Introduction
AI Columns are built for fast iteration. Test on a small sample in seconds, review results, adjust your prompt or categories, and test again. The entire cycle takes minutes, not hours.
In this guide, we'll show you how to use the Results Report and Ask Rippit to quickly improve your AI Column accuracy.
What is the Results Report?
By default, after you click Test or Run, you should see the Results Report appear which shows you:
How your AI Column has performed and organized your results
How many rows were analyzed
An auto-generated summary analyzing the results of your run while providing feedback on how to improve
If you don't see this, just click on Results or re-run your classifier again!
General Guidelines
Here is a quick checklist of what to keep an eye out for as you are evaluating the general quality of AI Column results.
✅ Good signs:
Relatively even distribution across categories (unless you expect skew)
Low percentage in "Other" or catch-all categories (under 15%)
Categories are being used (no 0% categories)
⚠️ Warning signs:
One category dominates (80%+ of results) - might be too broad
High percentage in "Other" (30%+) - missing important categories
Empty categories (0%) -
either unnecessary or prompt needs adjustment
Two categories with very similar counts - might be overlapping
Reviewing your AI Column Results with Ask Rippit
Built Without Ask Rippit
If you created the AI Column without using Ask Rippit, you will see an auto-generated AI Summary along with the button View AI Summary.
Clicking View AI Summary opens a new Ask Rippit conversation with your results pre-loaded and a starter prompt to help you analyze them.
Sourced from Ask Rippit
If your AI Column was originally generated by Ask Rippit, you will have the opportunity to use your original Ask Rippit conversation by clicking on Continue in Ask Rippit.
This allows you to unite any previous context that helped you build this prompt with updated insights you just obtained.
Targeting an AI Column Response
Regardless of how the AI Column was generated, you can focus on understanding why a specific response by clicking on the response in the report.
This will automatically filter your worksheet to all rows that match that response.
You can click on the Ask Rippit button right next to the response to start a conversation that will focus on these rows.
An example of this flow can be seen here!
What to ask Ask Rippit
Here are the types of questions that help you understand and improve your AI Column results:
Understanding Your Results
"Show me examples from [category]"
Example: "Show me 5 examples from the 'High Churn Risk' category"
"Why are so many results in [category]?"
Example: "Why are 40% of tickets classified as 'Other'?"
"What patterns exist in [category]?"
Example: "What patterns exist in tickets classified as 'Technical Issues'?"
Discovering What's Missing
"What categories am I missing?"
Example: "What categories am I missing based on the 'Other' results?"
"Are any of my categories overlapping?"
Example: "Are 'Billing Questions' and 'Account Issues' overlapping?"
"What percentage of [category] actually belongs in [other category]?"
Example: "What percentage of 'Medium Priority' should actually be 'High Priority'?"
Getting Refinement Suggestions
"How can I improve this classification?"
Example: "How can I improve my support ticket theme classification?"
"Should I split [category] into subcategories?"
Example: "Should I split 'Technical Issues' into subcategories based on the data?"
"Suggest better category names"
Example: "Suggest better category names for my churn risk classifications"
"What context should I add to my prompt?"
Example: "What context should I add to my prompt to reduce 'Other' classifications?"