Use an AI column when the desired value requires interpretation or generation rather than a deterministic formula. AI columns can summarize research, classify accounts, explain a recommendation, or draft personalized outreach content for each row.
#Use an inline AI prompt
- Add a column and select Use AI.
- Choose the inline prompt option.
- Write the Prompt Template. Type
{{to insert list variables or values from other columns. - Choose the output value type. Configure allowed options when returning a Select or Multi-select value.
- Optionally add run conditions, then create the column.
- Run one cell first to review the prompt and output before running the complete column.
A useful classification prompt states the task, supplies the row context, and defines the allowed answer. For example: classify the account as Enterprise, Mid-market, or SMB using its employee count, revenue, and account research, and return only one category.
#Use a list agent
Choose the agent option when the task should use a configured Bigmind agent rather than a single inline prompt.
- Select an agent that has a mode available for Lists.
- Select the agent mode.
- If the mode expects additional context, enable Variables and map each variable to a fixed value, list variable, or column value.
- Choose the output type and add run conditions if needed.
- Create and test the column on one representative row.
#Choose the right output type
- Long text: Research, summaries, explanations, and account theses.
- Select or multi-select: Qualification, segmentation, intent, or persona labels. Explicitly tell the AI to use only the configured options.
- Number: A score or estimated quantity. Define the scale and what each end means.
- Checkbox: A decision that should resolve to true or false.
- JSON: Structured data for a downstream Formula or Targetable People column. Describe the required keys and provide a short example in the prompt.
- Message: An email, call script, LinkedIn message, connection request, or generic message.
#Make AI columns reliable
- Reference only the row context the task needs.
- Describe the expected output and constraints explicitly.
- Use a typed output instead of free-form text when the answer must drive filters or automation.
- Add run conditions to avoid generating content for incomplete or disqualified rows.
- Test rows with complete, incomplete, and unusual source data before running the whole column.