ChatGPT Virtual Try-On Takes on Google’s AI Shopping Lead
ChatGPT virtual try-on arrived on October 1 with two connected shopping features, despite Google already spending years building its own AI fitting room. OpenAI now lets shoppers generate pictures of themselves wearing selected clothes or accessories. They can also save products to a Favorites collection inside ChatGPT.
The update turns a shopping conversation into a more continuous decision process. A shopper can describe an outfit, browse matching products, preview an item, and preserve promising options without leaving the assistant. However, the generated picture remains a visualization, not evidence that a garment will fit.
That distinction defines the contest with Google. Google introduced personal-photo try-on during 2025 and has since expanded the experience across Search, Shopping, Images, Chrome, and experimental fashion products. OpenAI is entering later, but ChatGPT can connect product discovery, conversation, image generation, and saved choices within one interface.
ChatGPT Virtual Try-On Connects Discovery With Visualization
The important change is not image generation alone. OpenAI has connected visual experimentation to an existing shopping journey inside ChatGPT.
A new Try on button can appear on eligible clothing and accessory listings in ChatGPT shopping results. Selecting it lets a user take or upload a selfie, then request a generated image showing the product on that person.
OpenAI says ChatGPT Images creates the result. A shopper can also upload a separate picture of an item, including a screenshot, and ask ChatGPT to visualize it on them. That path matters because inspiration does not always begin with a structured product listing.
A user might photograph an accessory in a store, capture an outfit from social media, or save an image from a retailer. ChatGPT can place that reference inside the same conversation used to discuss colors, styles, occasions, and alternatives.
The company announced the features on October 1, 2026. They are available through ChatGPT on mobile devices and the web, according to its release notes.
OpenAI’s documentation says a reference photo is saved for later try-ons. Users can replace or delete that image under Personalization settings. Reusing the same reference reduces the friction of testing several products during one shopping session.
Favorites provides the second part of the update. A bookmark control lets users save products into the ChatGPT Library, while folders can organize those products into smaller collections.
A shopper planning a trip could create separate folders for shoes, jackets, and accessories. Someone comparing interview outfits could preserve several combinations before choosing which retailers to visit.
The Library does more than store bookmarks. OpenAI has been developing it as a persistent area for uploaded documents, generated files, images, products, and places. The new shopping controls make product decisions another type of reusable material.
That persistence changes ChatGPT’s role. The assistant no longer needs to start every shopping discussion from an empty prompt or a forgotten product link.
A saved reference photo provides visual continuity. Favorites preserve the candidate set. Conversation supplies preferences and feedback that can refine the next request.
OpenAI’s shopping instructions also contain a crucial warning. Generated try-on images might not represent the customer or product exactly, and they do not guarantee size or fit.
Users still need to check a merchant’s measurements, product description, and return policy. The tool answers a style question more convincingly than a sizing question.
That limitation keeps the launch grounded. ChatGPT can help someone imagine an outfit, but it cannot reproduce the physical information available inside a real fitting room.
OpenAI Wants to Own More of the Shopping Decision
Virtual try-on and Favorites push ChatGPT deeper into the stages before checkout, where shoppers narrow choices and form preferences.
OpenAI has been assembling that path through several product releases. In November 2025, it introduced shopping research, a specialized experience that investigates products and prepares personalized buying guides.
Shopping research asks clarifying questions and searches for details such as availability, specifications, reviews, and images. Users can mark suggestions as uninteresting or request more products with similar characteristics.
In March 2026, OpenAI added richer product displays, side-by-side comparisons, visual browsing, and image-based product discovery. The company also expanded its Agentic Commerce Protocol, which allows merchants and commerce providers to send structured product information into ChatGPT.
The latest features fill two remaining gaps. Try-on adds personal visualization, while Favorites gives users somewhere to keep the products that survive an initial comparison.
Consider a shopper looking for an outfit for an October wedding. The person can specify the location, expected weather, dress code, colors, and clothing preferences in ordinary language.
ChatGPT can return product candidates, compare them, and accept feedback. The shopper can then preview a dress or jacket using a reference photo and save the strongest options into a wedding folder.
That sequence is more commercially valuable than answering an isolated product question. Each additional step keeps the user inside ChatGPT while the purchase moves closer.
OpenAI does not need every shopper to complete every step. It needs ChatGPT to become the place where a vague intention turns into a short list of products.
The company already tested a more direct commerce strategy. In September 2025, it announced Instant Checkout for eligible products, initially connecting ChatGPT with Etsy sellers and planned Shopify availability.
OpenAI worked with Stripe on the supporting commerce standard. The move positioned the assistant closer to a transaction platform, as an earlier commerce report explained.
Virtual try-on takes a different route. It focuses on the uncertain period before a purchase, when people compare options and ask whether a product suits them.
That stage offers more room for conversation than checkout does. It also plays to ChatGPT’s ability to interpret loosely expressed preferences and adjust recommendations through follow-up questions.
Favorites strengthens that strategy because shopping decisions often stretch across several sessions. A shopper might discover an item on a phone, reconsider it later on a laptop, then return after checking a retailer’s measurements.
The Library supplies continuity across that process. It also gives ChatGPT a structured record of what the user deliberately chose to preserve.
OpenAI says its product results are not advertisements and are not influenced by company partnerships. It also says ads remain separate from product recommendations.
That promise will receive more scrutiny as ChatGPT becomes more involved in commerce. Product ranking can shape purchasing decisions even when a recommendation is not labeled as an advertisement.
The harder issue is transparency. Shoppers need to understand why a product appeared, how current its details are, and whether comparable alternatives were omitted.
OpenAI says factors can include the prompt, conversation context, price, availability, reviews, and ease of use. Memory and custom instructions can also influence which items ChatGPT surfaces.
Those signals can improve relevance. They can also make two users receive meaningfully different product sets without seeing every reason behind those differences.
Google Still Has the Stronger Shopping Foundation
OpenAI is not introducing the first personal-photo fitting room. It is challenging Google with a more conversational route into an established category.
Google began offering generative virtual try-on for apparel in 2023, initially showing garments on a selection of real models. That approach was designed to represent different body shapes, skin tones, sizes, and poses.
In May 2025, Google added the ability to upload a personal photo. Shoppers could select eligible garments and generate a visualization using their own image rather than a predefined model.
Google’s try-on guide instructs users to provide a well-lit, full-body image with fitted clothing. It also allows them to save or share a generated look.
The capability now reaches a large catalog across Search, Shopping, and Images in supported markets. Google says its system can work with billions of garments, although eligible categories and regional requirements still apply.
Google also benefits from its Shopping Graph, a large collection of product listings, sellers, brands, reviews, and inventory information. That infrastructure gives it direct experience organizing rapidly changing retail data.
ChatGPT approaches the same problem from the opposite direction. Google begins with search results and adds a personalized image layer. OpenAI begins with conversation and brings product results into that dialogue.
The contrast shapes each company’s advantage.
Google can connect try-on to broad product coverage and familiar shopping searches. It can place the feature near users who already show strong commercial intent by searching for a specific dress, jacket, or pair of shoes.
OpenAI can retain qualitative context more naturally. A ChatGPT conversation might include the event, weather, existing wardrobe, preferred silhouettes, disliked colors, budget constraints, and feedback on earlier suggestions.
That contextual record helps when the customer has not yet decided what product to search for. The assistant can participate while the user converts an uncertain goal into concrete criteria.
Google has not ignored that territory. Its AI Mode and other conversational search experiences can interpret complex questions, while products such as Doppl have explored AI-generated personal styling.
Doppl lets users upload outfit images and preview them on an animated digital version of themselves. Google has also introduced Wardrobe features that can organize clothing detected in a user’s photo library.
These experiments show why OpenAI faces a difficult opponent. Google can connect product search, personal photos, browser activity, image discovery, advertising, merchant feeds, and fashion experimentation across several established surfaces.
OpenAI’s advantage is interface concentration. A user does not need to understand which specialized Google product handles each stage. ChatGPT presents the experience as another capability inside a familiar conversation.
That simplicity carries its own risk. Combining discovery, recommendation, visualization, memory, and saved items can make the assistant’s output feel more authoritative than the underlying evidence supports.
A polished picture might persuade a shopper emotionally, even when garment measurements or generated details are inaccurate. The conversational presentation can further blur the line between an informed recommendation and a plausible suggestion.
This is why the central contest is not merely whose images look better. It is whether ChatGPT can make a multi-step shopping process feel coherent without encouraging misplaced confidence.
Google’s existing lead does not settle that question. Search dominance provides distribution, but shopping behavior can shift when a new interface reduces repeated work.
OpenAI must prove that the convenience survives real product catalogs, imperfect images, changing inventory, and the complicated reality of apparel sizing.
How ChatGPT Try-On Works, and Where It Stops
ChatGPT try-on generates a visual interpretation of a product and person. It does not simulate physical fit or verify the merchandise.
A virtual try-on system combines information from a person’s image with the visual characteristics of a garment or accessory. The model then generates a new picture that attempts to preserve both identities.
That task is harder than pasting one image onto another. Clothing folds around a body, responds to posture, covers existing garments, and changes shape across different materials.
Patterns, logos, fasteners, straps, and fabric textures can also shift during image generation. A visually convincing result might still alter details that matter to the buyer.
OpenAI does not publish a detailed technical breakdown for this shopping feature. Its support material states that ChatGPT Images produces the generated try-on.
The workflow accepts either an eligible product listing or an uploaded product image. It also uses a selfie or another saved reference photo supplied by the user.
This mechanism is useful for evaluating broad appearance. Color combinations, silhouette, overall styling, and compatibility with visible features can become easier to imagine.
However, the result cannot reveal how fabric feels, whether a seam rubs, or how a shoe supports the foot. It cannot guarantee that a labeled size matches the shopper’s measurements.
It also cannot show exactly how the manufactured item will drape under real lighting. Even a faithful product photograph captures only one sample, angle, and set of conditions.
Researchers working on virtual try-on systems have repeatedly treated size control as a separate challenge from appearance generation. A model can create an attractive image without having enough physical data to predict fit.
OpenAI states this boundary directly. Its documentation tells users that the result does not guarantee size or fit and advises them to consult merchant information.
That language should guide how shoppers use the tool. ChatGPT virtual try-on is closer to a visual mood board than a remote tailoring service.
The feature could still reduce uncertainty. A shopper deciding between two colors might gain useful direction without expecting centimeter-level accuracy.
It could also help users explore styles that they would not normally test in a store. Generating several looks carries less social or logistical friction than collecting garments inside a fitting room.
The product-image upload path widens that experimentation. It can work with inspiration found outside ChatGPT, provided the image is suitable and the request complies with the platform’s rules.
Still, screenshots introduce another reliability problem. The model might not know the exact product, fabric, cut, or available variations represented by an image.
A conversational request can clarify some of that context, but it cannot recover facts absent from the source material. Shoppers should return to the retailer’s listing before treating a generated look as a purchase reference.
The generated result can also affect expectations about body appearance. Image models might smooth textures, shift proportions, alter facial features, or change the garment to produce a more coherent composition.
OpenAI acknowledges that neither the product nor the user may be represented exactly. That warning deserves more attention than the realism of the finished picture.
The best outcome is decision support with visible limitations. The worst outcome is a persuasive but inaccurate image that overrides better evidence from measurements and return conditions.
Saved Photos Create a Privacy Test
The convenience of repeated try-ons depends on retaining a personal image, making data controls part of the shopping experience.
OpenAI says a reference photo remains saved for future virtual try-ons. The user can manage it through Settings, Personalization, and Reference photos.
That design removes repeated uploads and makes experimentation faster. It also means the feature is not simply processing a temporary image and immediately discarding it.
A selfie or full-body photograph can contain sensitive contextual information. It may reveal a face, body shape, approximate age, home interior, location clues, or other people in the background.
Users should therefore choose reference images deliberately. A plain background and a photo created specifically for try-on can reduce unnecessary information.
OpenAI’s shopping page explains how to replace or delete the saved reference. However, that page does not provide the same detailed try-on-specific assurances that Google publishes about biometric data and model training.
Google’s photo policy states that its try-on experience does not collect biometric data. It also says uploaded images are not used for training and are not shared with third parties.
That does not automatically make one service safer than the other. It does give users a more explicit set of claims to compare.
OpenAI users should review the data controls attached to their account and organization. Consumer, business, education, and enterprise environments can have different data-handling arrangements.
People should also avoid uploading photographs of someone else without permission. A realistic try-on result can feel harmless, but the source photo still belongs to an identifiable person.
Retailers face a related challenge. Generated images might misrepresent a garment’s details, creating expectations the actual product cannot meet.
A distorted logo, length, neckline, or material could affect the shopper’s decision. The consumer might associate that error with the merchant even when the retailer did not generate the image.
OpenAI’s product-ranking claims will also require observation. The company says shopping results are independent and not influenced by commercial partnerships.
Yet product discovery systems depend on merchant feeds, structured metadata, third-party content, and availability signals. Those inputs are not equally complete across every retailer.
A merchant with cleaner data can become easier for an automated system to understand. That does not mean the system intentionally favors the business, but it can still affect visibility.
The same issue already exists in search engines and marketplaces. Chat-based shopping makes the selection less visible because users may see a concise recommendation rather than pages of alternatives.
Favorites can deepen that effect. Once a shopper saves several products, future conversations could increasingly revolve around the preserved set instead of the wider market.
That persistence is convenient, but it can narrow exploration. Users should periodically ask for alternatives and verify whether saved products remain available at the stated specifications.
OpenAI must make deletion, correction, and recommendation controls easy to find. A feature built around personal images and accumulated preferences depends on user confidence.
Accuracy also affects trust. If repeated try-ons alter body features or garment details inconsistently, the saved reference stops feeling like a stable representation.
The company’s own limitation notice sets an appropriately cautious standard. OpenAI should preserve that warning wherever generated images appear, not only inside support documentation.
What Will Decide the AI Shopping Contest
The next test is whether people repeatedly use ChatGPT for shopping, not whether they generate one entertaining try-on image.
The first signal is sustained use across several products and sessions. Favorites should reveal whether shoppers return to earlier decisions instead of treating virtual try-on as a novelty.
OpenAI has not disclosed adoption figures for the new features. Evidence of repeat use, folder creation, or movement from saved products to merchant pages would strengthen its integrated-shopping strategy.
A brief burst of image generation would suggest something else. It would show interest in the visual effect without proving that ChatGPT influences serious purchase decisions.
The second signal is product and image accuracy. OpenAI needs generated clothing details to remain recognizable while preserving the user’s appearance consistently.
Errors will matter most when they change a product’s defining characteristics. A shifted hem, missing pocket, softened pattern, or altered accessory can turn visualization into misinformation.
The system also needs current product data. A compelling recommendation loses value if the color, size, seller, or availability has changed before the user reaches checkout.
OpenAI’s expanded commerce protocol is meant to improve catalog representation. Merchant participation and data freshness will determine whether that technical foundation matches Google’s shopping infrastructure.
The third signal is Google’s response. Google can expand virtual try-on into more surfaces, connect Wardrobe more closely with Shopping, or increase conversational guidance around products.
Google could also use its browser position to bring try-on directly onto retailer pages. Its existing Chrome implementation already points toward a shopping layer that follows users across the web.
A faster Google rollout would weaken OpenAI’s interface advantage. A fragmented or confusing response would leave room for ChatGPT to become the simpler starting point.
Retailers should watch both companies instead of assuming one will control the journey. Product feeds, accurate imagery, complete sizing information, and consistent identifiers help any AI system represent merchandise correctly.
Shoppers should apply a simpler test. Use ChatGPT virtual try-on to compare styles, not to replace measurements, material details, or return policies.
Try the same item with more than one reference image if the result looks inconsistent. Check whether distinctive product details survive generation before relying on the visualization.
Review saved photos under Personalization and remove images that are no longer needed. Revisit Favorites before buying because product information and availability can change.
Most importantly, compare the generated picture with the retailer’s original images. The assistant’s version is an interpretation, even when it appears photographic.
OpenAI has created a coherent path from a vague clothing idea to a saved product shortlist. Google still holds the stronger foundation in product search and virtual try-on history.
The outcome will depend on behavior rather than launch-day novelty. Will shoppers trust one conversation to guide discovery, visualization, comparison, and recall, or return to specialized shopping tools?
For now, treat ChatGPT virtual try-on as a useful visual filter. Test it against real product details, keep its saved-photo controls in view, and watch whether OpenAI improves consistency over the coming months.
