Agentic Commerce Query Shopping Agents

Purpose

The signature skill of the agentic-commerce-skills library. Sends a single shopping query to 5 AI shopping agents in parallel (ChatGPT, Gemini, Claude, Mistral, DeepSeek) and returns each agent's ranked product recommendations.

This is what makes AI catalog visibility measurable. Without cross-agent ground truth, "AI visibility" is a vibe; with it, it is a metric you can A/B test fixes against (see [agentic-commerce-visibility-report](../agentic-commerce-visibility-report/SKILL.md)).

Together with the Agentic Commerce Benchmarks dataset, this skill lets you measure where your products rank vs. competitors across the 6 most important AI shopping agents.

Prerequisites

API keys for the agents you want to query (each is optional; missing keys = that agent skipped):

• ANTHROPIC_API_KEY — for Claude
• OPENAI_API_KEY — for ChatGPT
• GEMINI_API_KEY — for Gemini (free tier available, https://aistudio.google.com)
• MISTRAL_API_KEY — for Mistral Le Chat
• DEEPSEEK_API_KEY — for DeepSeek

Parameters

| Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | query | string | yes | — | Shopping query in natural language. Examples: "best winter jacket under $200 for hiking", "sustainable bamboo socks for sensitive skin" | | agents | array | no | all-configured | Subset of agents to query: claude, chatgpt, gemini, mistral, deepseek. Skip with missing API keys. | | vertical | string | no | — | Hint to the agents about category: apparel, beauty, home, electronics, fitness, food, pets, baby, outdoor, gifts | | priceceiling | float | no | — | Max price in USD (passed to agents in system prompt) | | format | string | no | human | Output format: human or json | | timeoutseconds | integer | no | 60 | Per-agent timeout |

Workflow Steps

1. Build the shared shopping prompt:

``` You are a shopping recommendation assistant. Given the query below, provide a numbered list of 5-10 specific product recommendations.

For each: brand + product name, approximate price, why you recommend it (1-2 sentences), and where to buy it (merchant/website).

Format each recommendation as:

1. Brand ProductName - $XX.XX

Reason for recommendation. Available at: https://merchant.com/product-link

Query: {query} {vertical and price ceiling hints if provided} ```

1. Call each configured agent in parallel with the shared prompt. Each agent runs in its own coroutine to keep total latency near the slowest agent (~10-15s, not 5×).

1. Parse responses uniformly: extract per-mention brand, product_name, price, merchant_domain. Store the raw response too (parser may need updating as agents shift output format).

1. Output ranked mentions per agent:

``json { "query": "best winter jacket under $200 for hiking", "captured_at": "2026-05-11T15:30:00Z", "agents": { "claude": { "ok": true, "latency_ms": 4521, "mentions": [ { "rank": 1, "brand": "Patagonia", "product_name": "Nano Puff Jacket", "price": 199, "merchant_domain": "patagonia.com" }, ... ] }, "chatgpt": { ... }, ... } } ``

1. If format: human, also render a side-by-side comparison table of mentions across agents (great for spotting that ChatGPT mentions you, but Gemini does not — actionable insight).

Safety

> Read-only. Sends data to external APIs (the 5 agent providers). Do not include PII in the query string. > > Costs: $0.003-0.01 per agent per query depending on model. A 100-query × 5-agent run costs roughly $1.50-5.

Companion skills

• [agentic-commerce-visibility-report](../agentic-commerce-visibility-report/SKILL.md) batches this skill across 100+ queries to compute a store's cross-agent visibility score.
• The Agentic Commerce Benchmarks dataset is captured continuously using this same pattern (2400 captures/day).

Reference

The agent prompt template comes from the reference ground-truth collector at services/ground-truth/runners/. The output parser is at agentic-commerce-tools.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: commerce-agentic
• Source: commerce-agentic/agentic-commerce-skills
• License: MIT

Install and usage instructions live in the source repository linked above.