Chartjunk names the category of visual elements that consume display space, attention, and cognitive processing without delivering data. Tufte coined the term in 1983 to describe the gridlines, hatching patterns, decorative borders, three-dimensional effects, drop shadows, and pictorial embellishments that had become standard features of statistical graphics in newspapers, corporate reports, and government publications. His argument was not that these elements were merely unattractive — though they usually were — but that they actively degraded the viewer's capacity to extract evidence from the display. The human perceptual system cannot distinguish data from decoration automatically; it processes all visual elements with roughly equal initial attention. When the decoration dominates, the signal is drowned. Applied to AI, chartjunk describes any element of generated output whose function is to perform fluency rather than transmit information.
The taxonomy of chartjunk that Tufte developed is precise. The worst category is moiré vibration — the optical effect produced by densely patterned fills, such as cross-hatching used to distinguish bars in a chart. The patterns create visual noise that interferes with the eye's ability to compare values. The second category is the grid — the gridlines that many spreadsheet programs produce by default, which add visual weight without adding information when the axis labels already encode the relevant values. The third is the duck — Tufte's term for graphics in which the decorative pictorial element dominates the data, borrowed from a Long Island building shaped like a duck that sold duck eggs and meat. A three-dimensional rendering of oil barrels for a chart about oil prices is a duck: the picture has overwhelmed the information.
Contemporary chartjunk has evolved. The drop shadows and gradient fills of 1990s PowerPoint have largely disappeared, replaced by flatter aesthetic conventions that achieve higher data-ink ratios on average. But new forms have emerged: the infographic with its cartoon characters and narrative arcs that buries the data beneath storytelling; the dashboard with its gauges, dials, and speedometer metaphors that add decorative machinery around numbers that would be clearer as plain text; the animated transition between chart states that prioritizes the viewer's engagement over the viewer's comprehension.
Applied to AI-generated output, chartjunk describes a subtler category. The model has learned from its training corpus that expert communication carries certain surface markers: qualifying hedges, structured enumeration, rhetorical framing that signals intellectual seriousness. Some of these markers carry information. Many are pure performance — verbal chartjunk that consumes the reader's attention without advancing the argument. The AI that produces a three-paragraph introduction before addressing the question, or the summary that restates what the preceding text has already established, is generating chartjunk at the level of prose structure.
The diagnostic question Tufte applies to graphics transfers directly: if the element were removed, would the communicative function of the display degrade? If the answer is no, the element is chartjunk. Applied to prose: if the sentence were deleted, would the reader lose information? If not, the sentence is verbal chartjunk. The discipline of output interrogation includes the reflex of asking this question of every paragraph AI produces.
Tufte introduced chartjunk in Chapter 5 of The Visual Display of Quantitative Information (1983), dedicating extensive analysis to specific examples from newspapers, academic journals, and government reports. The term was instantly canonical — by the late 1980s it had entered the vocabulary of statisticians, journalists, and designers, and remains the standard term for the category of visual failure it names.
Moiré, grid, duck. Tufte's three-part taxonomy covers most of what degrades statistical graphics: optical interference from patterns, unnecessary grid structures, and decorative elements that dominate the data.
The perception problem. Viewers cannot sort data from decoration automatically; the sorting is cognitive work, and high chartjunk raises that work prohibitively under time pressure.
The default problem. Most chartjunk is produced not by malicious designers but by software defaults that embed decorative elements as standard features. The designer must actively remove what the tool automatically adds.
Verbal chartjunk. The category extends beyond visual displays to any communicative medium where structural or decorative elements compete with substantive content — including AI-generated prose that performs expertise without transmitting it.
Cultural reinforcement. Chartjunk persists because it signals effort and sophistication in contexts where simple accurate displays would signal inadequate investment. The three-dimensional chart looks more expensive than the flat bar graph, and the decoration is read as diligence.