazon’s homepage is dying and that retail is being unbundled the way media was a decade ago.
Here is where executives go wrong. They assume that because more pages once won at SEO, more AI-generated content will win at GEO. The opposite is true. AI trained on AI degrades, and model providers actively suppress pages that read like machine-generated filler. Pumping out slop is the fastest way to become invisible to the very engines you are chasing.
The fix is to give models what they actually reward: novel, unique and authentic content. That is exactly why large language models lean so heavily on sources like Reddit and LinkedIn when they train. Then measure how you show up. I recently interviewed Alex Dees, the founder of Meridian, on this very topic, and I built my own monitoring engine, QueryEdge. The payoff is real. In my research, traffic that arrives from LLMs converts up to nine times better than ordinary channels.
Mistake 2: Chatbot Without Needs
The interface is becoming a conversation. Shoppers already use ChatGPT and Claude to decide what to buy, and I have shown how I bought coffee without ever touching a browser. So brands rush to add a chat box. They try to mimic ChatGPT, and they fail. We saw the same pattern in the early days of Google, when every site tried to build search as good as Google’s and could not.
I am guilty of this mistake myself. In 2024 I launched a consultative bot for Decent that nobody asked for. It was fun users did not really used it. Amazon later shipped Rufus, in my view a clunky bolt-on that is often more wrong than right. A chat window you have to summon is friction, not service.
The fix is to weave conversation into the experience instead of pasting a clone on top. The clearest example is the product page. A product page no longer has to be average, stuffed with every keyword to please search while burying the reader in detail they never asked for. It can personalize to each shopper, based on their clicks and queries, surfacing what matters and hiding what does not. Conversation still belongs in customer care and in product questions. Based on my own A/B tests, this approach delivered 8.6 times better conversion, because the experience adapts to the shopper rather than forcing the shopper to adapt to it.
Mistake 3: Manual Brand Workflows
Modern search is multimodal and conversational. You can search with a picture, ask in full sentences, and rely on embeddings to handle misspellings and intent, as I demonstrate here how “good search & discovery” should look.
Brands however follow their old systems to control what is shown when. Brand teams overwrite AI results by hand, pinning products and forcing rankings the algorithm did not choose. Fighting a modern AI system this way usually lowers revenue, not raises it.
The fix is to let transformers and longitudinal behavioral data predict the next best product. As surely as we can finish “life is like a box of,” and you already thought chocolates, we can predict what a shopper most wants after a given query. Stop overriding the ranking and start predicting it. The result is higher conversion and a lower return rate, because shoppers end up with the product they actually wanted. Fewer returns is not a soft metric. It converts straight into savings.
The Future Of E-Commerce
Notice the common thread. Each fix runs on the same fuel: your own data. Every conversation and every click is, by definition, unique and authentic, the one signal no general model has. With it you can power discovery, personalize the page, and earn your place in AI answers. On-page and off-page stop being separate projects and become one system.
I argued back in 2023 that AI models do not create moats. Models commoditize weekly. What cannot be copied is your record of real intent and real purchases. That is why the future belongs to branded AI models, trained on a brand’s own behavior. The future of e-commerce is based on shopping data and fine-tuned AI models.
