TaskbyTask.Practical AI guides for work
ChatGPT, Claude, Copilot or Gemini · Practical walkthrough

Ask AI for a chart people can understand without colour alone

The chart looks great in teal and lime, until someone prints it in black and white or can't tell the two greens apart. Ask for labels and a sentence that carry the meaning, so colour is a bonus rather than the key.

Task by Task6 min read + practice

Example output

A chart with readable values and labels

Example chart (fictional September requests): Email 12, Chat 18, Phone 9. Every value is written on its bar; colour only supports it.

Email
12
Chat
18
Phone
9

Fictional request counts · direct labels retain the measure.

Read the prompt
Create a chart answering [specific question] from this supplied dataset.
Use [fields], [units/time period] and [missing-value rules]. Preserve
source values and definitions. Use direct series/category/value labels
and a readable text explanation plus source table. Do not rely on colour
alone for meaning. Keep qualifications and unknowns visible; no invented
trend, cause or benchmark. Provide the actual editable file/source and
explain what was calculated, not just a screenshot of a chart.

What you'll make

a chart with values and labels written on it, a one-sentence text alternative, and the source table, placed where readers will find them.

What you'll need

a small dataset you’re allowed to use, what each column means, and a chat that can make files: ChatGPT, Claude, Copilot or Gemini. Excel, PowerPoint or Google Sheets for the final placement. About fifteen minutes.

1. Ask for a chart that works without colour

Attach or paste the data and ask:

Create a chart answering [question] from this data. Use [fields] and state the units and period. Keep the values exactly as they are.
Label every bar or line directly with its category and value, so nobody needs the legend or the colours to read it. Use colour only as support.
Also give me a one-sentence text alternative and the source table. Do not add a trend, cause or benchmark that the data does not show.
Give me the chart as an editable file, not only a picture.

[question] is what the reader should learn, such as “which channel had the most requests in September”. [fields] are the column names.

Why this works: direct labels move the meaning out of the colours and into words, which survive printing, colour-blindness and a quick glance.

What you should see: a chart where every value is written on it, a sentence, and a table.

2. Read the text alternative with the chart hidden

Cover the chart and read only the sentence and the table. Can someone understand the main point, and the numbers behind it? A good text alternative names the biggest and smallest values and any important change, in one or two sentences, with the table for the detail.

Why this works: the text alternative is what a screen reader announces and what remains when an image fails to load. If it doesn’t stand on its own, the chart isn’t accessible however good it looks.

What you should see: a sentence such as “Chat had the most requests in September (18), up from 11 in August; Email and Phone both fell slightly.”

3. Put the text where readers will find it

A label in a chat answer doesn’t travel with the chart. Add the text alternative and table in the place the chart will be used:

Where the chart goes Where the text goes
Excel or PowerPoint (Microsoft 365) Right-click the chart border › View Alt Text; put the table on the same sheet or in the notes
Google Sheets or Slides Select the chart › More (three dots) › Alt text
A web page or report The sentence in the body text beside the chart, with the table below it

Why this works: the alt text field is what assistive technology reads. A sentence elsewhere in the chat doesn’t reach anyone.

What you should see: the sentence in the alt text field, and the table next to the chart.

Check the result

  • Every value on the chart matches the source table.
  • Units and period are written on the chart or in its title.
  • You can read the chart in greyscale, without the legend.
  • The text alternative is in the file’s alt text field, not only in the chat.

The usual failure: the chart still depends on colour. Two series in two shades with a legend in the corner, and no values on the bars, look tidy and tell a colour-blind reader nothing. Spot it by switching the chart to greyscale, or printing it, and trying to read it.

The chart relies on colour and a legend to tell [series] apart, and the values are not shown. Label each bar or line directly with its series name and value, keep the source values unchanged, and give me the one-sentence text alternative and table again. Do not add any trend or cause.

[series] is what the colours were separating, such as “August and September”.

Worked example: two months of support requests

Fictional example. A team lead pasted this into a chat:

channel,august,september
Email,14,12
Chat,11,18
Phone,10,9

The first chart used teal for August and lime for September, with a legend and no values on the bars. In greyscale the two months were indistinguishable, and the summary said “Chat is growing fast”.

  1. The colours were doing all the work. Without the legend, a reader couldn’t tell which bar was which month, and couldn’t read a single value. The repair prompt added “August 11” and “September 18” style labels to each bar.
  2. “Growing fast” wasn’t in the data. Two months show a change, not a trend. The new text alternative says what the table says: “Chat had the most requests in September (18), up from 11 in August; Email fell from 14 to 12 and Phone from 10 to 9.”
  3. The totals reconcile. August: 14 + 11 + 10 = 35. September: 12 + 18 + 9 = 39. The September bars match the example chart at the top of this page.

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

Go deeper

Keep missing and zero distinct

A blank cell shouldn’t quietly become a zero-height bar. Say what a missing value means before you ask for the chart, and label it on the chart. The missing-months guide shows how that one assumption changes the analysis.

Don't let the chart explain why

Counts show what happened, not why. Ask for a description, and keep any explanation labelled as a guess. See explain a chart without inventing.

Fix a misleading label without touching the data

If a title or axis label overstates the chart, change the words, not the numbers. The chart-label guide covers it.

Sources

Optional analytics

With your permission, Google Analytics uses cookies to measure visits and which guides people read. It stays off until you accept. Use Analytics choices in the footer to change your choice. Rejecting after accepting refreshes this page. Privacy details.