> ## Documentation Index
> Fetch the complete documentation index at: https://user-guides.demand-genius.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Creating Smart Columns
> Run your own analysis across the library with a custom prompt or a template, choose output type and sample size, test on ten rows, and understand what it costs in credits.
The standard KPIs answer a fixed set of questions about every page. A Smart Column answers yours. Write a prompt, choose whether you want a tag or free text back, pick how much of each page the model reads, and run it across ten pages, a filtered set or the whole library. Tone of voice, authority signals, product mentions, competitor references: anything you can describe, you can score.
Smart Columns are the only feature that uses credits. Everything else, including KPI scoring, runs outside the credit system.
## Create a Smart Column
The **Add Smart Column** dialog opens on the **Template** tab.
Templates are pre-built prompts maintained by Demand-Genius, filterable by persona and purpose. Choosing **Use** on a template switches to the **Custom** tab with the form pre-filled so you can adjust it. Or select **Custom** to start blank.
The prompt runs against every page in scope, so be specific. The hint on the field says it well: think task, then context, then output.
Two help links sit on the left of the dialog: general prompt guidance and a Demand-Genius prompt-builder GPT. [You can access the prompt-builder GPT directly here](https://chatgpt.com/g/g-68bea35c87808191b2ab5850ed1d6c97-demand-genius-promptbuilder) → enter your use case or goal, and it'll help you work towards an effective prompt.
**Tag** returns a defined label per page, giving clean data you can filter, chart and compare. It's great for assessing and quantifying problems or trends in your content library.
**Text** returns more detail per page, for analysis and nuance - for example, if you'd like AI to suggest possible improvements to a piece. Under Tag, **Reporting Tag** makes the tag available to [Performance tracking](/content-intelligence/performance-tracking) so you can filter the Performance columns by it; it needs Performance enabled on your workspace.
**400 Words** reads the opening of each page; fast and cheap, enough for tone or topic. **Full Page** reads everything; needed for anything about completeness or specific mentions.
**10 rows** runs on a sample. **Selected rows** runs on the rows you ticked in the grid. **Filtered rows** runs on the current filter. **All rows** runs on the whole library. Tick **skip rows that already have a value** to avoid re-running pages you have already scored.
If you have multiple directories, pick which libraries the column runs across.
The footer shows **credits/row**, the estimated total and your balance. **Run** starts the job. **Save** stores the column definition without running it.
\*\*The column will appear to the right of the database, with the relevant cells highlighted while the AI runs. \*\*
Test every new prompt on 10 rows first. Read the output, tighten the prompt, then run the full library. The dialog says the same thing and it is the single biggest way to avoid wasting credits.
## What it costs
| Output type | Sample size | Credits per row |
| ----------- | ----------- | --------------- |
| Text | Full Page | 6 |
| Text | 400 Words | 4 |
| Tag | Full Page | 3 |
| Tag | 400 Words | 1 |
Your balance shows at the bottom of the sidebar and under **Settings > Billing**. Credits are allocated to your workspace by Demand-Genius. If you start a run with too few credits you are warned that it will complete partially: rows are processed until the balance runs out.
## Working with the results
* We recommend treating the database like a **waterfall**. Use **tags** to narrow down issues. For example, progressively narrow down 1,000 pieces to the 20 that you'd like to prioritise in light of a given goal. Then use **text** to help you figure out the exact actions / updates that are needed.
* Tag columns behave like any other tag: filter the grid by them, pivot the Insights heatmap on them, export them.
* **Edit Column** on the column header changes the name or the prompt. Re-run to apply a changed prompt to existing rows.
* **Re-run classification** from the row action bar re-scores selected rows for a column.
* Delete a column from its header menu. The values go with it.
## Troubleshooting
* **Some rows came back empty.** The run hit the credit balance, or the page was not scrapable. Check the Scrapable column and your balance.
* **The tag values are inconsistent.** Tag prompts work best when the prompt lists the allowed values explicitly. Edit the prompt and re-run on 10 rows.
* **A job seems stuck.** Check **Settings > Jobs** for a Category Classification job. Use **Stop classification** on the row action bar if you need to halt it.