Data Infographic
Turn materials into a vertical infographic where relationships are immediately clear and every claim can be traced back to its source. Communicate in the user's conversational language, and carry over the already confirmed audience, visual reference, and delivery format from the current project.
First, Confirm the Topic Selection Method and Target Language
For each new project, confirm the following two items in the same round. If the user has already specified, adopt those directly and only ask about what has not been decided; then wait for the user's response. Proceed to analyze materials, check colors, or draw only after both items are resolved.
- How to select the topic? Do I pick a key point from the materials and create the graphic directly, or should I first give you a few directions, let you choose, and then I supplement the data and cases?
- What language should the final output be in? For example, Simplified Chinese, Traditional Chinese, English, or another language. You can also say "same graphic in both Chinese and English" or "one copy in Chinese and one in English."
You can answer in one sentence, e.g., "You decide on the topic, and make the graphic in English."
For example, if the user sends English materials in Chinese but only says "You decide on the topic," then only ask about the output language. If the user also leaves the language to your decision, choose based on the target audience; when the audience's language is unknown, use the language of the conversation. When modifying, extending, or exporting an existing graphic, reuse the original topic selection method and output language, and adjust according to the user's new requirements. If materials are missing, request them in the same round.
Write titles, captions, legends, axes, unit names, and source descriptions in the target language. When using templates, write the target language into the lang field, and use that language for page footer titles and expand buttons. Keep proper names and original sources in a recognizable form; mark translated direct quotes as translations. For bilingual graphics on the same page, follow the order specified by the user. When producing separate graphics for each language, all language versions share the same set of values, sources, and timestamps.
Select the Topic Based on the Materials
First, identify what the materials are, what they are discussing, and what units are appropriate for description. Articles, interviewees, feedback tickets, chapters, and table rows each have different meanings. For large volumes of text, first build a list by source, read in batches, and summarize the evidence, keeping the original location of each finding. For specific methods and optional chunking tools, see Large Text.
- Generate directly. Identify a main question that has sufficient evidence and is valuable to the reader, then explain it using comparisons, changes, distributions, or specific examples. Routine content and visual trade-offs are handled by the Agent.
- Provide directions first. Typically propose 3–5 distinct options, adjusted according to the richness of the material. For each option, give a reader question, the actual scope it covers, one representative piece of evidence, and a feasible graphic. For example, a user survey could compare the most frequently mentioned issues across different groups, or it could follow a complete experience as a flowchart. Inform the user which materials support comparisons and which are better suited as case studies.
- Wait for the user to select before producing. Support multiple selections, self-defined directions, and exclusions. Use the chosen questions to structure the sections; each additional module should help answer one of the chosen questions. If the user leaves the remaining decisions to the Agent, the Agent may fill them in independently. Only ask the user to decide again if the chosen direction truly lacks material and a change of topic is needed.
Draw the Relationship First, Then Write the Explanation
When a visual example is available, first look at the position and size of the main chart, which comparisons share a scale, and where annotations are placed. Apply this reading order to the current topic. First, draw a main chart for the core question that can answer it on its own, placing it below the title so that the data or relationship is visible on the first screen. Only when there is an archive size, place counts such as number of articles or duration in a short range bar.
- Rich materials default to a multi-chart report. First, write down a few things a reader would naturally ask after the main question, and pair each question with verifiable data and its source. Long-form investigations or interview collections often support about 5–8 analytical charts with independent findings. Use this range to detect gaps; the actual number depends on the evidence and how deep the user wants to go. A single interview may only need one relationship chart and one direct quote. If the user says "add some more data charts," return to the original materials to add comparisons not yet answered.
- Organize the reading order by questions. On the first screen, establish the topic with a result from a clear source, indicating whether it belongs to a case study or the entire population. Below, use adjacent charts to answer complementary questions, such as "how many mention" and "which topics are mentioned together," "how much is saved" and "where it is used," or "how much do the two groups differ" and "how stable is the estimate." Only give shared scales to comparisons with the same scope. Use a well-positioned case study to explain the difference the reader just saw. Titles, total article counts, big numbers, quotes, and repeated graphics of the same data do not count as new findings.
- Let the material type determine the evidence. For collections, prioritize statistics from the original records; all topic counts and trace-back lists are generated from the same detailed dataset. Write the denominator, multiple selections, missing data, or non-matches next to the chart, along with the meaning of self-reported, expected, measured, assumed, and interval values. Within the user-approved scope of materials, if the original text only provides a case study but the reader also needs a measure of prevalence, you may supplement with directly related original surveys, experiments, or public tables—noting the sample, time period, and table number—and present them in a separate chart. When the user specifies using only the given materials, find distributions, groupings, and exceptions within the materials themselves.
- Interviews can also be visualized. Identify who gives what to whom, at which node the path branches, and how two approaches differ, using equal-width lines, step diagrams, or side-by-side comparisons. When there is a direction but no magnitude, use arrows with short labels like "increases." Clearly mark the scope of case studies, surveys, and self-reported estimates. Only data that can be compared horizontally should share a common scale.
- Distinguish between chart captions and in-chart labels. Use the title to point out the question or finding. The caption below the title is usually one sentence, explaining what a dot, a line, or a color represents. Add a second sentence if needed to highlight a key difference. Follow the column width and line count from the user's example; for Chinese, you can start with about 12–30 characters, typically one to two lines. Simple charts where labels already make everything clear may omit the caption. For example, "Each dot is an event; the black line marks this year's median." Next to icons and nodes, place only short category or action labels, such as "for client" or "trace root cause," with values including units. Keep the denominator and necessary qualifications where the reader looks at the numbers.
- Validate with the final output. Open the full-size thumbnail of the infographic, then view the longest labels and densest matrix at mobile width. Pick any one cross-cell, recalculate using the detailed data, click through to check the original source, and verify the record count and location. Check the exported PNG and PDF to ensure that the denominator, values, units, main sources, and method are fully readable even in static format. Go through each chart and state its unique finding; merge charts that answer the same question.
Bring the Topic's or Creator's Color Palette into the Final Output
For new graphics, after the topic is determined, default to using available web search and browsing tools to actually open public pages related to the topic or creator. For named bloggers, first check their personal page, series covers, or public visual style guide; for general topics, compare the reports, charts, or infographics from different publishers in that field to find recurring colors. If the user specifies a palette or requires offline work, follow that request. When modifying a finalized graphic, reuse the original palette; search again only when changing topics or bloggers or when a different color is requested.
Select primary and accent colors from the actual materials that suit this graphic, and map them to titles, indicators, bars, lines, and matrices. Save the color reference URLs and the basis for picking the colors along with the graphic; export them together with HTML, PNG, and PDF. For retrieval methods, palette fields, readability, and handling when no palette is found, see Online Color Palette Retrieval. This step is performed for both topic selection modes, and it usually does not require the user to select colors again.
From Evidence to Final Graphic
- First verify the statistical unit, deduplication, time range, denominator, and missing data, then write the totals and proportions. Limit sampling conclusions to the materials actually read. Qualitative content is suitable for case comparisons, timelines, classifications, or flowcharts. When numeric values are insufficient, these graphic types can express the original logic.
- Ensure that key conclusions can be traced back to a table range, report page number, article section, feedback ID, or recording timestamp. For Clipto share links, use the corresponding connector or shared transcription skill to read the collection and segments. Note missing data in the coverage description.
- Use verifiable data. When the user explicitly permits demonstration data, you may simulate it, but indicate this on the first screen or in the corresponding module. Quotes from identifiable individuals and actual performance metrics must have a source. Archive web color references and factual evidence separately.
- Default to vertical layout. When extending, add graphics that reveal new relationships or case comparisons. Closely related small graphics can be placed side by side; main charts and wide matrices can span columns.
Prefer following existing design files when available. The accompanying template uses lead for the main chart on the first screen and sections for subsequent comparisons. When the format is not specified for a multi-chart report, the default delivery is a self-contained HTML for trace-back, a shareable high-resolution PNG, and a single-page vector PDF. When the user specifies a format, follow it. For interview or feedback collections, you can start from the Reader Letters Multi-Chart Example. The template supports trace-back per cell, two-group shared-axis point charts, and intervals with point estimates. Template Fields and Exporting