• MATCH
  • DRIFT
  • GAP
  • HUMAN

How to audit search and recommendation logic with your AI

Connect Constructor to your AI client to pull live site data. You can then ask your agent to check if results match your specific business rules.

Ask AI about this page

Short answer

How can I use AI to audit my search and recommendation logic?

Your AI pulls live product data and recommendation sets to compare against your business rules. It identifies when ranking logic drifts from your intended strategy. You'll see exactly where the results fall short.

Audit outcomes.

Where your findings land.

The AI categorizes its findings based on how well the live data matches your rules.

MATCH

Logic is working

The AI confirms that the products returned by the search or recommendation tools align with your constraints. No action is needed.

DRIFT

Ranking issues detected

The AI finds that the ML is surfacing products that violate your rules. You get a list of the specific items that shouldn't be there.

GAP

Missing product coverage

The AI identifies when relevant products are being buried or ignored by the current ranking logic. This shows you where to tune your models.

HUMAN

Manual review required

The AI flags edge cases where the logic is technically correct but feels wrong for the brand. You decide if these results need a rule change.

The workflow

What your AI does when the audit starts.

The AI acts as an automated auditor, pulling live data and comparing it to your requirements.

  1. Pull search results

    The AI fetches the current product list for a specific query to see what's ranking highest.

    search_products

  2. Check recommendations

    The AI pulls the specific product pods shown to users to verify the items match the intended intent.

    get_recommendations

  3. Verify sort order

    The AI checks if the sorted product lists follow your specific business or popularity rules.

    search_sorted

  4. Compare and flag

    The AI compares the retrieved data against your written business rules and highlights any discrepancies.

    search_productsget_recommendations

Try it

Copy these to start.

Use these prompts to kick off an audit in your AI client.

Starting points

Replace the bracketed text with your specific query or product category.

Constructor Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_7QdndEEXMElkrXV72k1k9X9VY0HS8rfUr6VIhrH5/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Constructor capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "constructor-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_7QdndEEXMElkrXV72k1k9X9VY0HS8rfUr6VIhrH5/mcp"
    }
  }
}
Copy into chat04

  • Run a search for [category] and tell me if the top 5 products match our 'high margin' rule.

  • Get the recommendations for [product ID] and check if any out-of-stock items are appearing.

  • Search for [query] using the sorted method and verify that the most popular items are at the top.

  • Compare the results of a search for [query] against our brand safety guidelines.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Constructor once, then ask.

Just link your account to start. Your credentials stay encrypted, and you can immediately start asking questions about your live site data.

Connect Constructor to your AI

FAQ

How this task behaves.

  • 01

    Can the AI change my search rules?

    No. The AI can only read the data and report on it. You must make all changes within the Constructor dashboard.

  • 02

    Does the AI see real-time data?

    Yes. It uses the tools to pull the current state of your search results and recommendations as they exist right now.

  • 03

    Can the AI fix broken recommendation pods?

    No. It can identify that a pod is broken or irrelevant, but it cannot modify the logic or the products being served.

  • 04

    How does the AI know my business rules?

    You must provide your rules in the chat. Tell the AI, 'Check these results against this rule: [your rule].'

  • 05

    Can the AI delete products from my catalog?

    No. This MCP only has read capabilities for searching and getting recommendations. It cannot delete or edit anything.

  • More questions about Constructor? See everything the Constructor Connector can do

Connect Constructor to Claude, Cursor, ChatGPT & more