Gains:
- Ability to put findings into a context-tension-resolution narrative and highlight insight ('what does this mean')
- Ability to sharpen the message with contrasts such as before/after and target/reality and connect each graphic to a sentence in the narrative
- Ability to build an honest story without falling into the traps of selective data, exaggeration, and false causality
People forget paintings, they remember stories. Data storytelling is transforming numbers from being a dry pile of graphics into a narrative with a beginning, middle and end. The aim is not to embellish the data; It is to answer the audience's questions "what happened, why is it important, what should I do" in the correct order. In this unit, we will learn how to use artificial intelligence to build an honest and memorable narrative from your data.
Three parts of the story: context, conflict, resolution
Every data story has a backbone:
- Context: Where do we stand? "Last year we entered 3 new markets."
- Conflict / tension: What has changed, what problem/opportunity exists? "But in one of them, sales have been falling for 6 months."
- Solution / action: What should we do? "We need to review the price in this market."
There is also the concept of insight: a meaningful finding from the data that the audience has not seen before. The graph shows the data; insight tells you what it means. Storytelling is fitting insight into a narrative.
Tip: Don't say "The data shows this," think "The data says this." The chart is proof; The story is the meaning of that evidence.
From insight to narrative: step by step
- Write the main insight in one sentence. "In market X, the decline is not due to price, but to delivery time."
- Establish the context. Where should the viewer start?
- Show tension. The difference between what is expected and what actually happens.
- Insert the evidence (graph). Each graph proves a sentence.
- State the solution/action. What should the viewer do at the exit?
- Reduce it to a single message. The story should be summarized in one sentence.
Your role: data story editor. My findings (raw): [item by item numbers/observations]. My main insight: "[single sentence]". Task: Turn this into a narrative with a 'context → tension → resolution' structure. For each section: 1-2 sentences of text + a suggestion for a graphic type to prove that section. Don't exaggerate, make up numbers; mark claim not in data as '[confirmation]'. Tone: [objective / inspirational / stimulating].
To connect the chart to the narrative:
Make this graphic part of a story: write the ONE message the graphic shows in one sentence, then suggest a 'setup' sentence that prepares the audience for the graphic and a 'so what' sentence after the graphic.Graphic: [recipe]. Insight: [ ... ].
"Before/after" and comparison narratives
The most powerful data stories establish an opposition: before/after, target/actual, us/competitor, expected/what happened. Contrast creates tension in the mind of the audience and sharpens the message.
Contrast type
sample narrative
suitable graphic
before/after
"9 days before process change, 4 days after"
Dual bar/slope chart
target/actual
"The target was 10% growth, we stayed at 4%"
target line bar
expected/happened
"We expected a summer decline, but the increase came"
Line + shaded expectation band
Part/whole
"70% of the losses are from one step"
Stacked bar with hover
Set up a 'before/after' data story. Before: [case+number].After: [case+number]. The reason for the change: [...].Task: Suggest a 3-sentence mini-narrative (context, change, result) and the type of chart that will show this contrast most honestly. Do not try to make the change bigger than it is.
Integrity: don't let the story distort the data
The shadow side of storytelling is selectively presenting data to strengthen the narrative. Beware of two traps:
- Selective data (cherry-picking): Showing the period/segment that fits the narrative and hiding the one that does not.
- Pseudo-causation: “A increased, B increased, so A caused B” — correlation is not causation.
Caution: A good story persuades; But persuasion should not turn into permission to distort data. “What data does not fit this narrative?” Ask yourself and show it if necessary.
three mini cases
Case 1 — From painting to story. A team presented a 12-line table to management, but no one understood anything. With AI, they found the insight: “80% of the profit comes from 3 products.” They built the presentation around this one sentence. For the first time, management made a clear decision (focus on these 3 products).
Case 2 — Before/after. A hospital had made a change that reduced the waiting time, but the presentation consisted of dry numbers. They set up a slope chart with the contrast "First 47 minutes, then 19 minutes". The impact of the change was immediately understood and the budget was expanded.
Case 3 — Correcting selective data. A marketer had charted only the top 2 weeks to show the success of the campaign. Ask AI “is there data that doesn't fit this narrative?” he asked; the model recalled the post-campaign decline. The team showed the entire period and gave the honest message that "follow-up is required for lasting impact."
Weak prompt / Strong prompt
Weak:
A fascinating story emerges from these data.
The desire to “impressive” pushes the model towards exaggeration and selective presentation; honesty is lost.
Strong:
Your role: data story editor. Key insight: "70% of losses are in the delivery step". Results: [numbers]. Task: build an honest narrative in the structure of context→suspense→solution, match a graph to each section, and indicate if there is data that DOES NOT fit this insight. Don't exaggerate, don't present correlation as causation, write [confirmation] for what is not in the data.
The latter both gives structure and requires honesty (show data that does not fit, do not claim causality).
One graphic, one message: distributing the narrative across slides
The strongest rule of thumb when putting a data story into a presentation is: one chart on one slide, one message on one chart. The viewer cannot look at three graphs at the same time and draw a conclusion. Rather than showing a complex chart all at once, it is very effective to progressively reveal it: first the empty axis, then the single line, then the comparison line, finally the emphasis. At each step you explain what you are saying; the viewer "reads" the chart with you.
Your role: data presentation editor. I have a single dense graph: [recipe]. Instead of showing this on one slide, make a 3-4 stage 'cascade reveal' plan that will take the viewer step by step. What should appear on the chart at each stage and what should I say? The only message to highlight at the end is: "[sentence]".
Make the chart title part of the narrative, too: “West region alone drives growth,” not “Regional Sales.” The title tells the result of the chart; The viewer looks knowing what he will see.
Tip: If there are more than two charts on a slide, you're probably cramming two or three separate messages onto one slide. Split: give each message its own slide.
Common mistakes
- Presenting a stack of graphics: Stringing graphics in a row without a narrative.
- Story without insight: Saying "what happened" but leaving out "what it means".
- Selective data: Hiding data that does not fit the narrative.
- False causality: Presenting correlation as "caused."
- Choosing genres for exaggeration: Using misleading graphics to magnify the effect.
- Multiple messages: Fitting more than one "main story" into a presentation.
In summary
Data storytelling is putting numbers into a context-tension-resolution narrative and highlighting insight (“what does this mean”). Contrasts (before/after, target/actual) sharpen the message. AI is powerful at turning your findings into narrative; But don't fall into the traps of selective data, exaggeration and false causality. A good story persuades and does not distort data.
Application task
Draw an insight from your data set. (1) Write it in one sentence. (2) Have the artificial intelligence produce a narrative in the context-tension-resolution structure and match a graphic to each section. (3) “What data does not fit this narrative?” and include the answer honestly in the narrative. (4) Reduce the story to one sentence.
checklist
- [ ] I have one main insight and I can write it in one sentence.
- [ ] The narrative was structured in a context-tension-resolution structure.
- [ ] Each graph proves a sentence in the narrative.
- [ ] I did not hide data that did not fit the narrative.
- [ ] I did not present correlation as causality.
- [ ] The story can be reduced to a single sentence.