Skip to main content

​
LLM System Prompt Tuning Workflow

A well-tuned system prompt can be the difference between an LLM that gives generic responses and one that nails your use case. But finding the right prompt takes iteration—testing different instructions against real inputs and comparing outputs. This guide shows you how to build a workflow that automates this process. Put your prompts and test cases in a Google Sheet, run the workflow, and get all responses written back automatically.

​
What You’ll Build

Complete LLM prompt tuning workflow
The workflow:
  1. Reads system prompts and test messages from Google Sheets
  2. Iterates through each row
  3. Sends each message to an LLM with its system prompt
  4. Writes the response back to your sheet

​
Prerequisites

  • A NoClick account (sign up here)
  • A Google account for Google Sheets access

​
Step 1: Set Up Your Google Sheet

Create a Google Sheet with four columns:
ColumnPurpose
System PromptYour instructions for the LLM
MessageThe test input
Expected AnswerWhat you’re hoping to get (for comparison)
Actual AnswerWhere results get written (leave blank)
Google Sheet setup
Add a few rows with different system prompts you want to test. For example:
System PromptMessageExpected AnswerActual Answer
You are a helpful assistant. Be concise.What is Python?A programming language
You are a coding expert. Always include examples.What is Python?A programming language with code example

​
Step 2: Import the Workflow

Click the copy icon in the top-right corner of the code block below, then paste into NoClick.
{
  "type": "noclick-workflow",
  "version": "1.0",
  "nodes": [
    {
      "id": "sheets_read",
      "type": "automation-google-sheets",
      "position": { "x": 200, "y": 285 },
      "config": {
        "operation": "read",
        "sheet_name": "Sheet1"
      }
    },
    {
      "id": "iteration",
      "type": "iteration",
      "position": { "x": 380, "y": 285 },
      "config": {
        "items": "{{sheets_read.values}}",
        "header_row": true,
        "concurrency": 3
      }
    },
    {
      "id": "agent",
      "type": "agent",
      "position": { "x": 585, "y": 440 },
      "config": {
        "model": "openrouter/openai/gpt-4o-mini",
        "message": "{{iteration.item.Message}}",
        "system_prompt": "{{iteration.item.System Prompt}}"
      }
    },
    {
      "id": "sheets_write",
      "type": "automation-google-sheets",
      "position": { "x": 920, "y": 285 },
      "config": {
        "range": "D{{iteration.row_number}}",
        "values": "{{agent.response}}",
        "operation": "write",
        "sheet_name": "Sheet1"
      }
    }
  ],
  "edges": [
    { "source": "sheets_read", "target": "iteration" },
    { "source": "iteration", "target": "agent" },
    { "source": "agent", "target": "sheets_write" },
    { "source": "iteration", "target": "sheets_write" }
  ]
}
To import:
  1. Create a new workflow in NoClick
  2. Press Cmd/Ctrl + V to paste the JSON
  3. The workflow will appear in your canvas

​
Step 3: Connect Your Google Account

Click on either Google Sheets node and select Connect Google Account. This opens a new tab where you authorize NoClick to access your sheets.
Google account connection
NoClick only accesses the specific sheets you select. We never store your Google credentials—we use secure OAuth tokens that you can revoke anytime.

​
Step 4: Configure the Sheet Nodes

​
Read Node (First Google Sheets Node)

  1. Click the first Google Sheets node
  2. Select your spreadsheet from the dropdown
  3. Verify “Sheet1” is selected (or change to your sheet name)

​
Write Node (Second Google Sheets Node)

  1. Click the second Google Sheets node
  2. Select the same spreadsheet
  3. The range D{{iteration.row_number}} writes to column D for each row
Node configuration

​
Step 5: Run the Workflow

Click Run in the top right corner. The workflow will:
  1. Read all rows from your sheet
  2. Process up to 3 rows concurrently (configurable in the iteration node)
  3. Send each message with its system prompt to GPT-4o-mini
  4. Write responses back to column D

​
Step 6: Review Results

Go back to your Google Sheet—the Actual Answer column now has all your LLM outputs.
Results in Google Sheet
Compare against your expected answers, tweak your prompts, and run again. No more manual copy-pasting!

​
Customization Options

​
Change the LLM Model

Click the Agent node and select a different model:
  • openrouter/anthropic/claude-3-5-sonnet - More capable, higher cost
  • openrouter/openai/gpt-4o - OpenAI’s flagship model
  • openrouter/google/gemini-pro - Google’s model

​
Adjust Concurrency

In the Iteration node, change concurrency to control how many requests run in parallel:
  • 1: Sequential processing (slower, lower API costs)
  • 3-5: Balanced (default)
  • 10+: Fast but may hit rate limits

​
Add More Test Data

Simply add more rows to your sheet—the workflow processes all rows with data.

​
Next Steps

  • Connect NoClick to your AI assistant to build workflows conversationally
  • Explore other workflow nodes like Slack notifications, email, or webhooks
  • Schedule your workflow to run automatically on a timer