OpenAI has launched the "Shopping Research" feature on both the web and app versions of ChatGPT, available to Plus/Pro/Team users. The system real-time crawls e-commerce sites, forums, unboxing videos, and price tracking websites, automatically filters out fake reviews, generates structured shopping reports, and ensures no brand advertisements are involved throughout the process.

The shopping research feature is based on a specially optimized GPT-5 mini model, which improves the accuracy of shopping-related tasks through reinforcement learning. Detailed.jpg

Report Structure  

1. Top 3-5 Cost-Performance Rankings: Sorted by performance, reputation, and price with weighted scores  

2. Honest Pros and Cons Table: Lists the drawbacks of each product item (such as poor battery life, heavy, or poor customer service)  

3. Multi-Platform Price Comparison: Marks historical low prices and current good deals, and alerts about inventory and shipping time  

4. Personalized Optimal Solution: Provides a single recommendation based on user memory (budget, brand preferences, usage scenarios)

Core Process  

- User inputs needs → Model breaks down keywords → Parallelly searches over 30 sources → Outputs collapsible report cards within 10 minutes → One-click jump to official store or price comparison page

Data Coverage  

The initial release supports five categories: electronics, home appliances, beauty, home goods, and apparel; in the future, it will include return rates, repair rates, and carbon emissions data, with coverage expected for cars and second-hand houses by Q1 2026.

Test Feedback  

- Netizen "Choosing a laptop in 10 minutes" saved 800 yuan  

- On average, it saves users 3 days of price comparison and review watching  

- Reports support one-click sharing, and friends can see real-time updated prices when they open them

OpenAI stated that Shopping Research adopts a "zero-commercialization" strategy, does not accept brand paid rankings, and all revenue comes from subscriptions, ensuring a stance that only stands for users.