Ask Pinterest App Tests Agentic Visual Search in US Sandbox

On June 17, 2026, Pinterest introduced Ask Pinterest, a standalone web application that replaces traditional keyword discovery with a conversational interface. The release serves as a US-only sandbox for agentic shopping features, rolling out alongside a suite of enterprise-focused AI tools for advertisers. The initiatives follow a recent $4 billion cloud deal with AWS designed to accelerate model training and AI inference using Amazon’s custom silicon.

Ask Pinterest Application

The experimental app shifts discovery to natural language processing for complex, multi-step queries. Users can prompt the system with extensive constraints, such as planning a dinner party on a budget or furnishing a room over several months.

The application engine relies on Pinterest’s proprietary Taste Graph, which maps user interests, aesthetic preferences, and intent. When authenticated, the app personalizes its outputs against the specific Pins and Boards a user has saved. Pinterest categorizes the experience as agentic, meaning the application maintains context across multiple sessions to facilitate long-term shopping journeys rather than isolated, transactional searches.

Enterprise MCP and Ad Workflows

Alongside the consumer app, Pinterest released integration infrastructure to pipe its discovery data into professional marketing workflows. The Pinterest Model Context Protocol (MCP) provides an AI-native infrastructure layer for advertisers. This enables brands to route Pinterest campaign and analytics data directly into the third-party copilot applications they already use. The alpha rollout includes partnerships with Dentsu, Havas, Pacvue, and Omnicom’s Jump450. Developers can review the underlying MCP architecture to understand how these programmatic data connections function across different agent environments.

The platform also introduced an internal Business Assistant integrated into the Pinterest Ads Manager. The tool processes visual data, including trend graphs and top-performing Pin examples, to help marketers optimize active campaigns.

For automated campaign execution, the new Performance+ Creative Model evaluates multiple ad variants concurrently. The model predicts and selects the specific variant most likely to perform for an individual impression. Initial testing data shows a 7.5% increase in click volume compared to previous singular variant models.

Strategic Shift

The move positions Pinterest against visual search and shopping initiatives from Google, OpenAI, Meta, and Shopify. Chief Business Officer Lee Brown noted the platform intends to leverage context, taste, and trusted recommendations as the primary drivers of product discovery.

For developers working in the e-commerce space, the release signals a clear structural shift away from index-based product matching. If you build discovery systems, you must now account for session persistence and personalized aesthetic mapping alongside traditional keyword retrieval.