With the holiday shopping season approaching, both OpenAI and Perplexity unveiled new AI-driven shopping features this week. These tools plug directly into their existing chatbots to help users research potential purchases.

 

The two offerings are strikingly similar. OpenAI suggests that users might ask ChatGPT to find “a new laptop suitable for gaming under $1,000 with a screen over 15 inches,” or upload photos of a luxury garment to get recommendations for a less expensive alternative.

 

Perplexity, for its part, is emphasizing how its chatbot’s memory can tailor recommendations based on what it already knows about a user, such as their location or profession.

 

Adobe expects AI-assisted online shopping to grow by 520% this holiday season — a trend that could benefit niche AI shopping startups like Phia, Cherry, and Onton (formerly Deft). But with OpenAI and Perplexity leaning further into shopping experiences, does that put these smaller players at risk?

 

Zach Hudson, CEO of interior design shopping platform Onton, doesn’t think so. He argues that startups with deep domain expertise will continue to offer superior results compared to general-purpose chatbots.

 

“Any model or knowledge graph is only as good as its data sources,” Hudson told TechCrunch. “Right now, ChatGPT and LLM-based tools like Perplexity piggyback off existing search indexes like Bing or Google, making them only as good as the first few results returned.” (Perplexity clarified to TechCrunch that it operates its own search index.)

 

Daydream CEO and longtime e-commerce executive Julie Bornstein shares this view. She noted earlier this year that search has long been “the forgotten child” of the fashion industry — largely because it has never worked particularly well.

 

“Fashion is uniquely nuanced and emotional — finding a dress you love isn’t the same as finding a TV,” she said on Tuesday. “That level of understanding comes from domain-specific data and merchandising logic that grasps silhouettes, fabrics, occasions, and how people build outfits over time.”

 

AI shopping startups build their own datasets so they can train on higher-quality, richly structured information — something far easier to achieve when cataloging fashion or furniture than modeling all human knowledge.

 

For Onton, that meant building a data pipeline to cleanly catalog hundreds of thousands of interior design products, then training internal models on that refined dataset. Hudson believes that without this kind of specialization, smaller companies will inevitably fall behind.

 

“If you’re using only off-the-shelf LLMs and a conversational interface, it’s hard to see how a startup can compete with the larger companies,” he said.

 

OpenAI and Perplexity, meanwhile, benefit from massive existing customer bases — and the scale to forge immediate partnerships with major retailers. While tools like Daydream and Phia send shoppers to retailers’ websites (sometimes earning affiliate revenue), OpenAI and Perplexity can enable in-chat checkout through partners like Shopify and PayPal.

 

Both companies still face the challenge of finding sustainable business models, given the heavy compute costs behind their products. Following the lead of Google and Amazon, e-commerce could become a key revenue stream — with retailers potentially paying to promote their products within AI-generated results.

 

But that might simply worsen the problems users already have with search.

 

“Vertical models — whether in fashion, travel, or home goods — will outperform because they’re tuned to real consumer decision-making,” Bornstein said.