Two Rails of Agentic Commerce
I see the emerging battle as a two-rail split:
Rail | Players | Discovery | Check out Route | Data Control | What brands must do |
Open Rail | ChatGPT + Walmart/Etsy/Shopify etc. | Agent surfaces via conversation prompts | Instant Checkout (or multi-item cart later) | Shared / mediated by the assistant | Optimize content for agent access, in addition to SEO and humans, enable bundling, stitch identity, capture downstream LTV |
Closed Rail | Amazon / Rufus / Proprietary Assistants | Discovery inside a retailer’s AI layer | Checkout inside Amazon / via its system | Full control of signals, ranking, ads, attribution | Feed agent-friendly product data, seed Q&A, manage adversarial prompts, measure agent conversions internally |
So rather than “which side do I pick?”, brands must optimize for both rails simultaneously. The risk is a lopsided bet: gain traction in ChatGPT but lose in Amazon’s own closed loop or vice versa.
The Reality Check: Amazon Still Owns Discovery
For all the headlines about “open” commerce and agentic shopping, the data tells a harder truth: Amazon still owns the discovery moment and that dominance is actually growing.
According to Feedvisor’s 2025 Consumer Behavior Report, 80% of shoppers now begin their product search on Amazon, up from 60% the year prior. This signals more than incremental growth, it’s an acceleration of habit. Consumers trust Amazon’s reviews, delivery reliability, and frictionless path to purchase.
“A standard search bar is no longer the fastest path to purchase; rather, we must use technology to adapt to customers’ individual preferences and needs.”
— Suresh Kumar, Global CTO & Chief Development Officer, Walmart Inc.
Even with Walmart’s ChatGPT integration, that search gap remains wide. Walmart follows at 50%, while Google sits at 42%. Social channels—Facebook, Instagram, and TikTok—collectively influence 48% of product discovery. This fragmentation underscores a key truth: AI assistants and conversational commerce aren’t replacing search yet, but they’re augmenting it.
And when shoppers do turn to AI assistants, Amazon is still part of the conversation. Among consumers who have used an AI assistant for product searches:
- 26% used ChatGPT
- 18% used Google’s Gemini
- 13% used Amazon Rufus
- 7% used Microsoft Copilot
- 4% used Anthropic’s Claude
A striking 93% found them helpful, suggesting that while the volume of users remains modest, satisfaction and trust are high, especially for Amazon’s Rufus, which has evolved from early criticism (“unhelpful,” “just plain wrong”) into a functional, intuitive shopping guide that can interpret prompts like “help me find a gift for Mom.”
What it means for brands
AI discovery may be fragmenting, but Amazon still sets the baseline expectations for speed, accuracy, and relevance. Walmart and ChatGPT are building an open alternative, but the reality is, most consumer journeys still begin and end on Amazon.
That makes the “two rails” approach even more urgent:
- The closed rail (Amazon) isn’t just walled, it’s winning.
- The open rail (Walmart + ChatGPT) is the emerging counterbalance, creating optionality and future-proofing for brands.
Optimizing for both is more risk mitigation than redundancy, but if you had to choose between one or the other, Amazon is still your best bet.
How to Win on Both Rails
Rail A: Open (ChatGPT: Walmart, Etsy, Shopify)
- Structure for agents, not search engines
- Write titles, bullets, and specs that disambiguate variants, sizes, and compatibilities.
- Include “best for” statements (“best for travel,” “under $50,” “vegan-safe”) so AI assistants can surface your products confidently.
- Prepare for agent-assembled carts
- Design pre-bundled kits and complementary products so ChatGPT can assemble multi-item baskets that make sense.
- Include metadata like “compatible with X” and “pair with Y” to improve bundle accuracy.
- Fix the attribution gap
- ChatGPT checkout flows often default to “guest” mode (like Etsy’s current integration).
- Add deep-link tracking, claim-your-order emails, and post-purchase account stitching to reclaim first-party data and LTV.
- Build a “Prompt Map”
- Identify your top 100 natural-language queries (e.g., “eco-friendly cleaning products under $20,” “beginner guitar kit for teens”).
- Map each to a curated bundle or PDP collection page.
- Plan for transparency and fallback
- Provide short, factual “why this product” explanations.
- Add clear return and guarantee policies the assistant can quote when users ask follow-up questions.
Rail B: Closed (Amazon: Rufus)
- Feed Rufus the right signals
- Optimize PDPs for readability: comparisons, benefits, proof points, and certifications.
- Use high-quality reviews, full attribute coverage, and consistent pricing to ensure Rufus and AI summaries surface your brand correctly.
- Seed conversational Q&A
- Add FAQ content that matches real shopper phrasing: “Which is better for oily skin?”, “Can I use this for hiking?”
- This helps Rufus learn context from real questions, not just keywords.
- Reduce volatility
- Frequent price or stock changes can lower confidence in AI recommendations.
- Use stable signals during campaigns or promotions.