Make AI account for overlapping segments in a chart

Example result.

Note under the chart: “People could choose more than one workflow, so the shares add up to more than 100%.”

1. Say whether the groups can overlap

Before asking for a chart, say whether one person can belong to more than one group. Separate column names don’t make groups exclusive. The everyday case is a “select all that apply” survey question: if 62% of staff use Teams and 48% use Slack, the shares add up to 110% and both numbers are right. They just can’t be slices of one pie.

Upload the counts or records and tell the model the rule:

Chart these group counts for [population and period].
The groups [can / cannot] overlap, because [definition].
Use the total population as the denominator for shares.
Don't add overlapping counts and call the sum unique people or a whole.
If you split the groups into exclusive parts, show how each part was
calculated. If the overlap isn't in the data, say so; don't invent it.

[population and period] is who was counted and when, such as “30 survey respondents, September”. [can / cannot] and [definition] state the rule, for example “can overlap, because the question allowed several answers”.

Why this works: the model’s default is to treat counts as parts of a whole. Stating the rule first changes which chart it reaches for.

What you should see: a reply that repeats the overlap rule, names the denominator and either uses the overlap from your data or says it’s missing.

2. Choose a chart that fits the overlap

Side-by-side bars show each group against the whole population without implying the groups are exclusive. A labelled Venn diagram shows the shared membership. A pie works only when the parts are exclusive and together make the whole.

Ask the model to explain its choice before you accept it:

Before drawing the chart, tell me its denominator and whether it assumes
the groups are exclusive. Show any calculation of "both", "only" or
"neither" from my numbers. Don't use a pie unless the parts are exclusive
and cover the whole. If you draw a Venn diagram, say whether the circle
sizes are to scale.

You don’t need AI for the chart itself. In Excel, select the group names and shares, choose Insert › Recommended Charts and pick a bar chart. In Google Sheets, choose Insert › Chart, then under Setup › Chart type pick a bar chart.

Why this works: a stated denominator is checkable. A chart without one leaves the reader to guess which whole the percentages belong to.

What you should see: a denominator, a chart type that doesn’t imply a whole, and any derived numbers with their working.

3. Read the chart as your audience will

Open the exported chart and read only what’s on it: labels, title and note. Would someone who hasn’t seen the data think the groups add up to the population? Direct labels and a short note under the chart do more than a colour key here.

Why this works: the error usually lives in the picture, not the numbers. Reading the chart cold is how you catch it.

What you should see: labels with counts or shares, a stated population and a note when shares add up to more than 100%.

Check the result

• Each group’s share uses the total population, not the sum of the groups.
• “Both”, “only” and “neither” add up to the population.
• No pie or 100% stacked bar treats overlapping groups as slices.
• A note explains why shares can add up to more than 100%.

The usual failure: a pie chart of the raw group counts. It adds the groups together, uses that sum as the whole and counts everyone in both groups twice. Spot it by adding the slice counts: if they come to more than the number of people, the chart is wrong.

The chart treats [overlapping groups] as exclusive and uses
[wrong denominator] as the whole. The data has [population and overlap].
Correct the counts, the denominator and the chart type together.
Keep my numbers unchanged and show how you worked out each part.
Don't invent a missing overlap or force the shares to add up to 100%.

[overlapping groups] are the groups the chart split wrongly; [wrong denominator] is the total the chart used.

Fictional example. A team survey asked: “Which shared workflows do you use? Select all that apply.”

Respondents: 30
Marketing workflow: 20
Finance workflow: 15
Both: 8

The first draft was a pie: marketing 57%, finance 43%, out of a whole of 35. There are only 30 people, so eight of them had been counted twice. It also hid the three people who use neither.

The repair prompt, filled in with “marketing and finance workflows”, “35” and “30 respondents, 8 in both”, produced this:

The chart became the bar chart in the example result, with the note about shares above 100%.

• The pie’s whole was 35. Adding the slice counts and comparing them with 30 is a 10-second check that catches it.
• Neither is a real group. Three people used no shared workflow; the pie had no slice for them.
• The union is 27, not 35. 20 + 15 − 8 = 27 people use at least one workflow. That only works because the survey recorded “both”; without it, the overlap is unknown.

Task kit for this job: Download (ZIP) · Reusable instructions

Go deeper

Sources

• Create a chart from start to finish (Microsoft Support)
• Add and edit a chart or graph (Google Docs Editors Help)
• Work with files in ChatGPT (OpenAI)
• Create and edit files with Claude (Claude Help Center)
• Complex images (W3C Web Accessibility Initiative)