On October 1, OpenAI began rolling out a Try on feature inside ChatGPT shopping, letting users upload a selfie and see how clothing and accessories look on them before buying, alongside saved favorites. Retail Dive and TechCrunch both reported the launch as a step deeper into the pre-purchase path. The implication for apparel brands is direct: fit confidence is now being generated inside a conversational surface the brand does not control, and the quality of product data feeding that surface will determine which items get recommended and which get returned.
Key Conclusions
The core conclusion is that virtual try-on shifts the apparel decision from the product page to a conversational interface, which changes what data quality means. A try-on result is only as good as the garment imagery, size data and fit attributes behind it. Brands with clean, structured product attributes will see higher conversion and lower returns from AI-mediated traffic, while those with inconsistent sizing data will absorb the cost of mismatched expectations at scale.
The scale of the shift is easier to see against holiday numbers. Adobe projects $275.1 billion in U.S. online holiday sales with AI-driven traffic to retail sites up 130 percent year over year (Adobe), and ICSC's holiday survey found 88 percent of consumers intend to spend the same or more. When AI channels grow that fast, product data quality stops being a back-office concern and becomes a traffic-allocation variable that merchandising teams must manage directly.
How Try-On Entered the Chat Flow
The feature arrived as an incremental capability rather than a standalone app, which matters for how brands should respond. OpenAI added a Try on button to eligible apparel and accessory listings and paired it with saved favorites, keeping the interaction inside the same assistant where shoppers already ask comparison questions and check availability.
From product page to conversation
Historically, try-on lived on the retailer's own site or in a dedicated app, where the brand controlled imagery, sizing guidance and upsell placement. Inside ChatGPT, the same decision happens in a surface the brand does not design. That removes some merchandising control but also removes friction, because shoppers no longer have to switch context to evaluate fit before deciding.
What the assistant needs to work well
For try-on to produce a useful result, the assistant needs consistent garment imagery, standardized size charts and clear fit descriptors such as cut, rise and stretch. Where these attributes are missing or contradictory, the generated image is plausible but misleading, which converts a conversion tool into a returns driver. Data hygiene is therefore the precondition rather than a follow-up task.
Best Practices
Three practices help brands get value from conversational try-on without inheriting its risks. They focus on making product data machine-readable, measuring outcomes by surface, and closing the loop between returns and product attributes so that each season starts from evidence.
Standardize fit attributes first
Normalize fit language across the catalog before optimizing for any AI surface. If one product line describes a garment as relaxed and another calls a similar cut oversized, an assistant cannot reconcile them, and recommendations become inconsistent. A controlled vocabulary for fit, material and sizing is a modest investment with an outsized effect on AI-mediated conversion and on customer trust.
Measure conversion and returns by surface
Track conversion and return rate separately for conversational surfaces versus owned channels. Blended metrics hide the effect, because AI traffic tends to arrive with higher intent but less brand context. Segmenting by surface reveals whether try-on is genuinely improving fit confidence or simply moving returns from one channel to another without changing the net cost.
Common Mistakes
The first mistake is treating try-on as a marketing novelty and staffing it with campaign budget rather than data investment. The feature's performance depends on catalog quality, which is an operations problem. The second mistake is measuring only conversion. If returns rise at the same rate, the net effect on margin can be negative, and the brand will have paid for a feature that shifted cost rather than removed it.
A third mistake is ignoring the assistant's limitations on non-visual attributes. Fabric weight, stretch recovery and care requirements cannot be inferred from an image, so shoppers still need structured text alongside the render. A fourth mistake is assuming all categories behave alike. Try-on helps most where fit is the primary uncertainty, and much less where performance or durability drives the purchase, so investment should follow category-level evidence.
What It Means for Returns and Reviews
Returns are where conversational try-on will be judged. Apparel returns are driven mainly by fit and expectation gaps, and both are addressable with better attributes. If a brand can show that try-on traffic returns at a lower rate than non-try-on traffic for the same item, it has a defensible case for shifting catalog investment toward attribute quality rather than commissioning more imagery.
Reviews will change too. When shoppers arrive already having seen a rendered image of themselves in the garment, review content shifts from basic fit questions toward durability and washing experience. Brands that monitor review themes can detect this shift early and adjust both the product page and the try-on inputs accordingly, closing a loop that most catalogs currently leave open for months.
There is also a competitive dimension. Retail Dive reported that Walmart is expanding one-hour express pickup and adding in-store sign language interpretation, and that roughly half of shoppers worry about affording gifts. Convenience and value remain decisive, so try-on will not rescue a poorly priced or poorly stocked item. It improves the odds at the margin, and margins are where apparel economics are decided.
Summary
Virtual try-on inside ChatGPT moves fit evaluation into a conversational surface and makes product data quality a growth variable. Brands should standardize fit attributes, measure conversion and returns by surface, and route investment toward categories where fit drives the decision. The feature does not replace pricing, availability or convenience; it sharpens the outcome of everything else the brand already does.
Data Sources
1. Retail Dive: Virtual try on launches on ChatGPT (retaildive.com)
2. TechCrunch: ChatGPT can now virtually try on clothes for you (techcrunch.com)
3. Retail Dive: Walmart debuts 1-hour express pickup, ASL service (retaildive.com)
4. Adobe: U.S. Holiday Shopping Season to Hit Record $275.1 Billion Online (news.adobe.com)
FAQ
Q: Why does try-on inside an assistant change anything for apparel brands?
A: Because fit confidence is generated in a surface the brand does not design, so the assistant's recommendation depends on the structured product data the brand supplies rather than on page design or merchandising layout.
Q: Which product attributes matter most for try-on quality?
A: Consistent garment imagery, standardized size charts and controlled fit descriptors such as cut, rise and stretch. Missing or contradictory attributes produce plausible but misleading renders.
Q: How should success be measured?
A: Segment conversion and return rate by surface. If conversion rises while returns stay flat for the same item, the feature is adding value rather than shifting cost between channels.
Q: Is try-on equally useful across categories?
A: No. It helps most where fit is the main uncertainty, such as tailored apparel, and less where performance, durability or specification drive the purchase decision.
Q: Can a small brand start without new systems?
A: Yes. Begin by normalizing size charts and fit language across the existing catalog, then measure conversion and returns by traffic source before investing in new tooling.
Q: Does try-on reduce returns automatically?
A: Only if expectations match reality. Without accurate attributes, a rendered image can raise expectations and increase returns, which is why data quality must come first.
References
1. Virtual try on launches on ChatGPT
2. ChatGPT can now virtually try on clothes for you
3. Nearly half of shoppers are worried about affording holiday gifts
