Setting Up AI Columns

Add quick-decision fields directly to your results grid

The One Thing to Know.

AI columns are defined in your system prompt, not created anywhere in the app. SKU Manager reads your prompt, auto-detects the columns it defines, and you tick the ones you want in the grid. If you remember only that, the rest of this page is detail.

Table of Contents #

1. What Are AI Columns?
2. When to Add Them
3. How They Work
4. Step-by-Step Setup
5. What You Can Do With Them
6. Tips & Best Practices
7. Troubleshooting

---

What Are AI Columns? #

AI columns are custom fields the AI fills in for every listing it analyzes, right in your results grid, so you get at-a-glance decisions without opening each analysis. In the grid the column header starts with ai, so Can Return reads ai Can Return.

	Display Template	AI Columns
Where	AI Sku Analysis panel	Results grid
Format	Full HTML formatting	Short exact values (Yes/No, High/Low)
Reading style	One listing at a time	Scan many listings at once
Best for	Understanding a listing	Fast decisions

Example columns:

• Can Return → Yes / No / Unknown
• Condition Grade → Excellent / Good / Fair / Poor
• Risk Level → High / Medium / Low
• Est. Profit → $45 / $20 / −$10

---

When to Add Them #

Add AI columns after you have:

• A working prompt returning good analysis
• A display template showing correct results
• Tested with 10+ real listings
• Noticed which decisions you make on every single listing — those are your columns

Don’t Start with Columns.

Get the basic round-trip working first. Columns add moving parts, and a broken column on top of an untested prompt is twice as hard to debug.

---

How They Work #

Your prompt outputs JSON. Each field in that JSON can become a grid column:

{
  "Can Return": "Yes",
  "Risk Level": "Low",
  "Condition Grade": "Good"
}

Prompt runs → JSON comes back → SKU Manager fills your ticked columns with the values.

The app finds your columns by reading the JSON keys in your prompt’s example output, such as "Can Return": "Yes". A name that is not a JSON key there is not listed. Open the Fields & Columns tab, or click the Auto Detect Custom Columns icon at the top of the columns panel, and the list reads the prompt again. The exact allowed values are for the AI and for your sorting, not for the detector.

---

Step-by-Step Setup #

1. Decide What Columns You Need #

Before touching anything: what quick decisions do you make on every listing? Keep it to 3–5 columns to start — a buy/pass decision, a risk or confidence level, and a key identifier is a solid first set.

2. Have your AI Project Update the Prompt #

Go back to your SKU Manager Assistant 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. "Risk Level" column (High / Medium / Low / Unknown)

Please update my prompt to output these as separate fields with exact values.

The assistant adds a section like this to your prompt:

## AI COLUMN CONFIGURATION
### "Can Return" — AI Column
Must return EXACTLY one of: Yes, No, Unknown
### "Risk Level" — AI Column
Must return EXACTLY one of: High, Medium, Low, Unknown

Exact Values are the Whole Game.

The grid sorts, groups and filters on these values verbatim. Yes, not yes, Y or Yes!. Give every column an Unknown fallback so it’s never blank, and make sure values can’t overlap.

3. Paste the Updated Prompt into Your Profile #

Open Data → Ai → AI Settings. While the server runs the configuration is locked: click Stop Service [Edit Configuration] first. Select your profile in the Profile dropdown, and on the AI Configuration tab replace the System Prompt with the updated one — paste the whole prompt from your master file, don’t hand-edit in the box.

4. Tick the New Columns #

Go to the Fields & Columns tab. Under User-Defined Custom Columns (Results), your new columns appear under Detected — not selected. Tick them — they move up to Selected.

Ignore the Junk.

The detector lists every JSON key it can find in your prompt, including ones that aren’t real columns. Tick only the columns you defined under ## AI COLUMN CONFIGURATION and leave the rest unticked.

5. Save and Test #

1. Click Start Service [Save Configuration] — that button is the save.
2. Run a search your filter matches. New listings fill the columns.
3. Rows analyzed under the old prompt read Outdated. Click the refresh icon in a row’s Current AI Status cell to analyze it again.

After you change the prompt or model, rows analyzed under the old settings read Outdated. To analyze one again, click the refresh icon in its Current AI Status cell, or click the row if it is not the selected one. This is paid, and the app does not ask first. To re-run a row that reads Analyzed, right-click it and choose Refresh AI analysis. The app asks first.

---

What You Can Do With Them #

Scan instead of read #

With a Can Return and a Risk Level column showing, you can work the grid directly — skip every No, look first at every Low risk — and only open the full analysis for listings that survive the scan.

Sort and group #

Because the values are short and exact, the grid can sort and group on them like any other column. Put your decision column first and the grid becomes a worklist.

Show the same value in both places #

Many users put the key decision in both the column (for scanning) and the display template (with the reasoning behind it, for when you click in). Ask your assistant to keep the two consistent — that’s exactly what it’s for.

---

Tips & Best Practices #

• Start small: 3–5 columns, add more only after those work reliably. Too many slow processing and clutter the grid.
• Keep one master prompt file — every column change is a prompt change; re-paste the whole prompt each time.
• Test edge cases — listings with missing data, odd conditions, ambiguous titles. The Unknown fallback should catch them, not a blank.
• Ask for reasoning in the template — when a column value looks wrong, the template’s reasoning section tells you why the AI chose it. See Building Your AI Analysis.

---

Troubleshooting #

The column isn’t in the list at all — not even under Detected #

The app finds your columns by reading the JSON keys in your prompt’s example output, such as "Can Return": "Yes". A name that is not a JSON key there is not listed. Open the Fields & Columns tab, or click the Auto Detect Custom Columns icon at the top of the columns panel, and the list reads the prompt again.

The column is there but blank #

• Is it ticked under User-Defined Custom Columns (Results)?
• Does the prompt’s field name match the column name exactly? Case matters.
• Does the row read Outdated? It was analyzed before the change.

Check the row’s Current AI Status column. Not analyzed means it has not run yet, so click the row. Outdated means it ran under old settings, so click the refresh icon in its Current AI Status cell. Analyzed with errors means the AI request ended without a usable answer: check your API key and billing, then right-click the row and choose Refresh AI analysis.

The column shows wrong or inconsistent values #

• Tighten the allowed values in the prompt — the “Must return EXACTLY” line is doing the work.
• On a listing that got it wrong, click Feedback Trace at the top right of the AI panel and take the saved zip to your AI project for a targeted fix.

To report a wrong result, click the Feedback Trace link at the top right of the panel. Open your saved traces with Data → Ai → Open AI Traces.

Extra columns I never asked for #

That’s the detector listing every JSON key. Untick anything that isn’t a real column.

Questions? Live Support at uBuyFirst.com, or su*****@*******st.com.

See also: Quick Start Guide · Building Your AI Analysis · Troubleshooting & FAQ