How charts are chosen
Last reviewed against the app on 24 September 2026
Every chart in an AI answer, AI dashboard or AI report is drawn from a calculation on your data. The kind of calculation mostly decides the chart type.
The default choice
When clariBI works out an answer it runs one or more calculations. Each calculation has a natural chart:
| Calculation | Usual chart |
|---|---|
| A single total or average | Gauge or metric value |
| Top or bottom items by a value | Bar, pie, funnel or radar |
| A value over time | Line or area |
| This period against the previous one | Waterfall (A, change, B), bar or line |
| How often each value occurs | Histogram |
| Relationship between two numeric columns | Scatter |
| Relationships between many numeric columns | Correlation matrix |
Those eleven types (gauge, bar, pie, funnel, radar, line, area, waterfall, histogram, scatter, correlation matrix) are the ones clariBI picks on its own.
Other chart types
Dashboards and templates can also use heatmap, treemap, bullet, candlestick, bubble, box plot, violin, Sankey, sunburst, parallel coordinates, choropleth map, network graph and table widgets, plus goal widgets (goal progress card, milestone tracker, goal performance chart, goal summary). clariBI does not pick these by itself, because none of its standard calculations produce those shapes. A template that defines one of them uses it; in a question you can ask for one, but the answer may come back as one of the standard types if the data does not fit.
Asking for a specific chart
Say what you want in the question:
Show a line chart of monthly revenue for 2025.
Pie chart of orders by channel for last month.
Scatter plot of price against units sold.
Table of the 20 largest customers by revenue.clariBI follows the request when the calculated data fits the chart. A pie chart of a time series, for example, will not work well; ask for a line chart instead.
AI dashboards
When you create a dashboard with AI, clariBI looks at the columns of the data sources you picked (type, number of distinct values, dates) and recommends up to 15 widgets. The dashboard is built from the recommendations you accept. Each widget is then filled from a calculation on your data, and its caption goes through the number check described in the AI overview.
When the chart is not right
- Too many slices: pie charts with many categories are hard to read. Ask for a bar chart or the top 10.
- Line chart for categories: lines suggest a sequence. Use bars for unordered categories.
- Wrong aggregation: say "average order value" or "total orders" rather than "orders".
- Wrong column: if two columns look alike (amount and total_amount), name the one you mean.
Related
Try it on your own data
The 14-day trial includes 50 AI credits and needs no card. When it ends you move to the Free plan and keep your data.