Does This ChatGPT Thing Even Work For What I Sell?
For about a year now, “GEO” keeps coming up in marketing chats. GEO (generative engine optimization) is basically SEO for the answer ChatGPT or Perplexity gives, instead of the links Google shows. The thing you’re optimizing is whether the model recommends you at all.
And every time it comes up, someone asks the same question: does this actually work for my category?
I get why. It’s a real budget call — pour resources into GEO, or keep them on search and paid social. Guess wrong and it’s an expensive miss.
So here’s the short version, and then the data behind it. ChatGPT’s pull depends on how complicated your product is. The more complex the purchase, the more it pays off. On simple stuff, barely at all.
What 973 Stores Found: The Complexity Twist
Kaiser and Schulze, at Goethe University Frankfurt and Mannheim Business School, tracked 973 ecommerce sites for twelve months — about $20 billion in combined revenue — lining up 50,000-plus ChatGPT referrals against 164 million transactions from the traditional channels.
The headline result: ChatGPT’s conversion rate beat paid social, but trailed the mature channels like organic search and email. Fine — roughly what you’d expect from a young channel.
But the interesting bit is the catch. Product complexity moderated the effect. The more complex the category, the better ChatGPT referrals did — on both the money they brought in and their share of traffic. On simple products, the gap against traditional channels mostly vanished.
So this isn’t one number that applies to everything you sell. It bends with the product.
The Vertical Data Makes It Sharper
Alhena AI’s report on 329 brands across eight retail verticals draws the same curve. CVR (conversion rate — the share of visitors who actually buy) by category came out like this:
- Beauty & skincare: 5.36%
- Health: 4.68%
- Auto parts: 3.53%
- Fashion: 2.40%
- Electronics: under 2%
Look at what’s on top. Every one of them is a “do some homework before you buy” category — match skincare to your skin type, compare supplements to a health goal, find the part that fits your exact car. The more the decision needs a back-and-forth, the higher the AI-chat conversion.
And one more number worth sitting with: LLM (large language model) traffic is only about 1% of all visits, but it’s pulling roughly 10% of sales. Small in volume, heavy in quality — these visitors show up already close to buying.
Adobe Says The Same Thing About Complex Buys
Adobe’s report — built from trillions of site visits plus a survey of 5,000-plus shoppers — lands in the same place. Its highest-converting categories were electronics and jewelry; the lowest were apparel, household goods, and food.
The survey put it even more directly: 87% said they’re more comfortable using AI for big or complex purchases. Shoppers themselves think of AI chat as the tool for the hard decisions, not the easy ones.
One more tell. 86% of GenAI traffic came from desktop, against 34% for ecommerce overall. So AI product research isn’t thumb-scrolling on the couch — it’s the sit-down-and-compare kind of session. People settle in for it.
Why Complex Products Are Where AI Wins
So why does the pattern bend this way? It comes down to what a chat interface is good at.
Old-school search is keyword-in, list-out: type “cheap tissues”, pick one, done. For a simple product that’s plenty.
But make the purchase complicated and search starts to strain. Try filtering “a skincare set for dry, sensitive skin, under about $50 a month, that fits a morning routine” through a search box. It’s painful.
A chat handles exactly that kind of multi-condition sifting, as a conversation. You can ask follow-ups, or reorder what matters halfway through. (Though it’s not magic: how you ask, and what the model was taught, can swing the answer a lot, so it’s no oracle.)
So the messier the comparison, the more a conversational tool earns its keep. When the decision won’t fit in one search box, the LLM turns into a genuinely strong channel.
Conclusion: Start GEO Where The Buying Is Hard
LLM referrals don’t pay off evenly across a catalog. The room for AI to act as an advisor opens up on — the ones with lots of conditions to weigh and a long stretch of research before the buy. So rolling GEO out across everything at once is the wrong move. Sequence it.
The practical read: split your catalog into high-involvement and commodity, then pick two or three categories from the first bucket to test. Alhena’s vertical numbers and Adobe’s category ranking give you a starting map; lay your own conversion rates and support-inbox questions on top, and the list of where to invest gets short fast.
If you want the companion piece — how to read ChatGPT’s conversion numbers without getting fooled by a single figure — I dug into that from the same 973-store study over here.
Sources
- Kaiser, M. & Schulze, C. (Goethe University Frankfurt / Mannheim Business School), “ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels?”, SSRN Working Paper, 2025, DOI: 10.2139/ssrn.5585812
- Alhena AI, “329 Brands, 9 Channels: The AI Commerce Report for 2026”, alhena.ai, 2026
- Adobe, “Adobe Analytics: Traffic to U.S. Retail Websites from Generative AI Sources Jumps 1,200 Percent”, Adobe Blog, 2025
