As I was having conversations with marketplace founders about AI over the last few months, I felt that most founders were focusing on the wrong aspects of AI.

1. People’s Main Worry: LLMs Capture the Top of the Funnel.

Intuitively, this concern makes sense. If your first destination for everything is Chat GPT, it feels like they would capture a lot of value through some sort of lead-gen fee when referring transactions to your site. Worse, they can perhaps be the only destination by integrating payments and having people buy directly from Chat GPT, thus commoditizing e-commerce sites and marketplaces by sending transactions to the lowest price provider and obfuscating the ultimate provider, thus decreasing the value of your brand.

I’ve heard many variations of this concern. Honestly, it feels overblown.  There are three ways people shop online.

a. Shopping as Entertainment.

This is the online equivalent of window shopping. Consumers often go to sites like Vinted and classified sites like OLX without a specific idea of what they are looking for. They enjoy browsing through the listings. If something catches their eye, they buy it. The time on site for these sites is very high, as is the page view per visit, but the visit-to-purchase rate is relatively low, and prices are typically low, with an average order value often significantly below $100.

It’s hard to imagine LLMs playing a large role in disrupting this category. The point is not to make the experience more efficient, merely to see items that you like and might inspire you. It would take Open AI and its ilk the ability to create product browsing streams to match your preferences, despite not having the data and history of all your prior purchases on those sites. Moreover, even if they succeeded, given that these categories are winner takes most, the feed would essentially have most of its content from the dominant provider, who would still capture most of the transaction value.

b. Search

The second main way people shop is by searching for exactly the item they want. This is the main purchase behavior on Amazon and eBay. People know what they are looking for, and they enter the exact make and model. While AI can somewhat improve the search results and suggest related items people can buy, there isn’t a huge scope for LLMs to play. Because Amazon and eBay are dominant in many product categories, a lot of users just go there. Combined, they have 43% of the US e-commerce market and are hyper-dominant in certain categories like electronics and books.

Now, some people start their product search on Google, and it does capture some value through sponsored ads, but even then, many, if not most, of the results come from Amazon and eBay for many product categories, and they capture most of the transaction value.

To the extent people start their search journey on Chat GPT rather than Google, I can see them capturing the same type of value that Google captures through sponsored ads, but not more than that.

c. Considered Purchases.

If you do not know what you are looking for, generative AI can play a role in figuring out what you should be buying. This is true both for high-value items like houses and cars, but also for specialty items like high-end ski equipment or bespoke travel recommendations.

Sites like Curated, Fora Travel, and Stitchfix currently use human experts to help you buy outdoor equipment, plan your travel, and do your styling. I can easily imagine AI agents taking over from human experts. It’s perhaps the reason Curated only sold for $330M after raising $141.5M.

Anecdotally, I used GPT for car and real estate recommendations, so I can definitely see it play a role there. However, even here, it’s not obvious that ChatGPT ends up being the end-all be-all destination site for purchase recommendations. I can easily imagine Carvana having a better AI for car selection than Chat PT. Likewise, it could very well be that the current e-commerce transaction sites win their respective verticals. They have the data and specialized knowledge to make the perfect recommendations.

You may very well ask Instacart’s AI for cooking recommendations and Amazon’s AI for general purchase recommendations.

Conclusion

LLMs primarily serve as a direct substitute for search engines like Google because their core value lies in rapidly delivering information, answers, and synthesized content. This is the role search engines also do, albeit less effectively. This is why ChatGPT is an existential threat to Google.

Google does not pose an existential threat to e-commerce sites and marketplaces as they don’t, and don’t want to, handle inventory management, payment processing, fulfillment, and customer support.

The same applies to LLMs. Even if LLMs magically captured a large part of the top of funnel discovery, which as argued above, I do not think will happen, because the value of marketplaces is in their liquidity which makes them winner takes most, the dominant marketplaces per vertical would continue to dominate and have the selection of items people want with the best in class fulfillment and they would keep capturing most of the value.

Should you index yourself in LLMs?

The short answer is that if you index yourself in Google, you should index yourself in LLMs and take advantage of LEO (LLM Engine Optimization). Most e-commerce sites and marketplaces choose to index themselves in Google and, as such, should index themselves in LLMs. The exception is if you have 95% market share in a category, you should not index yourself in either, as you do not want to train users to go to search engines or LLMs to start their journey. However, this is exceedingly rare.

2. The Impact AI is Actually Having on Marketplaces.

While most people are overly worried that LLMs will capture most of the value in transactions, they ignore the actual impact AI is already having on marketplaces on a global level.

a. Cross-Border Commerce.

AI is facilitating true cross-border commerce for the first time. It used to be that when you launched a used good marketplace in Europe, you launched in the UK separately from Germany and France. The listings were unique per country, and there was no cross-border commerce.

With AI, listings and conversations between users are now automatically translated. With cross-border payments and logistics increasingly solved, this is leading to the emergence of truly pan-European companies like Vinted and the explosion in global cross-border commerce. It’s happening regionally with players like Wallapop using their Spanish dominance to sell in Italy and Spain and Ovoko sourcing car parts in Eastern Europe to sell in France.

It’s also happening in B2B with companies like CarOnSale now selling 30% of their cars cross-border.

b. Simplified Listing Process.

It used to be that to list an item on a marketplace you had to take lots of photos, select a category, enter a title, write a detailed description, select a price and follow a complicated multi-stage listing process.

Companies like Rebag, HeroStuff, and CollX are re-inventing the listing process. Their AI can use as little as one photo to identify the item, create the listing and recommend the price!

c. Improved Listings.

Beyond simplifying the listing process, AI can be used to improve listing quality to increase the sell-through rate. Products like Photoroom allow you to remove image backgrounds and replace them with the background that would best allow your item to sell given its category and the marketplace you are selling on.

d. Productivity Improvements.

Startups are all using AI to improve productivity. It’s true in every category, including sales and marketing, but it’s especially true in customer service and for programmer productivity.

i. Customer Service.
E-commerce sites and marketplaces are using AI to lower customer service costs and improve customer satisfaction. For instance, over the past two years, the percentage of customers interacting solely with Sebastian, Carvana’s AI customer support agent, without needing human assistance, has nearly tripled.

The use of AI to improve customer service is commonplace. Zappos leverages AI-driven FAQs and chatbots that understand sizing concerns, returns, and product queries. Chewy uses AI-assisted ticket classification to prioritize and route queries to the right human rep or auto-resolve common issues like order tracking. Glossier uses AI tagging for incoming messages to identify trending issues or pain points. The AI also helps auto-draft replies that human agents just tweak — making the service faster and on-brand. Expedia built an AI customer assistant that helps resolve flight changes, hotel cancellations, and payment issues faster than human reps in many cases.

ii. Programmer Productivity.
All startups are using AI to improve programmer productivity, and e-commerce sites and marketplaces are no exception. Here are the general categories of use.

• Code Generation & Autocomplete: GitHub Copilot & Cursor.
• Backend & API Optimization: OpenAI API + LangChain / Vercel AI SDK & CodiumAI.
• Frontend & UX Flow Automation: Builder.io + AI Copilots & Vercel’s AI SDK (for Next.js apps).
• Ops, Data, and Dev Workflow Boosters: AirOps / Hex / GPT-4 for Analytics & Replit Ghostwriter / Amazon CodeWhisperer.
• AI-Powered CI/CD & DevOps: Aider & Sweep.dev.
• AI for Security and Compliance (especially in fintech/kids/health verticals): Snyk + AI.

To give a specific example, Shopify engineers are rumored to use a combination of Copilot, in-house GPT tooling, and prompt-driven test writing for their Hydrogen + Remix stack. Startups building on Shopify or Stripe ecosystems tend to mimic this.

Conclusion
Too many marketplaces are fearful of the impact LLMs may have on their business, whilst ignoring the many opportunities AI can provide them. They should use the massive troves of data they have on consumer behavior and transaction history to create their internal shopping recommendation LLM. In addition, they should use AI to simplify listing processes, improve productivity, and potentially enable cross-border commerce. They should index themselves in various LLMs and be an early adopter of LEM, taking advantage of the free traffic to gain share against slower-moving and more conservative competitors.