When Chatbots Sell: How OpenAI, Perplexity and Startups Are Racing for Holiday Shopping

A shopper asks ChatGPT for a gaming laptop under $1,000 with a 15‑inch‑plus screen and checks out without leaving the chat. Two days later, a competing assistant promises recommendations tuned to what it already knows about the user. This season, conversational AIs are trying to replace search results with purchases.

The stakes are immediate: OpenAI and Perplexity announced AI shopping features in late November 2025 that push discovery, comparison, and checkout into conversational interfaces, leveraging partnerships with platforms like Shopify and PayPal to complete purchases in-chat. Adobe projects AI-assisted online shopping could grow by 520% this season, a figure startups and retailers are treating as both opportunity and threat.

Checkout in the chat: the giants buy instant access

For specialist startups that have spent years curating product catalogs and building vertical datasets, the arrival of general-purpose assistants raises a familiar question: compete on breadth and distribution, or double down on depth. The answer will hinge on three levers-data quality, checkout integration, and platform distribution-each of which is changing faster than many founders expected.

OpenAI and Perplexity are not merely adding shopping prompts; they are integrating checkout. OpenAI can let users find a laptop and complete a purchase within ChatGPT, while Perplexity touts memory-based, personalized recommendations and a PayPal partnership for in-chat payments. That changes unit economics: conversion rates, not clicks, will define success.

Vertical advantage: why deep data still wins

Partnerships matter. Shopify and PayPal integrations give those platforms a fast path to transactions and data on basket composition, margins, and returns. For retailers, a deal with a big assistant can drive immediate volume; for the assistant, it supplies telemetry to refine ranking and personalization models in production.

The commercial math is stark. Adobe’s 520% growth projection for AI-assisted shopping implies a multifold lift in sessions that convert inside a conversation rather than on product pages. For large LLM operators, that means the ability to monetize through affiliate fees, direct checkout take-rates, or promotional placements-each a lever smaller startups must counter with either superior conversion or niche loyalty.

Platform politics and distribution risk

Several founders argue the core secret is not dialogue design but the dataset beneath it. Zach Hudson, CEO of Onton, told TechCrunch that “any model or knowledge graph is only as good as its data sources,” and that vertical models tuned to fashion or furniture will outperform general-purpose tools. Onton has built a pipeline to catalog hundreds of thousands of interior-design SKUs, normalizing attributes that generic indexes miss.

Domain models capture nuance-fabric hand, silhouette, interchangeability across collections-that gnaws at recall and precision in commodity search. Julie Bornstein, CEO of Daydream, said fashion search requires merchandising logic that understands occasion, fit, and outfit-building. In plain terms: a generic LLM can point at similar images; a vertical model knows which dress sells with which shoe and why.

Technically, verticals exploit smaller, cleaner corpora to fine-tune or train retrieval-augmented generation pipelines with high-signal embeddings and curated knowledge graphs. That reduces hallucination in product recommendation and improves post-click conversion, metrics that matter when partners demand ROI for featured placement.

The operational edge: compute, latency and trust

Platform politics and distribution risk

Distribution is a second vulnerability. Microsoft’s decision to pull Copilot from WhatsApp after January 15, 2026 highlights how platform policy can erase a channel overnight. The change came after WhatsApp revised rules limiting general-purpose AI bots on its Business API, and Microsoft warned users that unauthenticated chat histories will not transfer when they migrate to Copilot’s apps.

For startups, that is a live risk: building on another company’s messaging surface, app store, or social graph creates single-point failures. Startups with direct merchant relationships and first-party checkout data can survive a messaging cull more easily than those that depend on free distribution through major assistants.

A practical fallout: expect more startups to negotiate first-party integrations and authenticated identity flows. Where an assistant can remember a user across sessions-Perplexity emphasizes memory as a differentiator-that persistence becomes a moat. But it also raises privacy and fairness trade-offs: memory improves personalization but concentrates behavioral data with a few players, which regulators and industry groups are watching.

• Microsoft's AI chatbot Copilot leaves WhatsApp on January 15 - TechCrunch, 2025-11-25

• Speechify adds voice typing and voice assistant to its Chrome extension - TechCrunch, 2025-11-25