OpenAI Adds Shopping Features to ChatGPT, Improves Search, Citations and WhatsApp-Integration

OpenAI has announced it is starting to roll out an integrated shopping experience within ChatGPT, aiming to assist users with product research and purchasing decisions.

The new functionality, available to all users globally including those without an account, presents product suggestions featuring images, prices, ratings, and reviews, accompanied by buy buttons that direct users to merchant websites to finalize purchases.

OpenAI frames the update as part of broader improvements to its search function, which it noted has become highly popular.

Broader Search Enhancements and Access

The shopping feature launch coincides with other improvements to ChatGPT Search. The system now aims for better verifiability by providing multiple source links for information within its answers, using highlighting to show information origins.

OpenAI also added features like trending search suggestions and autocomplete to speed up queries. Extending access further, users can now also query ChatGPT Search via its WhatsApp integration by texting +1 (800) 242-8478.

These enhancements follow OpenAI making its basic web search function available without login required in February 2025 and expanding its advanced Deep Research feature to free users (with limits) on April 25th, using the more efficient o4-mini model.

Information for search and shopping is gathered via web crawls (using its OAI-SearchBot crawler) and through partnerships.

Monetization Questions and Market Dynamics

While emphasizing the current lack of paid placements in shopping results, OpenAI indicated it is exploring affiliate revenue models for the future.

OpenAI also stated it plans to “bring merchants along on our journey as we quickly learn and iterate,” suggesting potential for direct merchant integrations later.

This exploration occurs amidst considerable financial pressure – OpenAI reportedly faced a $5 billion loss in 2024 – and its ongoing transition to a for-profit structure, a move drawing criticism from some former staff concerned about safety. The company had previously acknowledged considering ads in December 2024 as a potential revenue stream.

The shopping launch intensifies competition with established services like Google Shopping and newer AI-focused tools from Perplexity AI (which launched its “Buy with Pro” feature in late 2024), Microsoft Copilot (with its own Merchant Program), Anthropic, and xAI.

OpenAI’s strategy of broadening access mirrors its April 25th decision to offer a lightweight version of its Deep Research tool to free users, potentially responding to market pressures after Google made its similar Gemini research feature free in March.

Personalization and Underlying Tech

The system aims for personalization by leveraging ChatGPT’s Memory feature, which allows the AI to recall user preferences and details from previous conversations to tailor results.

OpenAI began activating this cross-chat recall capability around April 11, following initial testing from February 2024. The specific integration of Memory with search and shopping functionalities is, however, still rolling out over the next few weeks for Plus and Pro subscribers and is initially unavailable in the EU, UK, and several other European nations.

While enabled by default where available, users can manage or disable Memory through settings or use a temporary chat mode.

Driving these new capabilities are OpenAI’s o3 and o4-mini models, introduced mid-April 2025. These models feature what OpenAI describes as “early agentic behavior,” allowing them to autonomously select tools like web browsing or image analysis needed to respond to a user’s request.

They are also natively multimodal, understanding text and images. The image generation itself stems from the `gpt-image-1` technology (based on GPT-4o) first added to ChatGPT in late March 2025 and opened via API on April 24th.

This allows the shopping feature to display relevant product visuals alongside recommendations.

Despite these advancements, the reliability of the new o-series models has faced scrutiny.

Reporting from April 19th detailed how both OpenAI’s own benchmarks and external analysis by Transluce AI found these models hallucinated or fabricated actions more often than predecessors on certain tests. This raises considerations about the accuracy of the product insights ChatGPT provides.

OpenAI acknowledged the higher fabrication rates in its model documentation, stating “more research is needed,” and telling TechCrunch, “Addressing hallucinations across all our models is an ongoing area of research.”