#68: Holiday Shopping in the Age of AI
Let the model do the work for you
Author’s Note: Hope everyone had a wonderful Thanksgiving. I’m grateful that I have an audience to share my ideas with. Candidly, without you all, I’d just be talking to myself, and I’d rather talk to others.
Last week, OpenAI and Perplexity launched personal shopping assistants just in time for peak holiday gifting season. OpenAI introduced ChatGPT’s Shopping Research, and Perplexity announced Shop with Perplexity.
I highly recommend checking out these shopping assistants, as I find them helpful and a peek into the future of commerce. Primarily, ChatGPT’s Shopping Research can be found by clicking the “+” in the prompt bar and selecting the “Shopping Research” option. Let ChatGPT know what you’re looking for, with a prompt like, “I’m looking for the best running compression pants for cold weather in NYC this winter.” After pressing enter, you’re presented with a series of multiple-choice clarifying questions about budget, material, and features. Next, the Shopping Research assistant “thinks” for seconds to a few minutes and then provides a detailed side-by-side comparison of a few product options, as well as a final recommendation.
On Perplexity’s front, they offer a general shopping section (look for this on the sidebar) where you can browse products to buy, as well as a “shopping” option as part of a prompt. There’s no Q&A, but Perplexity does provide a detailed work-up (very much appreciated for my running compression pants inquiry).
As someone who strongly dislikes shopping, I’m impressed by what OpenAI and Perplexity created. One of the reasons I don’t like it is because of the cognitive load stemming from too many options. Ever hear of the Paradox of Choice concept? I’d rather someone who knows my preferences and has subject matter expertise provide a few options to select from.
Both tech challengers claim that you can purchase within their interface, however, I have yet to stumble across a product I could buy on the AI platform. Yet, I believe it’s only a matter of time until shoppable products are widespread.
A few thoughts (in no specific order):
Foundational models* (like OpenAI) are disrupting commerce because the industry is so large. When you’ve raised many billions of dollars (Anthropic, xAI, OpenAI), you need to justify that investment through a mighty monetization plan. And a clear way to do that is to take a big chunk out of a gigantic business. I believe that foundational models will attempt to influence or own the commerce experience in the way of being a marketplace facilitator, as well as an advertising platform. If it may not be a purchase within the model, ChatGPT will tell you exactly which store you can buy it from and you can sure bet there will be an advertisement soon enough next to that suggestion.
*It is worth noting that Perplexity does operate its own LLM called Sonar, but it does not qualify as a foundational model as it was built on top of Meta’s LLaMA. For the more technical reader, Sonar is an LLM fine-tuned on Meta’s LLaMA model.
Here’s the thing: foundational models should be able to provide a more curated product selection or a tailored advertisement because they have your conversational data. Sure, Google search knows what you’ve searched in the past, and your cookies are scattered across the web (which help advertisers piece together where you’ve been to help influence where you’ll go). But chatting back and forth with AI allows the model to get a more accurate sense of who you are and what you may be looking for. In the case of my running pants prompt, ChatGPT knew that I tend to run warm (from a prior conversation about sweating) and suggested pants that wouldn’t be too constricting. It does sound a bit dystopian how AI will know our nuanced preferences, but hey, if this means I’ll reduce the amount of time I spend researching a purchase, I’m all for it. That said, this is assuming users opt to share conversational history, which is typically optional.
Don’t build a general-purpose AI agent. It must have a specific use case that isn’t disrupting every part of a trillion-dollar industry, as you’ll be competing with the likes of the foundational model companies. There’s plenty of opportunities to build on the application (vertical) layer, however you must possess an unfair advantage or operate in a niche enough market for the foundational model to avoid bothering with disruption. Remember, it’s likely not worth a foundational model’s time to allocate time and resources toward solving smaller problems.
Instead of building an AI shopping agent that can aggregate information and transact on your behalf, build an AI shopping agent that operates within a specific niche. Here’s an example: a beauty AI shopping agent that considers past skincare routines, facial skin analysis (see La Roche Posay’s image-based scan), and even a dermatology consult. After ingesting your personal data (which wouldn’t be available to a foundational model), the beauty AI shopping agent can curate a skincare routine and then purchase those products on your behalf (based on predefined spending guardrails). You can upload various pictures of your face over time, and the AI can assess whether the routine is working.
This scenario is too specific for OpenAI to spend time on. But it is still a massive market. If you don’t narrow your ideal customer profile (ICP) beyond “everyone”, good luck. You’re picking pennies in front of the steam roller that are foundational models. Now that pennies will no longer be minted, this metaphor may miss the mark. But you get the point.
Another thought. In addition to “Shop with Perplexity”, the tech firm launched a virtual try-on feature that allows users to see how clothing would look on them before buying. Perplexity allows users to create a digital twin and then dress their avatar in the jacket they’ve been eyeing. Note, this feature is not yet available to all users.
This feature seems like a signal that Perplexity is looking to position themselves more deeply within the commerce realm. They’ve had shopping infrastructure within their platform for most of 2025. Moreover, they’ve made a conscious push for advertising (yet paused due to limited traction) and also spun up a publisher program so legacy publishers and affiliates can get paid for their content.
This isn’t by any means the hottest take, but I believe that Perplexity will attempt to be the AI engine for commerce. Over the next two years, they’ll look to position themselves at the intersection of commerce and media as these two businesses go together when it comes to influencing and making a purchase. If you’re in the CPG or advertising industry, pay close attention to what Perplexity launches. If not, their updates are likely less relevant.