Gains:
- Ability to choose the chart type from the question asked (comparison, trend, relationship, part-whole, distribution) rather than from the shape of the data
- Ability to improve readability with sorting by value, single highlight color and direct labeling
- Ability to avoid false matches such as 3D cake, multi-slice cake and relationship stick
A graph is essentially a sentence: "This data says this." The wrong type of graph will either make that sentence unintelligible or make it lie. In this unit, we will learn how to choose which chart type is right for the question at hand and use artificial intelligence as an advisor in this selection. The rule is: know what you want to say first, choose the graphic later.
First the question, then the chart
The type of chart depends on the question you ask, not the "shape" of the data. Five basic question families:
- Comparison: "Which is bigger?" → bar chart.
- Change over time (trend): “How has it changed?” → line chart.
- Part-whole (composition): "What is its share in the total?" → stacked bar, sometimes cake (with a small number of slices).
- Relationship (correlation): “Do two variables move together?” → scatter plot.
- Distribution: “How are the values spread out?” → histogram, box plot.
A few terms: categorical data are classes (city, product); numerical data are measured values (income, temperature); time series are measurements over time (monthly sales).
you want to ask
Data type
correct chart
To be avoided
Compare categories
categorical + numerical
Horizontal/vertical bar
3D cake
trend over time
time series
line
Multi-slice cake
Part-whole (2-4 parts)
categorical share
Stacked bar/pound cake
8 slice cake
Relationship of two numbers
numeric + numeric
Scatter
rod
Distribution of odd number
numeric
Histogram/box plot
line
A single significant number
single value
Big number (KPI card)
force graphics
Tip: A pie chart only works for 2-4 slices and if the shares are significantly different. If 5+ slices are required, the horizontal bar is almost always more readable.
Step by step: choosing the right chart
- Write the question in one sentence. "Which region has grown the most?" like.
- Identify the question family. Comparison, trend, relationship?
- Count the number and type of variables. How many dimensions are there, categorical or numerical?
- Select the candidate chart (from the table).
- Simplify. Discard unnecessary axes, grids, colors and effects.
- "Is it obvious at a glance?" do the test; Show someone 5 seconds and ask what they understand.
To consult the graph type for the AI, give the structure of the data and your question (not the raw sensitive data):
Your role: data visualization consultant. I have data with the following structure: [column names and types, e.g. "month (date), region (category), sales (number)"]. Number of lines ~[N].The question I want to answer is: "[single sentence]".Task: Suggest the type of chart that will display this question in the most honest and readable way. Give 1 primary, 1 alternative and write WHY it is appropriate/inappropriate for each. Also tell me why I should avoid potentially misleading options (e.g. 3D cake).
Different questions require different graphs from the same data:
Generate 3 DIFFERENT questions for the same data set and match the most appropriate chart type for each question. So show me the principle of 'subtracts chart type from question'. Data structure: [ ... ].
Colour, sorting and labeling
The right kind is not enough; Readability is in the details:
- Sort: Sort bars by value (largest to smallest), not alphabetically; This is how the eye makes the comparison.
- Highlight color: Don't color everything; Highlight the single bar/line you want to express, leave the rest grey.
- Direct label: Write the name at the end of the lines if possible; the viewer does not have to skim the bar from the legend (color description box).
- Low grid: Keep background lines faint; Highlight data.
Give readability suggestions for a bar chart: sorting (by value?), single highlight color selection, direct labeling, removal of unnecessary grid/axis. My goal is to show: "[region X is clearly ahead]". Say it concretely.
three mini cases
Case 1 — From cake to stick. One team showed the market share of 9 products in a 9-slice pie chart; The slices were close together and nothing could be understood. They switched to a horizontal bar sorted by value. The audience saw at a glance that the top 3 products accounted for 61% of the total.
Case 2 — Wrong question, wrong graph. One analyst was trying to show the "relationship between ad spend and sales" with a bar chart; The relationship was not visible. Artificial intelligence said that this was a relationship question and a scatter plot was required. When we switched to Scatter, a positive but weak relationship (saturation after a point) became clear.
Case 3 — Accent color. One manager had displayed all 12 bars in different colors; The eye didn't know where to look. They made only the bar to be explained orange and the rest gray. The "real message" of the meeting was captured in the first second; The discussion revolved around the right bar.
Weak prompt / Strong prompt
Weak:
Graph this data.
No questions; the model randomly (and often incorrectly) chooses a type.
Strong:
Your role: data visualization consultant. Data structure: month (date), channel (category: 4 values), revenue (number). My question: "Which channel grew the fastest in the last 12 months?" This is a TREND + COMPARISON question. Suggest the most honest chart type, tell me how to show 4 channels in one chart, which one to highlight and whether to start the axis from 0. Eliminate misleading options with justification.
The second includes the question, data structure, and fairness concern (axis); The suggestion would be accurate.
Small multi-graph: very small instead of single large
When you fit too many lines/categories into a chart, a "spaghetti chart" occurs; No lines are read. A powerful alternative is small multiples: repeating the same graph, small and side by side, for each category separately. For example, you can show the sales trend of 8 regions side by side as 8 small line charts, not as 8 lines in a single chart. All thumbnails share the same scale so the eye can easily compare.
Your role: data visualization consultant. I have 12-month trend data for 8 regions. 8 lines mixed in one chart. How do I set this up as 'small multiples'? Give concrete suggestions such as common axis scale, layout (how many columns), what to highlight in each small chart. Purpose: to easily compare regions.
Hint: "Should I put 8 lines into one chart?" If you are hesitating, the answer is usually no. After 4-5 lines, a small multiplot is almost always more readable.
5 second test and simplification
Once you choose the right type, simplify the chart. Everything that can be deleted is deleted: unnecessary grid lines, box edges, shadow, background color, three dimensions. This is often referred to as "reducing graphics ink"; That is, any pixel that does not display data is distracting. Then do the 5 second test: show the graph to someone for 5 seconds, close it and ask “what do you understand?” ask. If it can't say the right message, the chart is either of the wrong type or too crowded. This test tells the truth much faster than long explanations.
Common mistakes
- Choosing the chart before the question becomes clear: Selecting a type first and trying to fit the data into it.
- Cake all around: A cake with multiple slices or similar portions is unreadable.
- Alphabetical order: Makes comparison difficult; Sort by value.
- Color inflation: Coloring each category and de-emphasizing it.
- Bar to correlation question: Correlation requires a scatter plot.
- Forcing the chart to a single number: A large KPI card is enough for a single significant figure.
In summary
The type of chart derives not from the shape of the data, but from the question you ask: comparison requires a bar, trend requires a line, part-whole requires a stacked bar/planet, relationship requires a distribution, distribution requires a histogram/box. Sort by value, highlight with solid color, label directly, discard excess. Given the data structure and your question, the AI is good at suggesting the right type; But you make the final decision and honesty check.
Application task
Write 3 different questions for one of your data sets. (1) Place each question into a question family (comparison/trend/relationship/part-whole/distribution). (2) Give the data structure and questions to artificial intelligence and get a chart type recommendation. (3) Compare a chart with the wrong type (e.g., pie) and the correct type (e.g., bar) and write in one sentence which one conveys the message more clearly.
checklist
- [ ] Before selecting the chart, I wrote the question in one sentence.
- [ ] I placed the question in a family (comparison/trend/relationship/...).
- [ ] Type appropriate to the question and variable type.
- [ ] I sorted the bars by value.
- [ ] I used one highlight color and neutralized the rest.
- [ ] I avoided misleading types (3D cake, multi-slice cake).