Editing and Overriding AI Output

After you review your enrichment results, you'll likely find some values that need correction. Merchkit makes it easy to edit AI-generated values directly in the Products table. But understanding how manual edits interact with future enrichment runs is important—it affects your entire workflow.

Editing a Single Cell

To fix any AI-generated value:

1. Click the cell containing the value you want to edit in the Products table.
2. Type your correction directly into the cell.
3. Press Enter to save.

That's it. Your manual edit now overrides the AI-generated value.

[SCREENSHOT: Products table with a single cell in edit mode, showing cursor and typed text]

How Manual Edits Interact with Future Enrichment

This is the critical part. When you run enrichment again, how your manual edits are treated depends on which Generate option you choose.

Scenario 1: You Run "Generate Empty Attributes Only"

What happens: The AI only fills in attributes that are currently empty. Your manually edited cells are preserved—they're not empty, so the AI skips them.

Example:

• You enrich "Description" for 100 products using "Generate Empty Attributes Only."
• You review and manually fix the description for 5 products.
• Later, you enrich "Key Features" for the same 100 products using "Generate Empty Attributes Only."
• The 5 descriptions you corrected remain unchanged. The "Key Features" attribute is generated for all 100 products, including the 5 with corrected descriptions.

This is the safest approach when you have a mix of AI-generated and manually edited values.

Scenario 2: You Run "Generate All Attributes"

What happens: The AI re-generates every AI-configured attribute for the selected products, including attributes with existing values. Your manually edited cells are overwritten.

Example:

• You enrich "Description" for 100 products.
• You manually fix the descriptions for 5 products.
• You realize your prompt wasn't clear enough, so you update it.
• You select all 100 products and run "Generate All Attributes."
• All 100 descriptions are re-generated, including the 5 you manually fixed. Your edits are lost.

Use this option carefully. Only use "Generate All Attributes" if you're intentionally overwriting previous work (e.g., updating all values to match a new prompt).

[SCREENSHOT: Products table showing a previously edited cell, with visual indication that it's about to be overwritten]

The Best Practice: Manual Edit → Improve Prompt → Re-run

The strongest workflow combines manual edits with prompt refinement:

Step 1: Edit and Fix Values

Review your enrichment results, spot-check quality, and manually fix any values that are wrong using the technique above.

Step 2: Identify the Pattern

After fixing a few values, ask yourself: Why were they wrong? Common reasons:

• Your prompt was too vague (didn't specify what you wanted).
• Your Product Data context was missing a key field.
• Your Acceptable Values list was incomplete or misaligned.

Step 3: Update Your Attribute Configuration

Go to Workspace → Attributes, find the attribute you want to improve, and refine it:

• Rewrite your prompt to be more specific. Add examples if helpful.
• Add missing Product Data fields if the AI doesn't have enough context.
• Expand your Acceptable Values to cover edge cases you've identified.

[SCREENSHOT: Attributes configuration page with prompt and Acceptable Values being edited]

Step 4: Test on a Small Batch

Select 5–10 products that previously had poor output, and run "Generate All Attributes" on just those products. This tests your improved configuration without affecting already-correct values.

Review the results. If they're better, great—move to Step 5. If not, go back to Step 3 and refine further.

Step 5: Scale Up

Once you're confident your improved configuration works, run enrichment on larger batches using "Generate Empty Attributes Only." This fills in new products while preserving any manual edits you've already made.

[SCREENSHOT: Products table showing successful enrichment with improved output quality]

Understanding When to Edit vs. Re-configure

Edit individual cells when:

• It's a one-off error specific to that product (e.g., wrong spelling, typo, misplaced word).
• The prompt is working well overall, but this product is an edge case.
• You need a quick fix and don't have time to refine the prompt.

Re-configure and re-run when:

• Multiple products have the same type of error (systematic issue).
• Your prompt is consistently too vague, too specific, or missing context.
• You've updated Acceptable Values and want to enforce them across all products.
• You've realized that the information the AI needs isn't in your current Product Data selection.

Common Scenarios

Scenario: You've Edited Several Values and Now Want a Fresh Start

If you've made manual edits but now want the AI to regenerate everything (maybe your prompt has improved significantly):

1. Select the products you want to regenerate.
2. Run "Generate All Attributes."
3. Review the new results—your manual edits will be overwritten.

This is clean and straightforward, but make sure you've actually improved your configuration first. Otherwise, you'll just get the same poor results again.

Scenario: You Want to Re-run Only One Attribute Without Affecting Others

If you've edited multiple attributes and want to regenerate just one of them:

1. Go to Workspace → Attributes.
2. Find the attribute you want to re-run.
3. Return to the Products view, select your products, and run enrichment.
4. Choose your Generate option.

Since the Generate function only touches attributes with "Use AI" enabled, other attributes (whether manually edited or formulas) remain untouched. Enrichment runs on the selected attribute(s) only.

Note: If you use "Generate All Attributes," all AI-configured attributes are re-generated, not just one. If you only want to regenerate a single attribute, use "Generate Empty Attributes Only" and manually clear the cells of the attribute you want to refresh (set them to blank) before running.

Common Issues

I ran "Generate All Attributes" and lost my manual edits

• Unfortunately, "Generate All Attributes" overwrites everything. If this was accidental, you can manually correct the values again using the editing technique above, or manually re-run your original edits from memory.
• Going forward: Use "Generate Empty Attributes Only" instead. It preserves manual edits.

I edited a value, but it changed back on its own

• This can happen if you ran "Generate All Attributes" later. Manual edits don't persist through "Generate All Attributes" runs.
• Solution: Use "Generate Empty Attributes Only" to preserve your edits, or be very intentional when using "Generate All Attributes" and only do so when you want to overwrite previous work.

My improved prompt worked on the test batch, but large batches still have issues

• Larger datasets include more edge cases. Run a few more small test batches from different categories, vendors, or price ranges to catch additional patterns before scaling.
• Alternatively, make manual edits on the edge cases and refine your Acceptable Values to cover them explicitly.

I want to compare my manual edit with the AI-generated value

• Merchkit doesn't store a version history of cell values. Once you edit, the previous AI-generated value is replaced. If you need to compare, take a screenshot or export your products before making bulk manual changes.

Next Steps

1. Re-run enrichment strategically: Check out Re-running Enrichment to learn how to progressively improve your configuration and scale enrichment across your entire catalog.
2. Refine your configuration: Return to Configuring an Attribute for AI Enrichment for tips on writing better prompts and setting up Acceptable Values that reduce the need for manual editing.