2.4Visualisations & charts
The essentials in 30 seconds
- Describe the message, not the type: “show the trend by region” — Copilot chooses, you narrate.
- Iterate in words: colours, labels, sorting, highlighting — conversation replaces menus.
- Judge honesty: zero-based axis, limited slices, non-misleading scale — a chart must tell the truth.
Running mini-case — Hicham and the North that is slipping
Hicham suspects that the North region is declining in “Ventes2025”. Rather than choosing a chart type, he describes his doubt: “Show the monthly sales trend by region, highlighting the North”. Copilot chooses the multi-region line chart and inserts it as a real Excel object; Hicham iterates: “add the labels, sort the legend by descending sales”. Before concluding, he checks that the axis starts at zero — otherwise the “slippage” would be an optical effect. In 2.5, this same question will become a Python regression.
Continuation of the 2.3 running thread: Hicham's column becomes a visual that tells the truth — and continues in 2.5.
1. Describe the message, not the chart
A decisive change of stance: do not ask “make a stacked bar chart”, ask for the message. “Show the sales trend by region over the year” — Copilot chooses the chart type suited to your intent (line, bar, pie). Hicham focuses on what he wants to say; Copilot handles the visual translation. That is the role of a good data-viz assistant: start from meaning, not mechanics.
2. Iterate and refine in words
The proposed chart inserts as a real Excel object — editable normally afterwards. But you can also iterate in natural language: “put the regions in distinct colours”, “add the data labels”, “sort in descending order”. This “I describe, it adjusts” loop replaces dozens of clicks in menus. Hicham gets his visual through conversation, not by hunting for options.
3. The right chart tells the truth
Rigour safeguard: a chart must tell the truth. An axis that does not start at zero, a misleading scale, a 12-slice pie chart that is unreadable — Copilot can produce them if you ask for them. It is up to Hicham to judge visual honesty. Good reflex: ask Copilot “which chart best tells this data?” and challenge its answer. The best visual is the one your audience understands in three seconds, without being misled.
🔎 Concrete example — 🔎 Concrete example — “Show me whether the North is slipping”
Hicham describes his doubt instead of choosing a chart type:
“Show the monthly sales trend by region, highlighting the North.”
Then: “Add the labels, sort the legend by descending sales.”Reading: a multi-region line chart where the North reads at a glance.
Lesson: check that the axis starts at zero before concluding there is a “slippage”.
Section vocabulary
- Describe the message
- Ask for the intent (“show the trend by region”) rather than the technical type (“stacked bar chart”): Copilot translates meaning into a visual.
Simple example: Hicham describes the suspected slippage of the North.
What it is not: a technical type imposed from the start.
Reflex: visual = intent first.
- Real chart object
- The proposed chart inserts as a real Excel chart: colours, axes and labels stay normally editable.
Simple example: Hicham retouches the North line after insertion.
What it is not: a frozen picture that cannot be retouched.
Reflex: inserted = retouchable.
- Natural-language iteration
- Refine with small instructions (“add the labels”, “sort by sales”) instead of re-asking everything or hunting through menus.
Simple example: Hicham adjusts the legend in two sentences.
What it is not: a full rebuild at each retouch.
Reflex: refine = small sentences.
- Visual honesty
- A chart must tell the truth: zero-based axis, non-misleading scale, reasonable number of slices — to judge before distributing.
Simple example: Hicham rejects a truncated axis that dramatised the North.
What it is not: a spectacular but misleading visual.
Reflex: distribute = axis and scale checked.
In practice — In practice — Get an accurate visual starting from the message
- State what you want to SAY with “Ventes2025” (a trend, a comparison, a share).
- Ask for the chart by intent, without naming the type: let Copilot choose.
- Refine in words: colours, labels, sorting, highlighting of a series.
- Check honesty: zero-based axis, reasonable number of slices, non-misleading scale.
Ready-to-say script — ask for a visual without jargon (30 s)
“For Hicham's sales, I did not ask for a chart type: I said “show the monthly trend by region, North highlighted”. Copilot chose the line chart, I refined it in two sentences — labels, legend sorting. And before concluding there was a slippage, I checked the zero-based axis: a chart must tell the truth, not frighten.”
Say while showing the intent-based request then the axis check.
Copy-paste template — visual request frame
PROMPT 1 (intent): “Show the monthly sales trend by region, North highlighted.”
PROMPT 2 (refinement): “Add the data labels and sort the legend by descending sales.”
PROMPT 3 (choice): “Which chart best tells this data? Propose 2 options.”
PROMPT 4 (truth): “Check: zero-based axis, non-misleading scale, readable slices.”Annual trend
Hicham tracks the year: line chart by region, North highlighted, sober labels.
Comparison between cities
“Compare turnover in Algiers and Oran by quarter”: bars chosen by Copilot, sorted.
Market shares
Five regions max in pie slices; beyond that, Hicham switches to sorted bars.
Executive presentation
Message-title dictated at the end: “The North slips in Q3” — the chart asserts.
✓ Best practices
- Always start from the message to get across: the right chart type follows from it.
- Iterate with small instructions (“add the labels”, “sort by sales”) rather than re-asking everything.
- Ask Copilot “which chart best tells this?” then challenge its proposal.
✗ Common pitfalls
- Asking for a precise technical type when another would serve the message better.
- Accepting a truncated axis or misleading scale: the chart must tell the truth.
- Stacking too many series or slices: an unreadable visual communicates nothing.
Key takeaways
- Describe the MESSAGE, not the chart type: Copilot chooses the suited visualisation.
- The chart inserts as a real Excel object, then is refined in natural language.
- Iterate with small instructions rather than re-asking everything.
- A chart must tell the truth: judge visual honesty (axis, scale, number of slices).
48h challenge — apply tomorrow
On your real Table: ask for 1 chart by intent, refine it in 3 sentences, compare 2 options, then validate its honesty (axis, scale) before distribution.
- I can ask for a chart by message, not by type.
- I know how to iterate in natural language (labels, sorting, colours).
- I check visual honesty before distributing.
Frequently asked questions
Does Copilot always choose the best chart?
Often a good one, not always the best. Ask it “which chart best tells this data?” and compare two options before deciding.
Can I edit the chart afterwards as usual?
Yes: it is a standard Excel chart. You can retouch it via the classic menus or keep adjusting it in natural language with Copilot.
How should you ask Copilot for the right chart?