The Chart Selection Process
When you ask a question or run an analysis, the AI does not just answer -- it also decides how to visualize the answer. This process follows a structured logic:
Step 1: Data Shape Analysis
The AI examines the result data:
- Number of columns -- one metric vs. multiple metrics
- Data types -- dates, categories, numbers, text
- Row count -- a few values vs. hundreds of data points
- Cardinality -- how many unique values a category column has
Step 2: Question Intent Detection
The AI classifies your question into an intent category:
| Intent | Example Questions | Typical Chart |
|---|---|---|
| Trend | "How has revenue changed?" | Line chart |
| Comparison | "Which product sells most?" | Bar chart |
| Composition | "What is the revenue breakdown?" | Pie or donut chart |
| Distribution | "How are deal sizes spread?" | Histogram |
| Relationship | "Is there a correlation between X and Y?" | Scatter plot |
| Single metric | "What is total revenue?" | Metric card |
Step 3: Chart Matching
The AI scores each candidate chart type against the data shape and intent. Factors include:
- Data type compatibility -- a line chart needs a sequential X-axis (dates), so it scores poorly if the X-axis is categorical
- Cardinality fit -- a pie chart works with 3-6 slices but scores poorly with 20+ categories
- Intent alignment -- trend questions boost line chart scores; comparison questions boost bar chart scores
- Readability -- the AI penalizes chart types that would produce cluttered or unreadable output
The highest-scoring chart type wins.
What the Confidence Score Means
The confidence score is a percentage (0% to 100%) displayed alongside each AI-generated visualization. It represents how well the selected chart type matches the data and question.
Score Ranges
| Score | Meaning | Action |
|---|---|---|
| 90-100% | Strong match -- the chart type is well-suited to the data and question | Accept the recommendation |
| 75-89% | Good match -- the chart works but another type might also work | Accept, or try an alternative |
| 60-74% | Moderate match -- the chart shows the data but may not be ideal | Consider overriding with a different type |
| Below 60% | Weak match -- the AI is not confident this visualization is right | Override with a more appropriate type |
What Lowers the Confidence Score
- Ambiguous questions -- "Show me the data" does not indicate a clear intent
- Mixed data types -- a result with both time-series and categorical columns
- Too many or too few data points -- 200 categories in a pie chart, or 2 data points in a scatter plot
- Missing expected fields -- the AI expected a date column for a trend but did not find one
Overriding the Chart Type
If the AI's chart choice is not what you want, you can change it:
In Conversational Analytics
Below the chart in the response, click Change Chart Type. A dropdown shows all compatible chart types for the data, each with its own confidence score. Select a different type and the visualization updates immediately.
In Dashboard Widgets
When creating a widget from an AI result, you can pick any chart type during the widget configuration step. The AI's recommendation appears as the default, but you are not locked into it.
In Reports
Reports display the AI's chosen chart. To change it, click the chart in the report editor and select a new type from the visualization options.
Common Overrides
| AI Chose | You Might Prefer | When |
|---|---|---|
| Bar chart | Horizontal bar | Category labels are long (product names, URLs) |
| Pie chart | Bar chart | More than 6 categories |
| Line chart | Area chart | You want to emphasize volume/magnitude |
| Scatter plot | Bubble chart | You have a third numeric dimension |
| Table | Bar chart | You want a visual comparison, not raw numbers |
| Metric card | Line chart + metric | You want the number plus its trend |
Improving AI Chart Selection
You can guide the AI toward better chart choices:
- Include the visualization type in your question. "Show me a line chart of revenue over time" removes ambiguity.
- Specify the comparison dimension. "Compare by region" tells the AI to use a categorical axis.
- Limit the data. "Top 5 products" produces cleaner charts than "all products."
- Use follow-ups. If the first chart is not right, say "Show that as a bar chart instead."
Multiple Visualizations Per Response
For complex analyses, the AI may generate more than one chart in a single response. For example, asking "Analyze our Q3 sales performance" might produce:
- A line chart showing the revenue trend across Q3 months
- A bar chart comparing revenue by product category
- A metric card with the total Q3 revenue and year-over-year change
Each chart has its own confidence score. The AI arranges them in the order it considers most informative -- the highest-confidence, most relevant chart appears first.
How Chart Selection Improves Over Time
The AI does not learn from your individual usage (your data is not used for training), but the chart selection logic benefits from general improvements to the AI engine's understanding of data visualization best practices.
Within a conversation, however, the AI does learn from your feedback:
- If you override a chart type (e.g., change a pie chart to a bar chart), the AI takes this as a signal for subsequent questions in the same conversation.
- If you ask "Show that as a scatter plot," the AI remembers this preference for follow-up questions.
- These preferences reset when you start a new conversation.
How Chart Selection Works
clariBI's chart selection is powered by an AI engine that analyzes both the data structure and the semantic meaning of your question. It considers:
- The field names and their likely meaning (e.g., a column named "date" is temporal)
- Common visualization practices in business analytics
- The specific wording of your question for intent cues
- The number of data points in the result set
- Whether the data has a natural ordering (dates, stages) or is unordered (categories)
This is why natural-sounding questions tend to get better chart recommendations than terse or ambiguous prompts.
Chart Selection vs. Data Accuracy
The confidence score reflects chart type suitability, not data accuracy. A chart can have a 95% confidence score (meaning "this is the right chart type") while the underlying data query could still have issues. Always check the data table alongside the chart to verify accuracy. For more on AI accuracy, see When AI Analysis Falls Short.