OpenAI on Thursday rolled out ChatGPT Images version 2.5, embedding virtual try-on and a product-save library directly into its chatbot for 1.2 billion weekly active users.

The release lets shoppers upload a selfie, then select "try on" inside shopping results or apply the feature to a style image spotted elsewhere online.

OpenAI framed the upgrade in technical terms: "Images 2.5 produces more natural lighting and richer textures, is better at preserving the subjects in your reference photos, and follows editing instructions more reliably across multiple turns," the company said.

What does this change for fashion sourcing and merchandising?

For brands, the operational question is no longer whether virtual try-on exists — it has existed on retail sites for years — but whether a free, model-agnostic tool inside the world's most-used chatbot can redirect merchandising and styling assets toward a non-owned channel.

Conversion math now tilts toward AI-referred traffic, per Adobe's retail research. "In May 2026, AI-referred retail visitors converted at a rate 54 percent higher than non-AI traffic," Adobe reported. That gap has fully flipped from a year earlier, when AI visitors converted at nearly half the rate of their non-AI counterparts.

The same dataset showed AI-referred visitors stay on site 53 percent longer than visitors arriving through other channels.

What should brand and sourcing teams verify before reallocating budget?

Three commercial variables now sit on the merchandising agenda:

• Which SKU imagery feeds into OpenAI's results
• Whether internal product metadata is structured for chatbot retrieval
• How return rates move when shoppers test fit remotely

A virtual try-on that boosts conversion but lifts returns erodes margin rather than expands it. Brand-side teams need to confirm whether the selfie-to-garment match-up is generated or anchored to asset packs supplied by the brand itself.

OpenAI's new favorites save function also raises questions about who owns the shopper relationship once a ChatGPT thread starts collecting looks.

What capacity and compliance questions follow?

Brands feeding imagery into OpenAI's pipeline will want written clarity on data residency, training opt-outs and the audit trail for any generated try-on image used in a marketing claim. None of those terms surface in OpenAI's consumer-facing release.

Version 2.5 also says nothing about integration with retail PIM systems, plug-and-play feeds from Shopify or Salesforce Commerce, or inventory-aware styling — features that would move the tool from novelty to daily merchandising infrastructure.

For sourcing teams, the immediate read-through is channel allocation. If a chatbot in front of 1.2 billion weekly users carries fit decision-making, the share of qualified traffic sent to retailer product detail pages, supplier imagery libraries and wholesale order portals is at risk of shifting.

Brand and vendor briefs from OpenAI's commercial partners over the next quarter will likely specify product-feed formats and image-asset standards tied to this rollout, determining whether the feature stays a marketing demo or becomes a working merchandising channel.