Observe-Only Prompt: Compare Two Charts or Maps

Start here

You need three things from your own material: Visual A (one chart, graph, map or diagram, or a text description of it), Visual B (the second one, or its description), and any context you already have such as axis labels, legend text, units or source notes. A focus question is optional.

1. Open an ordinary AI chat that accepts text.
2. Paste the complete prompt (appended below) as one message.
3. Paste VISUAL A, VISUAL B and your context after it, clearly labelled, then send.
4. If you have only one visual, the prompt tells the assistant to stop and ask for the second.

If an image is blurry, cropped, or you paste only a vague description, the assistant is instructed to move that aspect into a "Cannot compare" list instead of guessing. Context you do not paste — units, periods, category names — will not be invented.

What the prompt produces

The output is a Markdown draft in a fixed order: comparison scope, a differences table with one row per visible difference, a "Cannot compare" list, a "Same or unclear" list, a neutral 2–4 sentence summary, and up to five questions back to you.

Checking the output before you use it

The prompt ends with five review checks for you to run yourself: spot-check at least three table rows by eye; confirm units, periods and category names came from your context; delete any row or sentence that offers a cause, prediction or recommendation; make sure "Cannot compare" is not empty when the visuals are cropped, low-resolution or clearly different subjects; and confirm the summary does not rank one visual above the other.

The prompt also states a fictional worked example (a fictional 2024 bar chart versus a fictional 2025 map, both marked as illustrative, not real data) to show the expected output shape. Treat it as an illustration only.

Privacy and limits

The prompt works only from what you paste. You decide what to upload; avoid pasting confidential figures or personal data into a service you would not otherwise share them with. The prompt cannot open files, browse the web, or send anything on your side — it only writes a comparison draft for you to review. It is observation-only: it will not explain why a difference exists, rank the visuals, or give medical, financial or legal advice.

source reviewed; runtime not tested

FAQ

Can I use text descriptions instead of images? Yes. The prompt accepts "one chart, graph, map or diagram, or a text description of one" for each side. If a description is too vague for a specific aspect, that aspect goes to "Cannot compare".

What if the two visuals cover different subjects? The alignment check asks the assistant to state whether subject, category or geography and displayed period match, and to answer "not determinable" instead of assuming. Aspects it cannot tell apart are listed as unclear.

Complete reusable prompt

You are helping me compare two data visuals that I will paste into this conversation. Work only from what is visible in the two images or descriptions I provide.

INPUT I WILL PROVIDE (paste after this instruction, clearly labelled):

• VISUAL A: one chart, graph, map or diagram, or a text description of one.
• VISUAL B: the second visual, or a text description of one.
• OPTIONAL CONTEXT: axis labels, legend text, units or source notes I am pasting myself. Do not guess these.
• OPTIONAL FOCUS: the specific comparison question I care about, if any.

YOUR TASK: Produce an observation-only comparison. List what is visually different between A and B. Do not explain why a difference exists, do not infer causes, trends, intent, quality or consequences, and do not add outside facts. If I explicitly ask for interpretation later, treat that as a separate request.

HOW TO COMPARE (in this order):

1. Alignment check: state whether the two visuals appear to cover the same subject, same category or geography, and same displayed time or reference period. If any of these cannot be determined from the visuals, say 'not determinable' rather than assuming.
2. Structural differences: chart or map type, orientation, number of panels, grouping and ordering of categories.
3. Scale and axis differences: axis ranges, tick intervals, stated units, log vs linear where visible, colour-scale direction, and whether scales are shared or independent.
4. Data-mark differences: which categories, regions or series are present in one and absent in the other; differences in bar/line/dot positions or counts that are plainly visible.
5. Label and annotation differences: titles, subtitles, legends, footnotes, source lines and any on-figure callouts, including ones present in only one visual.
6. Presentation differences that could mislead a casual viewer: truncation, missing zero baseline, colour choices, dual axes, or inconsistent rounding of printed values.

OUTPUT FORMAT (Markdown, in this order, no preamble):

• Comparison scope: one or two sentences on what is being compared and the focus question if supplied.
• Differences table: columns = # | Aspect | Visual A | Visual B | Difference type (structural / scale / data-mark / label / presentation). One row per difference. Include only differences you can actually see.
• Cannot compare: bullet list of things you could not verify from the supplied visuals or context, each with the exact reason (for example 'tick spacing unreadable in Visual B').
• Same or unclear: bullet list of aspects that look the same, plus any aspects you cannot tell apart, marked as unclear.
• Neutral descriptive summary: 2 to 4 sentences that state the visible differences without causes or judgement.
• Questions for me: up to 5 short questions whose answers would let you resolve an 'unclear' item, for example missing axis label text.

RULES AND BOUNDARIES:

• Never claim a difference you cannot point to in the supplied visuals or context.
• Never state or imply popularity, traffic, quality ranking or importance.
• Never give medical, financial or legal advice, and never recommend a decision based on the comparison.
• If a visual is too low-resolution, cropped or described too vaguely to compare a specific aspect, move that aspect into 'Cannot compare' instead of guessing.
• Do not fabricate source names, dates, authors or units. Use only what I supply.
• Do not perform any external action: you are only producing a written comparison draft for me to review. You cannot open files, browse, or send anything.
• If I supply only one visual, stop and ask for the second before proceeding.

REVIEW CHECKS BEFORE I USE THE OUTPUT (state these at the end so I can verify):

1. Every difference row should be checkable by eye in the pasted visuals; I will spot-check at least three rows and remove any I cannot verify.
2. Confirm units, periods and category names were taken from my supplied context, not invented.
3. Check that no row or sentence offers a cause, prediction or recommendation; if one slipped in, delete it.
4. Confirm the 'Cannot compare' section is not empty when the visuals are cropped, low-resolution or clearly from different subjects.
5. Verify the summary is purely descriptive and does not rank one visual above the other.

WORKED FICTIONAL EXAMPLE (illustration only, not real data): INPUT: VISUAL A is a fictional bar chart titled 'Bike lane counts, District 4, 2024' with a y-axis starting at 0 and five labelled streets. VISUAL B is a fictional map titled 'Bike lane counts, District 4, 2025' with the same five streets shown as shaded circles on a 0 to 500 scale. FOCUS: do the two visuals show the same streets? EXPECTED OUTPUT SHAPE: the scope line names both visuals and the focus question; the table has rows such as 'visual type (bar chart vs map)' under structural, 'scale representation (axis vs colour scale)' under scale, and 'street names present (five listed vs five mapped, match to be verified from labels)' under data-mark; the 'Cannot compare' list notes that exact count values in Visual B cannot be read because the legend is not supplied; review check 1 asks me to confirm the five street names appear in both.

Begin by asking me to paste VISUAL A, VISUAL B and any context, then produce the output above.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

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