‍

In April this year, OpenAI announced that it had updated ChatGPT’s web search capabilities to improve online shopping for users with personalized product recommendations including images, reviews and direct purchase links. In July, the Financial Times reported that they plan to take a cut from sales made directly through ChatGPT.

Today, ChatGPT powers over 1 billion web searches per week and drives a growing percentage of retailer traffic — a source that has grown 1,200% in just seven months. For any online merchant, this represents a fundamental shift in how consumers shop.

In this post, I'll explain why OpenAI poses an existential threat to online retailers — and what merchants must do now to protect their businesses.

What threat does OpenAI pose online retailers today?

Simply put, your customers might end up buying your products on ChatGPT rather than your own website. And this might be most if not all of your customers: there is a very real possibility that ChatGPT becomes the default place for people to shop. OpenAI will not only own all the demand for your products, but they will be able to own the shopping journey to checkout (and potentially beyond), with you relegated to the role of fulfilling the order and processing returns.

The potential consequences are significant:

• You would no longer own the customer relationship, OpenAI would
• There is a real risk of disintermediation: it is not clear what options you will have to ensure that OpenAI keeps funnelling demand in your direction versus your competitors
• Your margins will suffer. If OpenAI owns demand, most of the customer journey and the customer relationship, they can command a larger percentage of the purchase price 

Why? Because shopping on ChatGPT will be incredibly compelling…

OpenAI is in a position to make the complete customer journey around online commerce much more compelling and efficient than it is today.

Product discovery in ChatGPT

Because consumers go to ChatGPT when they have problems, OpenAI is in a unique position to understand what products they need to solve those problems – before they have even identified those needs!

Over time, OpenAI will be able to build a sophisticated understanding of consumer preferences, and marry that understanding with the whole universe of products to deliver particularly effective product suggestions.

A great place to research products

ChatGPT is an amazing place for consumers to do product research. Instead of having to go out and search multiple different authorities (specialist publications, social forums etc.) on different websites, absorb and assimilate that information, and then apply it to their specific needs, the consumer can just ask ChatGPT to do it. ChatGPT will perform the search, collate the results, assimilate it and synthesize it with its own understanding of the consumer’s problem and the consumer’s perspective. As a result, this will save the consumer a huge amount of work and compress the entire consideration stage into a much shorter, more efficient journey. 

A great place to buy products

OpenAI is likely to make buying in ChatGPT very easy, with agents handling the complexity of submitting payment details and organizing delivery. It is much easier for a consumer to tell an agent to “pay with my American Express, and make sure the delivery is on either Tuesday or Wednesday after 1pm” then try to figure out how to arrange that via a traditional, static UI. 

And a great place for post-purchase support

Consumers can go back to ChatGPT once a product has arrived with questions on how to use it. This process is almost certainly a better experience than referring to the user manual directly (if they’ve kept it), or Googling it and then reading the online version. Similarly, if a user wants to return a product, they can simply ask ChatGPT “can you please print a return label for me, and let me know where the nearest drop-off location is?”

…More importantly, consumers will live in ChatGPT

As ChatGPT becomes more and more useful across a wider range of tasks, it is likely that consumers will spend more and more time in ChatGPT. 

ChatGPT is likely to become the “homescreen” for our everyday lives: somewhere we can check the news, plan our day, send messages to friends and co-workers, check our emails, shop and more.

Sam Altman, the CEO and co-founder of OpenAI, has had a strong consumer internet bias ever since co-founding Loopt, a “check-in”-based mobile social app, in 2005. Building the next billion-consumer internet platform, and achieving Artificial General Intelligence, are not conflicting goals – it is precisely the behavior of a billion consumers that gives Altman the real-world training data for AGI.

With OpenAI’s move to build its own browser, it is likely that even when consumers are not in the ChatGPT UI, ChatGPT will still be there, browsing alongside the consumer. And through OpenAI’s tie-up with Apple’s generational designer Jony Ive, acquiring his startup io Products, we can expect even more form factor innovation around how consumers spend their days with ChatGPT.

That means there will need to be really compelling reasons for consumers to leave ChatGPT to browse and buy elsewhere. 

It will be okay! Retailers are well-placed to compete

The good news is that retailers can develop those really compelling reasons for consumers to leave ChatGPT and buy from their own storefronts. The most important things that online retailers must do is invest in delivering compelling, agentic shopping experiences to rival those that OpenAI is creating. The future of commerce is agentic (see Alex’s earlier piece on ambient agents), and agents can make online shopping both more enjoyable and more productive.

There are really good reasons why online retailers can build agentic experiences that are more compelling than those developed by OpenAI, and so give consumers the good reasons to shop outside of ChatGPT:

Focus and domain expertise

ChatGPT is a general-purpose technology: it has to work across all product categories, as well as performing an enormous number of functions that are nothing to do with online commerce. 

As an online retailer, you don’t have to think so horizontally. You can focus relentlessly on your vertical and customer niche, and build a really compelling experience just around that. That’s a level of specialization that OpenAI will never be able to afford.

Customer expertise

You know your customers really well, from years of successfully serving them and growing their value. That means you’re incredibly well placed to anticipate what kinds of next generation digital shopping experiences are going to resonate with them.

So what should retailers do?

Build compelling agentic shopping experiences for your consumers, experiences that are better than those that OpenAI can provide

As discussed above, the ‘number one’ thing retailers need to invest in is building out great customer-facing agents and agentic experiences, with a view to making these significantly more compelling for your customers than anything OpenAI is in a position to develop.

But what does this look like? The answer is going to require a lot of creativity, and vary vertical-by-vertical, and brand-by-brand, but to give a couple of examples:

Fashion retailer

Imagine an agent that knows your customer's measurements, past purchases, and style evolution. When a customer says "I have three weddings this summer," your agent doesn't just show dresses, it:

• Suggests outfits that work with shoes and accessories the customer already own
• Warns that the outdoor vineyard wedding needs block heels
• Remembers that this customer hates strapless styles — and that she returns 60% of size M items despite identifying as size M
• Shows how each dress would look on her specific body type
• Offers to book alterations at their preferred tailor (knowing their usual adjustments: “$40 hem, take in waist 1")

ChatGPT can show products well. But the kind of personalized styling advice above is only possible to deliver based on years of building a loyal customer base, collecting high quality data and using that data to build customer and fashion intelligence.

Home improvement retailer

Imagine a customer points their phone at a brown patch on their lawn. Your agent:

• Identifies the likely cause (grub infestation versus drought stress)
• Calculates the exact amount of treatment needed for their 2,500 sq ft yard
• Checks local regulations for pesticide use
• Suggests pet-safe alternatives because it knows they bought dog toys last month
• Schedules curbside pickup for Saturday morning when they usually shop

This requires deep product knowledge, local expertise, and customer history. Whilst ChatGPT could do all of that, they’re not going to deliver it in a seamless out-of-the-box way – the use case simply is not valuable enough for them as a generalist vendor.

Turn your customer data into your competitive weapon

The examples above aren't just hypothetical! They are powered by the customer data and intelligence that retailers have spent years building. This is one of your moats against OpenAI.

Consider what makes those experiences possible:

• The fashion retailer knows that that customer returns 60% of size M items because they've tracked every transaction
• The home improvement retailer suggests pet-safe alternatives because they've connected purchase history across categories
• Both can predict preferences because they've analyzed patterns across millions of customer interactions

Most retailers are already using this data to power recommendation engines, dynamic pricing, and targeted promotions. But that same intelligence can now enable your agents that feel almost telepathic in their understanding of customer needs.

The key is making your data actionable for agents. This means:

• Connecting siloed data sources (transactions, returns, customer service interactions, browsing behavior)
• Building customer intelligence layers that agents can query in real-time
• Creating feedback loops where agent interactions further enrich your understanding

You currently have 5-10 years of behavioral intelligence that OpenAI can't access — actual purchase patterns, return rates, lifetime values, and service histories. But your data moat is eroding. Every month you delay is a month closer to OpenAI building its own transactional intelligence through ChatGPT checkout. The question isn't whether you have an advantage — it's how quickly you can weaponize it before it disappears.

Be ambitious and creative: explore opportunities beyond “chatbots” – like generative UIs 

When people start working on agentic applications they inevitably fall into the trap of starting with a chatbot. As OpenAI has taught us all, chat is an incredibly powerful interface. But boiling everything down to chat, or simply adding chat alongside an existing digital experience, is not a compelling user experience.

And crucially: your chatbot is not going to compete effectively with ChatGPT, which is the king of chat interfaces.

Agentic applications have the possibility of delivering generative user interfaces: interfaces that are dynamically composed based on an understanding of who this user is, and what she is trying to do right now. These can be rich, interactive, and highly specific. There is so much scope for retailers to innovate here and come up with much better ways to browse and choose products. 

Again – exactly what this looks like is going to vary by vertical and brand. Let’s look at a couple of examples to get the creative juices flowing.

Dynamic Destination Discovery (Travel Retailer)

Instead of traditional search filters and static destination pages, imagine an interface that builds the perfect holiday as the customer explores:

• The customer starts with: "I need to recharge somewhere warm"
• The UI generates an interactive globe that highlights destinations based on their current season, spinning to show real-time weather
• As they add context ("but I hate crowds"), popular destinations fade while hidden gems emerge and grow larger
• Mentioning "I love history" causes ancient sites to rise from the map like a pop-up book, complete with time-travel sliders showing how sites looked across centuries
• Budget constraints don't just filter — they transform the entire view: "$500" might show weekend escapes with detailed hour-by-hour itineraries, while "$5000" reveals three-week adventures with interactive route maps
• The interface will adapt to show "Morning person paradise" or "Night owl heaven" versions as it becomes clearer which resonates better with the customer
• Saying "traveling with teens" morphs the entire experience to highlight TikTok-worthy spots, adventure activities, and WiFi strength indicators

The magic moment: When they hover over Lisbon, the interface doesn't just show hotels — it generates a living preview of their potential trip: morning coffee in Bairro Alto, afternoon tram rides, sunset from a miradouro, all animated with their travel dates' actual daylight hours and weather patterns.

This is radically different from scrolling through static hotel listings and pre-written destination guides. The interface is literally creating their unique trip visualization in real-time.

Dynamic Product Configurator (Furniture Retailer)

Instead of static dropdown menus, imagine a living room that builds itself as the customer describes their needs:

• Customer says: "I have a narrow living room with lots of natural light"
• The UI generates a 3D room matching those dimensions with realistic lighting
• As they mention preferences ("I love mid-century modern"), furniture appears in the space
• They can grab and move pieces, and the UI automatically suggests complementary items that fit the remaining space
• The interface reshapes itself based on what they're trying to do — expanding the color palette when they linger on fabric options, showing child-safe alternatives when they mention kids

These generative UIs represent a fundamentally different shopping paradigm to what ChatGPT can offer. While ChatGPT excels at conversation, it's constrained to a chat interface with images and videos. Retailers can create immersive, visual, interactive experiences that make shopping feel less like asking questions and more like exploring possibilities. This is where the future of digital retail lies — not in better chatbots, but in interfaces that couldn't have existed before generative AI.

Note: These aren't far-future concepts. The building blocks — 3D rendering, generative AI, real-time personalization — exist today. The challenge is combining them into cohesive experiences that feel magical rather than gimmicky.

Start experimenting with new agentic shopping experiences now

Most enterprises, including online retailers, are early on their journeys in adopting agentic technology. This is a technology that is unlike anything that has gone before, so there is a lot of learning to do. There are also many other applications of this technology outside of digital customer experiences. Given that, it is not a surprise that most enterprises are piloting agentic applications internally (for employees) rather than externally (for customers), with a focus of driving employee productivity.

At the same time, the threat from OpenAI does not feel immediate. ChatGPT isn’t the default homescreen, yet. And the risks of delivering a bad experience (because a customer facing agent did something unpredictable, not aligned with the retailer’s brand) are high. Given all of that, doesn’t it make sense to look at this sometime later, maybe next year?

That is a mistake. There has never been a technology that has been adopted as fast or as widely as ChatGPT: an Evercore survey reported that 8% of U.S. consumers now use ChatGPT as their primary search engine — up from 1% just one year ago.Learning by doing takes time. Companies need to start now. The most forward thinking online retailers already are.

Invest in other competitive moats that OpenAI cannot compete with you on – think brand and community

Outside of digital experiences, retailers have lots of opportunities to invest and press home their advantage over OpenAI, including both brand and community.

Brand: Consider Patagonia. When customers buy from them, they're not just buying a jacket — they're supporting environmental activism and a company that told consumers "Don't Buy This Jacket" in the name of sustainability. This brand ethos creates emotional connections that transcend product features. ChatGPT can recommend outdoor gear, but it can't replicate the feeling of being part of Patagonia's mission.

Community: Sephora's Beauty Insider Community has over 3 million active members who share tips, post reviews, and answer each other's questions. When someone asks "What foundation works for combination skin in humid weather?", they get answers from real people with similar skin types who've tested products in similar conditions. This peer-to-peer validation creates trust and authenticity that AI cannot match.

As a major, horizontal, US-headquartered technology company that serves so many users and use cases around the world, there are real limits to how effectively OpenAI can leverage their brand in your niche. And while OpenAI has developer communities, they cannot build the intimate, interest-specific communities that retailers cultivate around their products and values.

Do not use OpenAI models. And if you do, definitely do not use their customer memory

OpenAI offers leading frontier models, so it is tempting to adopt and fund their technology in spite of the competitive threat OpenAI poses to online retailers. This is especially true as they build out a better ecosystem of tooling around their APIs, and so many “hello world” agentic application examples are built on OpenAI.

That is not necessary. We are fortunate to live in a world where there are many other great models to choose from: including Anthropic, Google, Meta, DeepSeek and others. 

More than their models, though, OpenAI’s real moat is its customer memory. ChatGPT sees much more consumer adoption than any comparable AI service, and the data and understanding that OpenAI is able to build on those users is unprecedented. On April 10th, Sam Altman announced that ChatGPT’s memory can “reference all your past conversations”. That is already one heck of a customer data set, and one that will grow exponentially as usage of ChatGPT continues to skyrocket.

Today, that memory is used by OpenAI to drive an improved consumer experience in ChatGPT – enabling it to provide tailored, personalized answers to the consumer’s questions. But it is likely that at some point OpenAI will offer enterprises that use its frontier models for consumer facing applications to “enable” that memory when they leverage those models: this will be compelling because then those applications will benefit from the customer understanding that OpenAI has been able to develop. That would be a huge mistake for online retailers. This is because it would give OpenAI the ability to add new memories based on the retailer’s digital experience for those same customers to its own customer memory bank, further deepening their competitive moat whilst eroding one of the retailer’s critical moats – their own customer data and associated customer knowledge. 

So avoid using OpenAI models and memory in your consumer-facing applications because it is a bad idea to fund the business that’s coming for your business, but more importantly to prevent you eroding one of your competitive moats: your customer intelligence.

Partner with companies that can help you achieve these goals faster

Building agentic shopping experiences is complex — but you don't have to do it alone. A broad ecosystem of companies has emerged to help retailers develop and launch consumer-facing agentic applications faster, bringing specialized expertise and proven tools to accelerate your journey.

Why partnering matters: The technology landscape is evolving rapidly, and the skills required — from prompt engineering to real-time personalization — are scarce. Smart partnerships can help you move from concept to production in months rather than years, while avoiding common pitfalls.

Key categories of partners include:

AI Development Platforms: Companies like LangChain, Mastra, Akka, and Boundary provide frameworks specifically designed for building agentic applications. These tools handle the complex orchestration between language models, your data, and external systems — letting your team focus on creating unique experiences rather than infrastructure.

Cloud Providers: The major clouds — AWS, Google Cloud Platform, and Microsoft Azure — offer comprehensive AI services beyond compute. They provide managed language models (like Bedrock, Vertex AI, and Azure OpenAI Service), vector databases for semantic search, and specialized tools for building conversational experiences. Importantly, they offer access to alternatives to OpenAI's models while providing enterprise-grade security and compliance.

Data Platforms: Your customer data is your competitive advantage, but only if agents can access it in real-time. Platforms like Confluent, Databricks and Snowflake have evolved beyond traditional analytics to offer AI-ready infrastructure. They can help you build the semantic layers, streaming aggregations and real-time feature stores that power truly personalized agent experiences.

Systems Integrators and Consultancies: Companies like Accenture, Deloitte, and specialized AI consultancies bring together strategy, technology, and implementation expertise. They can help you navigate the build-vs-buy decisions, integrate agentic experiences with your existing tech stack, and manage the organizational change required for success.

Specialized Retail Tech Vendors: A new category of vendors is emerging specifically focused on agentic commerce. These companies understand retail's unique requirements — from inventory integration to visual search — and can provide pre-built components that accelerate development.

Snowplow! Last but not least – at Snowplow, we are working with leading online retailers to help prototype and launch agentic shopping experiences. Snowplow’s low-latency digital shopper behavior is a great building block for not just driving these agentic experiences, but also for building customer memory that is owned by you, the retailer – not by OpenAI! 

The key is choosing partners who enhance your capabilities without creating new lock-in. Look for those vendors who understand that your goal isn't just to implement AI, but also to defend your customer relationships against the threat of disintermediation.

Whilst you invest in building agentic shopping experiences, also invest in Generative Engine Optimization (GEO) 

Generative Engine Optimization (GEO) is the process of optimizing your digital content to maximize the chance that it appears in answers provided by ChatGPT and other AI-driven platforms (e.g. Claude, Gemini, Perplexity etc.) This is important because the rise of ChatGPT in retail is inevitable, and it is important you compete effectively with the demand that is currently moving over to ChatGPT, whilst you invest in building the compelling digital proposition to compete with it.

Successfully implementing GEO for an online retailer is simple but not easy. It involves structuring your product data, creating AI-friendly content, and cultivating third-party mentions. 

We have had a lot of success implementing GEO at Snowplow. We are now working on a dedicated post for retailers on implementing GEO – watch this space!  

Here's what retailers should do immediately:

1. Structure your product data for AI consumption

• Add comprehensive schema markup (Product, Review, Offer, FAQ)
• Include detailed product attributes beyond basic specs
• Write clear, conversational product descriptions that answer "why" not just "what"
• Add comparison tables and pros/cons lists — AI loves structured comparisons

2. Create content AI engines trust

• Develop detailed FAQ sections for each product category
• Write "How to Choose" guides that address common customer questions
• Include statistics and specific numbers (dimensions, performance metrics, usage data)
• Add expert quotes and citations from authoritative sources

3. Optimize for third-party mentions Research shows over 90% of AI-generated shopping content comes from third-party sources. To capitalize on this:

• Actively participate in relevant Reddit communities (without being promotional)
• Encourage genuine customer reviews on multiple platforms
• Partner with credible bloggers and industry publications
• Create newsworthy content that earns media coverage

4. Keep everything fresh

• Update product information weekly — AI favors recent content
• Add "last updated" timestamps to all pages
• Regularly refresh seasonal content and buying guides
• Respond quickly to customer questions on public forums

The clock is ticking

OpenAI's shopping features are live today. Every week you wait is another week your customers build habits around shopping with ChatGPT. The merchants who win won't be those who perfect their agent strategy — they'll be those who start building now.

The path forward requires parallel efforts. While you're implementing GEO tactics to compete for today's ChatGPT traffic,  you must simultaneously build the agentic experiences that will define tomorrow's commerce.

This means reimagining the entire customer journey through the lens of what's now possible. Instead of forcing customers through static interfaces designed for the average shopper, you can create experiences that reshape themselves around each individual's context, intent, and history. The agents we have described don't just answer questions — they anticipate needs, understand nuance, and transform complexity into simplicity.

When a customer's problem becomes the starting point rather than a product category, when interfaces adapt in real time rather than remaining fixed, when years of relationship data inform every interaction — that's when you create shopping experiences that no horizontal platform can replicate.

Your 5-10 years of shopper behavioral intelligence gives you a head start, but that advantage shrinks daily. Every transaction that flows through ChatGPT adds to OpenAI's understanding. This is an arms race. The question isn't whether to build agentic experiences, but how quickly you can transform your customer understanding into experiences that feel magical.

Most importantly, resist the siren call of OpenAI's "easy" solutions. Their models and memory might seem like shortcuts, but they're paths to dependency and then lock-in. Choose partners who enhance your capabilities without competing for your customers.

The retailers who thrive in the AI era won't be those who simply optimize for other peoples’ chatbots — it will be the retailers who give customers a compelling reason to still come to them. That future is yours to build, but only if you start today.

Looking to build consumer-facing agentic applications?

At Snowplow, we are partnering with retailers to help them build agents that can better understand their customers – because they can perceive their digital behavior in real time. If you’re interested in learning more, get in touch with our team today.

Note: 

1. This post was inspired by Brian Balfour's excellent post The Next Great Distribution Shift, which shows how OpenAI is following a well trodden playbook already run by Facebook, Apple and Google.  
   

‍