OpenAI just quietly scaled back its ecommerce ambitions. According to OpenTools, the company has “recalibrated” its shopping strategy after early pilots didn’t deliver the transaction volume they expected.

They’re not abandoning commerce entirely, but they’re focusing less on being a checkout destination and more on what LLMs actually do well: helping people discover and research products.

It’s a clarifying moment, because the same week OpenAI pulled back, Microsoft Clarity published research showing that AI-generated traffic converts at 3x the rate of other channels.

And Search Engine Land found that while LLM referrals drive strong engagement, the conversion mechanics are still messy.

The pattern is clear: LLMs are exceptionally good at discovery. They’re not yet good at transactions.

And if you’re a brand trying to show up in AI-driven shopping experiences, understanding that gap is everything. Let me show you what the data actually says, and what you should do about it.

TL;DR:

• OpenAI scaled back shopping features after transactions underperformed, refocusing on discovery and research use cases
• Microsoft Clarity found AI traffic converts 3x better than other channels, but the sample size is still small
• LLMs excel at helping users explore options and compare products, but struggle with checkout friction and purchase completion
• The opportunity isn’t in selling through ChatGPT, it’s in being the brand LLMs recommend when users are ready to buy
• Right now, the battle is for discovery and consideration. Purchases may come later, but winning the research phase is what matters today
• Brands that optimize for AI visibility during discovery will own the consideration set, even if transactions happen elsewhere

Why OpenAI Pulled Back (and What It Means)

OpenAI’s recalibration isn’t about LLMs being bad at commerce. It’s about LLMs being bad at *checkout*.

The company reportedly found that users loved asking ChatGPT for product recommendations and research, but when it came time to complete a purchase inside the interface, friction killed conversion. Payment setup, trust signals, shipping details, all the operational machinery that makes ecommerce work, doesn’t translate well into a conversational interface. Not yet, anyway.

What OpenAI learned is what we’ve been seeing in our data for months: LLMs are discovery engines, not transaction engines. Users come to ChatGPT, Perplexity, or Google AI Overview to figure out *what* to buy, not *where* to complete the purchase. They want recommendations, comparisons, and context.

Once they’ve made a decision, they go directly to the brand’s site or Amazon to finish the transaction. The handoff is the hard part. This is actually good news for brands. It means the game isn’t “get featured in ChatGPT’s shopping cart.”

The game is “get recommended by ChatGPT when users are researching.” That’s a different challenge, and it’s one you can solve today.

LLMs (of today) Are Built for Discovery

Let’s talk about what LLMs are exceptional at. They’re great at synthesizing information, comparing options, and explaining tradeoffs. If a user asks “what’s the best moisturizer for dry skin under $30,” an LLM can pull from reviews, ingredient lists, brand positioning, and user sentiment to give a nuanced, personalized answer. That’s valuable. That’s where the magic happens.

The Search Engine Land study backs this up. LLM referrals drive high-intent traffic with strong engagement metrics. Users who land on a site from an AI recommendation spend more time browsing, view more pages, and show clear purchase intent. The problem isn’t quality of traffic, it’s conversion rate once they arrive.

That gap exists because the user experience breaks between discovery and purchase. They found the right product, but now they’re on an unfamiliar site, figuring out shipping costs, return policies, and whether to trust this brand.

Microsoft Clarity’s research found the opposite: AI traffic converts 3x better than other channels. But the nuance matters, the sample size is still limited, and the sites seeing that lift are likely optimized for high-intent visitors who’ve already decided. The takeaway isn’t “AI traffic always converts better.” It’s “when AI sends someone who’s done their research, they convert fast if your site is ready for them.”

The Discovery-to-Transaction Gap

What matters now is understanding the handoff. LLMs can tell a user “you should buy Product X from Brand Y” with incredible confidence and context. But getting that user from recommendation to checkout involves a transition that’s still clunky. They have to leave the LLM interface, land on your site, rebuild trust, and figure out your checkout flow. Every one of those steps introduces friction.

OpenAI’s pullback reflects this reality. Users don’t want to buy inside ChatGPT. They want ChatGPT to tell them what to buy, and then they want to complete the purchase somewhere familiar and trusted. That’s not a problem, that’s just where weare in the evolution of AI-driven commerce.

The brands that win are the ones that get recommended during discovery, then deliver a seamless experience once the user arrives. This is why visibility in LLM outputs is so critical. If you’re not showing up when users are researching, you’re not in the consideration set.

And if you’re not in the consideration set, it doesn’t matter how good your checkout flow is. The battle is being won or lost during discovery, not during transactions. For now, that’s where the opportunity lives. Purchases inside AI interfaces may come later, but right now, winning the discovery phase is everything. That’s the shift that matters.

What Discovery Optimization Actually Looks Like

Discovery optimization is about making your brand’s value proposition, positioning, and proof points legible to LLMs when they’re synthesizing recommendations. That means structured product data, clear differentiation, authentic customer reviews, and content that answers the “why this product” questions users are asking.

When an LLM is deciding which moisturizer to recommend for dry skin, it’s pulling from what it can access and understand. If your product pages are thin on detail, if your reviews are sparse or generic, if your positioning is vague, you’re making it harder for the LLM to recommend you with confidence.

The brands winning in AI discovery are the ones making it easy for LLMs to explain why they’re the right choice.

This also means thinking about the content layer that sits between product pages and blog posts. Comparison guides, ingredient breakdowns, use case explanations, this is the content LLMs love to cite.

If your site only has transactional product pages and top-of-funnel blog content, you’re missing the mid-funnel layer where discovery happens. That’s where you need to invest.

What You Should Do Now

1. Audit your brand’s visibility in LLM outputs. Use tools like our [Commerce GPT Visibility Tool](https://commerce-gpt.yotpo.com) to see if you’re showing up when users ask relevant product questions. If you’re invisible during discovery, you’re losing before the transaction even starts.

2. Optimize your content for LLM comprehension. That means clear product descriptions, structured data, authentic reviews, and content that answers “why this product” questions. LLMs pull from what’s available, make sure your brand’s positioning is clear and discoverable.

3. Build a post-recommendation experience that converts fast. If someone arrives at your site from an AI recommendation, they’ve already done their research. Don’t make them start over. Highlight trust signals, make checkout frictionless, and give them confidence they’re in the right place.

4. Track where your AI traffic is coming from and how it behaves. Install Microsoft Clarity or similar tools to see how users from LLM referrals move through your site. If they’re bouncing, figure out why. If they’re converting, double down on what’s working.

5. Focus on mid-funnel content that LLMs cite. Comparison guides, ingredient breakdowns, use case explanations, this is the content that gets pulled into AI recommendations. If your site only has product pages and a blog, you’re underinvesting in the discovery layer.

6. Don’t try to force transactions inside AI interfaces. The data says users don’t want that yet. Instead, make sure that when an LLM recommends you, the path from recommendation to your site to checkout is as smooth as possible.

7. Test your site experience for high-intent visitors. Users arriving from AI recommendations have different expectations than users arriving from paid search. They’ve already decided on a product category and often on specific features. Make sure your landing experience reflects that.

Track your visibility for free here: Commerce GPT Visibility Tool