New research on $20B in transactions reveals a counterintuitive problem—and an unexpected solution
A year ago, when ChatGPT started sending customers directly to e-commerce sites, the conventional wisdom was clear: these would be your best customers. They’d arrive educated, well-matched, and ready to buy…for free.
The conventional wisdom was wrong.
New research from Kaiser and Schulze analyzing 973 e-commerce websites reveals something nobody expected: ChatGPT traffic converts at roughly half the rate of Google organic search, generates significantly less revenue per session, and shows declining average order values[^1].
This isn’t a measurement problem. It’s a fundamental mismatch between how AI prepares customers and how your website converts them.
But here’s what matters more: the same questions AI-educated customers need answered are the exact questions your traditional traffic is struggling with. They just haven’t had ChatGPT to ask yet.
The Data Nobody Wanted to See
The study tracked 12 months of performance from August 2024 through July 2025—the entire first year after ChatGPT began providing e-commerce links. The scale is unprecedented: $20 billion in combined revenue, over 50,000 AI-referred transactions compared against 164 million from traditional channels[^1].
ChatGPT dominates AI traffic with more than 90% of all LLM referrals. And that performance is disappointing. Organic search converts at 3-3.5%. ChatGPT traffic converts at roughly 1.5-2%—a gap of 40-50%[^1].
The only bright spot? Bounce rates. ChatGPT traffic actually bounces less than most traditional channels, suggesting strong product-customer matching. Users who click through from ChatGPT are finding relevant products. They’re just not buying them.
Why Educated Customers Are Harder to Convert
The data confirms ChatGPT is doing its job well. Low bounce rates prove good matching. The problem isn’t the quality of the referral—it’s what happens after customers land on your site.
Think about what ChatGPT does before sending someone to you. It elicits preferences through conversation. It compares products across multiple attributes. It explains trade-offs between features and price points. By the time a user clicks through, they’ve had an extensive consultation.
Your website, meanwhile, was designed assuming customers arrive knowing almost nothing.
Traditional e-commerce architecture is built around discovery and persuasion. You guide users from homepage to category page to product page. You build trust through brand storytelling. You educate them about features. The entire funnel exists to move someone from “vague need” to “confident purchase.”
ChatGPT customers bypass this entire funnel. They land directly on product detail pages via deep links, having already completed the education, comparison, and decision-making process.
But here’s the critical insight: these educated customers aren’t just skipping your funnel—they’re arriving with fundamentally different psychology.
The architectural mismatch: Your product pages aren’t designed to be landing pages for cold traffic. Users miss your homepage trust signals, your brand story, your category-level merchandising, your cross-sell opportunities. They land on a page optimized for the middle of a journey they never took.
The psychological mismatch: Your pages are built for discovery mode—extensive product descriptions, educational content, comparison tools, reviews for confidence building. ChatGPT users are in verification mode. They want to confirm the price is accurate, check if it’s in stock, and complete the transaction quickly. The session duration and page view data confirms this: they’re task-oriented, not exploratory[^1].
The economic mismatch: This is the most concerning. The research shows ChatGPT traffic has declining average order values even as conversion rates improve. This isn’t random—it’s structural. ChatGPT shows customers multiple price points, cheaper alternatives, and explicit value-for-money reasoning. They arrive price-anchored to options your site never wanted them to see[^1].
Traditional channels let you position premium products without direct price comparison. You can guide customers from good to better to best. You create value perception before introducing price. ChatGPT eliminates this advantage, and the declining AOV trend suggests it’s getting worse, not better.
The Hidden Opportunity: Your 96% Problem is Bigger Than Your 0.2% Problem
Here’s the strategic insight the research reveals: ChatGPT traffic represents only 0.2% of total e-commerce volume[^1].
Your traditional channels—organic search, paid search, paid social, email—represent 99.8% of traffic and convert at just 3-4%. That means 96-97% of your visitors don’t convert.
And those non-converting customers are struggling with the exact same questions that ChatGPT answers before sending its educated 0.2%.
The difference? ChatGPT users got answers through conversation with AI. Your traditional traffic gets static product pages, overwhelming navigation, and zero guidance.
Why Conversational AI Solves Both Problems Simultaneously
Most companies will try to fix their ChatGPT traffic problem through architectural changes—redesigning product pages, implementing deep-linking infrastructure, building verification-optimized experiences. These projects take 18-24 months and massive investment.
There’s a faster, smarter path that solves both problems at once: Deploy conversational AI for your traditional traffic first.
First, you immediately impact your largest problem. 99.8% of traffic gets ChatGPT-quality guidance without leaving your site. Based on real-world deployment data, a fashion retailer saw conversion rates jump 147% in 30 days, with average order values up 42% and returns down 72%[^2]. Additional revenue lift was 4.2%.
Second, you build a learning flywheel. Every conversation with traditional customers reveals what questions people ask before buying, what trade-offs they care about, what causes hesitation, and what builds confidence. These are the exact same questions AI-educated customers asked ChatGPT before arriving at your site. By capturing this intelligence from your 99.8%, you learn how to serve your 0.2%.
Third, the system gets smarter over time. The fashion retailer’s data shows performance improving dramatically: 180-day conversion rate improvement was 14%, 90-day improvement was 21%, but the most recent 30-day period showed 147% improvement[^2]. Start earlier, learn faster.
Fourth, deployment is fast. Conversational AI takes hours on Shopify to two weeks on Salesforce[^2]. Structural website redesign takes 6-18 months. You’re generating ROI immediately versus waiting for delayed returns.
Finally, you use insights to fix structural issues faster. Once you understand what customers actually need from conversation data, you can prioritize the right architectural changes for both traditional and AI traffic.
The NewEcom.AI Approach: Built for E-commerce Profitability
This isn’t theory. NewEcom.AI proprietary technology (patent pending) has deployed conversational commerce specifically designed for these challenges.
At the core is what they call the “Intentional Data Model”—a learning system that enriches your product catalog based on real customer conversations[^2]. It takes your product catalog, descriptions, specs, and images, combines them with brand and industry contextual data, then learns from every customer conversation what works and what doesn’t.
The system correlates intentions with conversions, identifies which explanations drive purchases, learns which product attributes matter for different customer segments, and fine-tunes recommendations based on outcomes. Your catalog becomes smarter over time, understanding not just what products are, but how to explain them in ways that convert.
The conversational interface is purpose-built for e-commerce, not generic customer service. It handles multi-turn conversations that refine complex purchase intentions, provides side-by-side comparisons with contextual explanations, and uses matching logic that considers both product attributes and customer motivations.
Product pages become interactive, with predefined dynamic questions tailored to common inquiries and specific product context. Answers are hyper-personalized and automatically added to conversation history, building a complete shopping journey maintained across sessions.
The system maintains multi-tab conversations enabling parallel shopping, full conversation history for follow-up questions, and personal profile building with recommendations. It’s mobile-first by design—floating widget, movable and resizable, with multi-modal interactions including text, picture, and voice.
Between March 17 and September 18, 2025, a fashion retailer deployed NewEcom.AI’s platform. During the most recent 30-day period, sessions using conversational AI achieved a 147.82% increase in conversion rate, a 42.82% increase in average order value, an additional 4.20% lift in total revenue, and a 72.53% reduction in return rates due to sizing issues[^2].
Over the 180-day period, the conversion rate for sessions with conversational AI rose from 2.23% to 4.81%—more than doubling. This isn’t a marginal improvement; it’s a transformational impact on the 96% of traffic that previously wasn’t converting.
The Strategic Sequencing That Actually Works
The optimal path forward starts with deploying conversational AI for traditional traffic immediately. Focus on capturing the 96% conversion opportunity with proven technology. Platform integration takes hours to two weeks depending on your e-commerce platform. Catalog synchronization and enrichment is automatic. Start with paid social since it’s your worst performing channel with the most to gain, then expand to exploratory organic search with high bounce rates, while preserving your high-intent traffic experience to avoid disrupting what already works.
You’ll see immediate conversion rate improvement of 15-50% depending on segment, AOV lift from guided bundles and upsells, returns reduction from better product-customer matching, and rich conversation data revealing customer needs. Investment is platform deployment plus integration costs—significantly less than structural redesign.
Over months three through nine, use conversation data to understand both customer types simultaneously. From traditional traffic questions, you learn what customers need to know before buying, what causes hesitation, what builds confidence, and what drives premium purchases. Applied to AI traffic, these are the same questions ChatGPT answered—now you know what AI-educated customers already understand and what they still need verified.
You’ll continuously improve conversion rates and AOV through the flywheel effect while building a clear roadmap for structural changes based on actual customer behavior, not guesses. Your organization learns what customers actually need.
Then in months nine through eighteen, make strategic infrastructure changes informed by data. Redesign product pages as landing pages using conversation data to know what trust signals actually matter. Build deep-linking architecture for AI traffic based on where they actually need to land. Create verification-optimized flows addressing what AI customers need to confirm. Implement technical integration with LLM platforms for real-time pricing and inventory APIs.
This sequencing works because you’re not guessing what to fix—conversation data tells you. You’re generating ROI during the learning phase, not after. Infrastructure changes serve both traditional and AI traffic. You’ve built organizational capability in conversational commerce before competitors even start.
Why This Beats the “Wait and See” Approach
Consider the alternatives. You could ignore AI traffic since it’s only 0.2%, but when it scales to 5-10%, you’re 2-3 years behind. You miss the opportunity to learn from traditional traffic conversations and give competitors time to build moats.
You could immediately redesign for AI traffic, but that’s an 18-24 month infrastructure project requiring massive investment with uncertain ROI. It risks disrupting traditional traffic that’s already working and does nothing to improve your 96% conversion problem.
Or you deploy conversational AI for traditional traffic first. You get immediate ROI from 99.8% of traffic, learn customer needs that inform both traditional and AI optimization, achieve fast deployment measured in weeks not years, see compound improvements from the learning flywheel, and generate revenue while building strategic capabilities.
This approach addresses the three threats the research identifies. AI platform disintermediation becomes less threatening when you build world-class conversational commerce on your own properties, reducing dependence on ChatGPT sending you price-sensitive customers. You own the conversation and the relationship.
Declining AOV from AI traffic becomes manageable because your conversational AI can guide bundles, explain premium value, and overcome price anchoring by controlling the entire conversation—not just the final verification step.
Competitor advantage shifts in your favor. While others struggle with ChatGPT’s 0.2%, you’re capturing the 96% opportunity and building proprietary intelligence that compounds over time.
What This Means for Q4 2025 Planning
If you can improve conversion by 50-150% on traditional traffic while learning how to serve AI customers, why wait? The math is straightforward: allocate 20% of your paid social budget to conversational AI that converts 3x better and the ROI is immediate.
The question isn’t whether conversational AI is the future—BCG projects that $1 trillion in spending (50% of e-commerce) will be agent-assisted within a few years[^2]. The question is whether you’ll have spent 2025-2026 planning infrastructure changes, or whether you’ll have deployed, learned, earned ROI, and built advantages that compound.
When that shift happens, the companies with years of conversational commerce experience and proprietary customer intelligence will dominate.
ChatGPT traffic is 0.2% today. But it’s growing, improving, and will eventually become table stakes. The choice you make in Q4 2025 determines your competitive position for the next five years. The research provides the evidence. The proven deployment model provides the roadmap.
The window to act is now.
Ready to see how conversational AI transforms your conversion rates?
NewEcom.AI has helped e-commerce companies achieve 3x conversion rate improvements and 40% increases in average order value while dramatically reducing returns. Our intelligent commerce infrastructure learns from every customer interaction, continuously optimizing for profitability.
Contact: Francois Silvain, CEO | fsilvain@newecom.ai | +1 781 491 3808 | www.NewEcom.AI
References:
[^1]: Kaiser, M., & Schulze, C. (2025). ChatGPT Referrals to E-Commerce Websites: Do LLMs Outperform Traditional Channels? University of Hamburg and Frankfurt School of Finance & Management.
[^2]: NewEcom.AI (2025). Maximizing Ecommerce Profitability: Intelligent E-commerce Infrastructure. Internal company data and BCG US Agentic Commerce Consumer Survey (N = 2,532), July 2025.
