Start here

This is a reusable prompt for an ordinary AI chat that accepts text. It helps you draft colours and redundant cues for a chart you will build yourself.

What to paste: a list of your data series, one per line. Each line is a series label; you may add a short note about what it represents or how it relates to others. Optionally add chart context: chart type, background colour, printed or on screen, and any colour you must keep.

Where to paste it: into your usual AI chat together with the complete prompt text supplied below. No terminal, no account setup and no API key are needed.

How to check the output: the reply should first confirm how many distinct series it found, then give a colour table with columns Series | Hex | Description | Redundant cue | Contrast note, followed by a pairwise caution list, a text-label fallback per series, and a closing boundaries line. Confirm that every series from your list appears exactly once, that each has a cue surviving black-and-white printing, and that weak pairs are named with a fix. If a series is missing, a cue depends on colour, or a hard-to-separate pair is unflagged, treat the draft as failing and ask for a fix.

Known defect in the supplied source prompt you must watch for: in the source's fictional mini example, only four series are listed (reach, clicks, spend, conversions), yet its "Illustrative output shape" and "Check against my example" sections say five rows and "all five series appear". That internal self-check cannot be satisfied by the example as written. The prompt text is preserved unchanged as the original, so ignore the "five" count when it contradicts your own list: the only count that matters is the count of series you actually paste. Readers who need a clean example can mentally correct "five" to "four" for that fictional block.

What the assistant still cannot know: it cannot see your data values, group membership beyond what you typed, your charting tool or your final rendering. If you give no background colour, it will assume white and say so.

Introduction

Colour alone is a fragile way to separate data series, especially for readers with red-green, blue-yellow or low colour vision. This prompt asks the assistant to pair each series with both a colour and a non-colour cue, such as line style, marker shape, a direct label or a fill pattern, and to flag pairs of series that are hard to tell apart. The output is a written design draft you apply yourself in your own charting tool.

Prerequisites, setup and permissions

• A text-based AI chat you already have access to; nothing else is required.
• Your own series list: one line per series, optionally with a short note about what the series represents or how it relates to others.
• Optional chart context: chart type, background colour, print or screen, and any colour you must keep.
• No file access, integrations, background actions, paid capabilities or credentials are involved. You stay in control of what is pasted, so avoid confidential series labels if your chat account does not allow them.

What a correct output checks

1. A series count line matching your own list, plus any open questions instead of silent guesses.
2. A colour table where every series appears exactly once, each row having a hex code, a plain-language description, an explicit non-colour cue and a contrast note.
3. A pairwise caution list naming the pairs most likely to be confused, each with a one-line fix.
4. A text-label fallback giving each series the shortest label that identifies it in a monochrome version.
5. A boundary line stating these are design suggestions to be verified in your own tool and with a colour-blindness simulator.

Limitations

• This is a design draft, not a rendered chart. Nothing here has been executed or rendered by the assistant.
• Contrast notes are advisory; with no background supplied, the prompt assumes white and says so.
• The assistant cannot see your data values, open your charting tool or apply the colours for you, and it must not claim accessibility certification.
• The supplied source prompt contains the four-versus-five example-count defect described above; this is an artefact of the original text, not of your data.

FAQ

Do I need special software, a terminal or an API key? No. Any AI chat that accepts text is enough; the prompt produces only a written draft that you copy into your own design work.

What if my series list is empty, or two series share a label? The prompt tells the assistant to ask you for the list, or to ask you to distinguish duplicates, instead of guessing. It also must not silently add, drop or merge a series.

Is my pasted content private? The prompt itself runs inside whatever chat account you use, so its handling depends on that provider; do not paste confidential labels if your account settings do not allow it.

Verification note

source reviewed; runtime not tested.

Complete reusable prompt

Colour-Blind-Safe Chart Scheme Drafter

You are helping me choose colours and redundant cues for a chart I will build myself. I am pasting a list of data series below. Work only from that list. Do not invent data values, group membership, labels or extra series.

What I will provide

I paste a list with one line per data series, in this shape:

• Series label
• Optional: a short note about what the series represents or how it relates to others

Optionally I may also paste my chart context (chart type, background colour, whether it will be printed or shown on screen, and any colour I must keep).

Your task

1. Read my series list and confirm how many distinct series you found. If any line is ambiguous, ask me one question instead of guessing.
2. Propose a colour scheme in which no two series rely on hue difference alone. Aim for differences in lightness as well as hue, so the set still separates for people with red-green, blue-yellow or low colour vision.
3. For each series, give a candidate colour as a hex code, plus a plain description. Where I must keep an existing colour, work around it.
4. For each series, add a redundant cue that does not depend on colour: line style, marker shape, direct text label, fill pattern, or a short in-figure tag. Give the cue explicitly.
5. Note the contrast between each colour and my background. If I do not specify a background, assume white and say so. Where a colour-versus-background or colour-versus-colour contrast is weak, flag it and offer an alternative.
6. Produce a short fallback plan: if the colours still fail for a particular reader or output, what text label, ordering or annotation makes each series identifiable without colour.

Output format

Return these sections, in order:

• Series count and questions: the number of series, plus any open questions.
• Colour table: columns Series | Hex | Description | Redundant cue | Contrast note.
• Pairwise caution list: pairs of series you judge hard to tell apart, with a one-line fix for each.
• Text-label fallback: for each series, the shortest label or annotation that would identify it in a monochrome version.
• Boundaries: one short line stating that these are design suggestions and that I must verify the final chart in my own tool and with a colour-blindness simulator.

Uncertainty and missing input

If my series list is empty, ask for it. If two series share the same label, ask me to distinguish them before continuing. If I give no background colour, state the white-background assumption in the contrast notes. Do not silently add, drop or merge a series.

Review checks before you finish

• Does every series in my list appear exactly once in the table?
• Does every series have a cue that would survive a black-and-white print?
• Are there any two rows whose hex codes differ mainly in hue, with near-equal lightness? If yes, fix that pair.
• Have you avoided saying anything about how the chart will render in a specific product?

Boundaries

This is a design draft, not a rendered chart. You cannot see my data, open my charting tool or apply these colours for me. Do not claim accessibility certification. Tell me clearly when a judgement is uncertain.

Fictional mini example of the input style

• Ad reach
• Ad clicks, note: always lower than reach
• Ad spend
• Ad conversions
• Background: white. Chart: grouped bar.

Illustrative output shape (not real values)

A table with five rows, each with a hex code and a description such as "mid-dark blue", a redundant cue such as "solid fill plus direct end label", and a note such as "strong against white; keep when printed". Then a pairwise caution list naming reach vs clicks and clicks vs conversions, each with a fix. Then one short text-label fallback per series, e.g. "Reach", "Clicks", "Spend", "Conv." Do not treat these shapes as final content: derive the real rows and cautions from the series list I supply.

Check against my example

For the fictional list above, pass if: all five series appear; no two series are distinguished by hue alone; every row has a printable cue; weak pairs are named with a fix; and a monochrome reader could still tell reach from clicks from spend from conversions from the labels alone. Fail if any series is missing, if a cue depends on colour, or if a pair is left unflagged when it is hard to separate.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.