> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://help.ubuyfirst.com/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Setting Up AI Columns

Table Of Contents

* [What Are AI Columns?](#0-toc-title)
* [When to Add AI Columns](#1-toc-title)
* [How AI Columns Work](#2-toc-title)
* [Step-by-Step Setup](#3-toc-title)
* [Step 1: Decide What Columns You Need](#4-toc-title)
* [Step 2: Update Your Prompt](#5-toc-title)
* [Step 3: Create Column Profile in SKU Manager](#6-toc-title)
* [Step 4: Update Your Filter](#7-toc-title)
* [Step 5: Test It](#8-toc-title)
* [Common Use Cases](#9-toc-title)
* [Use Case 1: Filter Out Listings You Don’t Buy](#10-toc-title)
* [Use Case 2: Visual Scanning with Color Coding](#11-toc-title)
* [Use Case 3: Two-Stage Analysis (Save API Costs)](#12-toc-title)
* [Use Case 4: Hidden Columns for Logic Only](#13-toc-title)
* [Tips & Best Practices](#14-toc-title)
* [Keep Values Consistent](#15-toc-title)
* [Start Small](#16-toc-title)
* [Use “Return Original”](#17-toc-title)
* [Column Order Matters](#18-toc-title)
* [Test with Edge Cases](#19-toc-title)
* [Troubleshooting](#20-toc-title)
* [Columns Not Appearing in Grid](#21-toc-title)
* [Columns Show Empty/Blank Values](#22-toc-title)
* [Columns Show Wrong Values](#23-toc-title)
* [Want to Refresh Existing Results?](#24-toc-title)
* [Quick Reference](#25-toc-title)
* [What’s Next?](#26-toc-title)

**Add quick-decision fields directly to your results grid.**

---
## What Are AI Columns?

AI Columns are custom fields that appear in your results grid, giving you at-a-glance decisions without opening the External Panel.

| Display Template | AI Columns |
| --- | --- |
| Detailed analysis in External Panel | Quick values in results grid |
| Full HTML formatting | Simple text (Yes/No, High/Low, etc.) |
| Read one listing at a time | Scan many listings at once |
| Best for understanding | Best for fast decisions |

**Example AI Columns:**

* “Buy?” → Yes / No / Maybe
* “Condition” → Excellent / Good / Fair / Poor
* “Risk Level” → High / Medium / Low
* “Model” → iPhone 15 Pro / iPhone 15 / Unknown
* “Est. Profit” → $45 / $20 / -$10

---
## When to Add AI Columns

✅ **Add AI Columns AFTER you have:**

* A working prompt returning good analysis
* Display template showing correct results
* Tested with 10+ listings successfully
* Identified key decisions you make repeatedly

❌ **Don’t start with AI Columns** — get the basics working first. They add complexity.

---
## How AI Columns Work

Your prompt outputs JSON. AI Columns pull specific fields from that JSON into grid columns.

```
Prompt runs → JSON output → AI Columns extract values → Values appear in grid

```

**Example JSON output:**

```
{
  "shouldBuy": "Yes",
  "riskLevel": "Low",
  "estimatedProfit": "$35",
  "modelIdentified": "iPhone 15 Pro 256GB"
}

```

Each of those fields can become a column in your results grid.

---
## Step-by-Step Setup
### Step 1: Decide What Columns You Need

Before touching SKU Manager, decide:

* What quick decisions do you make on every listing?
* What values would help you scan faster?
* Keep it simple — **3-5 columns max** to start

**Good first columns:**

* Buy/Pass decision
* Risk or confidence level
* Key identifier (model, variant, etc.)

---
### Step 2: Update Your Prompt

Go back to your Claude or ChatGPT project and ask:

```
My prompt is working well. Now I want to add AI Columns for quick decisions.

In the grid, I need:
1. "Can Return" column (Yes/No/Unknown)
2. "Original Owner" column (High/Medium/Low/Unknown)

Can you update my prompt to output these as separate fields with exact values?

```

The AI will add a section to your prompt like:

```
## AI COLUMN CONFIGURATION

### 1. "Can Return" Field – For AI Column
Must return EXACTLY: "Yes", "No", or "Unknown"

### 2. "Original Owner" Field – For AI Column  
Must return EXACTLY: "High", "Medium", "Low", or "Unknown"

```

**Important:** The prompt must specify EXACT output values. AI Columns need consistent, predictable text.

---
### Step 3: Create Column Profile in SKU Manager
1. Open External Data panel (**Data tab > Get Data**)

2. Go to **“AI Columns”** tab

3. Click **“Add”** to create new profile

4. Name it (e.g., “Phone_Analysis_Columns”)

5. **Enter each column on a SEPARATE LINE:** `Can Return Original Owner`

6. Press **Enter** after each column name

7. Check ✅ **“Return Original”** (recommended) — keeps original values if AI doesn’t return a match

8. Click **Save**

📸 _[Screenshot: AI Columns tab with column names entered on separate lines]_

---
### Step 4: Update Your Filter
1. **Close** the External Data panel

2. Go to **Home > Filters**

3. Edit your existing filter

4. In the **Columns** dropdown, select your new columns profile

5. Click **OK** to save

Your filter should now have all four profiles connected:

* ✅ Prompt
* ✅ Display Template
* ✅ Fields to Send
* ✅ Columns ← **new**

---
### Step 5: Test It
1. Reopen External Data panel

2. Restart the script (**Stop Script** → **Script Started!**)

3. Run a search that triggers your filter

4. Check your results grid — new columns should appear with values

**To refresh columns on existing results:** Right-click the column header → **“Update LLM Columns”**

---
## Common Use Cases
### Use Case 1: Filter Out Listings You Don’t Buy

**Scenario:** You search “iPhone 15” but only buy iPhone 15 Pro models.

1. Create AI Column: `Buying` (Yes/No)

2. Prompt logic identifies Pro vs non-Pro

3. Create a second filter:

* Condition: AI Column “Buying” = “No”
* Action: **Remove Row** (or hide/dim)

4. Condition: AI Column “Buying” = “No”

5. Action: **Remove Row** (or hide/dim)

Now non-Pro models automatically disappear from your results.

---
### Use Case 2: Visual Scanning with Color Coding
1. Create AI Column: `Risk` (High/Medium/Low)

2. Create filters to highlight rows:

* Risk = “High” → Red background
* Risk = “Medium” → Yellow background
* Risk = “Low” → Green background

3. Risk = “High” → Red background

4. Risk = “Medium” → Yellow background

5. Risk = “Low” → Green background

Instantly spot risky listings while scanning.

---
### Use Case 3: Two-Stage Analysis (Save API Costs)

**Problem:** Full analysis on every listing is slow and expensive.

**Solution:** Chain your analysis:

1. **First filter** — Quick screening prompt with AI Columns only

* Display Template: (leave blank or minimal)
* Columns: `Worth Analyzing` (Yes/No)
* Fast, cheap analysis

2. Display Template: (leave blank or minimal)

3. Columns: `Worth Analyzing` (Yes/No)

4. Fast, cheap analysis

5. **Second filter** — Triggers only when `Worth Analyzing` = “Yes”

* Full detailed prompt
* Complete Display Template
* Only runs on promising listings

6. Full detailed prompt

7. Complete Display Template

8. Only runs on promising listings

This can significantly reduce API costs and speed up your workflow.

---
### Use Case 4: Hidden Columns for Logic Only

You can create AI Columns that **don’t display** in the grid but power your filters.

Example: Create a `category_match` column that outputs internal codes, then use filters to route listings to different prompts based on the code.

---
## Tips & Best Practices
### Keep Values Consistent

AI Columns work best with a small set of exact values:

* ✅ Good: “Yes”, “No”, “Unknown”
* ❌ Bad: “Yes!”, “yes”, “Y”, “Definitely yes”

Your prompt should specify exact allowed values.

### Start Small

Begin with 2-3 columns. Add more only after those work reliably. Too many columns slow processing and clutter your grid.

### Use “Return Original”

Check this option to preserve values when the AI doesn’t return a recognized response. Helps with debugging.

### Column Order Matters

Columns appear in the order you enter them. Put most important columns first.

### Test with Edge Cases

Try listings with missing data, unusual conditions, or ambiguous titles. Make sure your columns handle them gracefully.

---
## Troubleshooting
### Columns Not Appearing in Grid

* Is the Columns profile selected in your filter?
* Did you restart the script after changes?
* Are column names spelled exactly the same in prompt and profile?

### Columns Show Empty/Blank Values

* Check that your prompt outputs the exact field names
* Verify JSON structure is valid
* Look at the full AI response in External Panel for clues

### Columns Show Wrong Values

* Review your prompt’s column configuration section
* Make sure allowed values are clearly specified
* Submit feedback trace and refine with your AI project

### Want to Refresh Existing Results?

Right-click on any AI Column header → **“Update LLM Columns”**

---
## Quick Reference

| Task | Where |
| --- | --- |
| Create column profile | External Data panel > AI Columns tab |
| Connect to filter | Home > Filters > Edit > Columns dropdown |
| Refresh existing results | Right-click column header > Update LLM Columns |
| Add column logic to prompt | Your Claude/ChatGPT project |

---
## What’s Next?

Once your AI Columns are working:

* Create filters that act on column values (highlight, remove, trigger additional analysis)
* Experiment with chained analysis for speed/cost optimization
* Use Feedback Traces to refine column accuracy

---

See also: [Quick Start Guide](https://claude.ai/chat/ee92e4d4-6210-417b-917b-d9a11d0f392f#) | [Troubleshooting & FAQ](https://claude.ai/chat/ee92e4d4-6210-417b-917b-d9a11d0f392f#) | [Using Feedback & Traces](https://claude.ai/chat/ee92e4d4-6210-417b-917b-d9a11d0f392f#)