Virtual try-on inside an AI chatbot stopped being a demo and became a mainstream retail feature in 2026. L’Oréal announced on June 17, 2026 that Maybelline’s Makeup Virtual Try-On would launch directly inside ChatGPT, marking the first time a major beauty brand built real-time AR makeup simulation into a conversational AI used by more than 900 million people weekly. For every brand that isn’t L’Oréal, though, that specific route doesn’t exist: the deal came out of a bespoke commercial partnership with OpenAI, not a feature anyone can switch on.

The practical question this raises is narrower and more useful: which real tools actually let a brand put virtual try-on inside a chatbot it controls — its own website assistant, a WhatsApp bot, a custom GPT — rather than a chat surface owned by Google or OpenAI? Google has separately built its own version directly into Search, Shopping and the Gemini app, retiring its standalone Doppl app on April 30, 2026 in the process; retailers from Walmart to Macy’s now plug into that first-party surface rather than embedding it elsewhere. Neither of those paths, though, gives a brand a chatbot of its own with try-on inside it.

There is no single best answer to that question either. Google’s built-in try-on has unmatched reach but cannot be embedded anywhere else. TryOn API ranks second as the strongest building block for a brand that wants its own version of what L’Oréal built: a single OpenAI-compatible endpoint routing 14 try-on models that drops straight into an existing chatbot’s tool-calling layer. Perfect Corp and Revieve offer more turnkey, licensed conversational agents for beauty and skincare brands specifically, and Tolstoy is the fastest path for a Shopify apparel store that wants a working chatbot-plus-try-on experience without building one. Below we rank six ways to get virtual try-on into a conversation, and what each one trades off to get there.

Why “inside ChatGPT” and “inside your own chatbot” are different problems

L’Oréal’s Maybelline integration is a brand-level partnership: OpenAI built the feature into ChatGPT itself, with Maybelline’s ModiFace technology plugged in behind the scenes. That is not a capability a smaller brand can request or buy — there is no self-serve “add my products to ChatGPT’s try-on” product. What a developer can do is call a try-on model as a tool from inside a chatbot they already own, whether that bot runs on a custom GPT, a website widget, or a messaging platform like WhatsApp Business. That is the gap TryOn API and fal.ai fill from the infrastructure side, and the gap Perfect Corp, Revieve and Tolstoy fill with more finished, licensed conversational products. Google’s own version sits outside this framing entirely, since it isn’t something any brand embeds — it is a feature Google runs on its own product listings, reached through direct retailer partnerships rather than a public integration path.

What the Macy’s number does and doesn’t tell you

Macy’s reported 4.75x spending lift is the clearest public data point so far connecting a conversational shopping assistant to real revenue impact, and it is worth citing carefully. “Ask Macy’s” is a broader Gemini-powered shopping assistant that includes Google’s try-on capability alongside styling advice and product discovery, not a virtual-try-on-only feature, and Macy’s own team has attributed part of the lift to self-selection — shoppers who open an AI stylist are often already close to a purchase decision. Treat it as evidence that conversational commerce as a category converts well, not as proof that try-on specifically was the deciding factor. It is still the most concrete, named-source number in a category where most claims are directional.

Where the market is heading

The direction is toward every major AI chat surface treating commerce, and specifically visual commerce, as a native capability rather than a bolt-on. OpenAI’s arrangement with L’Oréal and Google’s consolidation of Doppl into Search and Gemini point the same way: the platforms that already have the largest conversational audiences are building try-on in-house rather than waiting for third parties to bring it to them. For everyone else — the vast majority of brands without an OpenAI or Google-scale partnership — the practical path stays what it has always been for underdogs in a platform-dominated category: use an API that routes across multiple underlying models, so a brand isn’t locked to a single provider’s roadmap or pricing, and build the experience into a channel it actually owns.