Seedlabs

ChartCritic: One-Click Design Feedback Inside Chart Editors

A plugin for chart editors that suggests concrete design fixes to reduce clutter and improve color.

A plugin for chart editors like Datawrapper, Flourish, and Observable that acts like a design mentor: it reads the rendered chart, measures salient properties like hue count and clutter, and returns a short ranked list of concrete fixes before the chart ships.

Computer ScienceData Visualization and Analytics
Cognitive-Adaptive Visual AnalyticsData-journalism and BI charting tools — embedded AI design feedback in chart editors.

Concept

A one-click "Critique my chart" button embedded in browser-based and desktop chart editors (Datawrapper, Flourish, Observable, Tableau/Power BI extensions). The pipeline mirrors the demonstrated system: (1) the editor renders the current chart to an image; (2) a suite of perceptual filters extracts salient metrics from that image (e.g., distinct hue count, contrast, mark density, aspect ratio); (3) a curated preamble of visualization design guidelines is prepended; (4) an off-the-shelf LLM returns a ranked list of specific, plain-language suggestions, each tied to a guideline and a measured metric ("9 distinct hues detected; guideline suggests 6 or fewer for categorical encoding — consider grouping or direct labeling").

Two deliberate design choices keep it simple and safe: it is image-in/text-out, so it never touches the underlying dataset (sidestepping data-privacy objections in newsrooms and enterprises), and it suggests rather than auto-edits, preserving designer agency — important because the evidence shows even experts used the guidance to refine their work rather than obey it.

Why now

  • Paper [0] demonstrates precisely this mechanism with an off-the-shelf LLM (ChatGPT): a preamble of visualization design guidelines plus perceptual filters extracting salient metrics from the chart image produced actionable, customized design feedback.
  • The longitudinal study covered 13 designers — 6 novices, 4 intermediates, 3 experts — each authoring a new visualization from scratch over several days, and found the natural-language guidance aided even seasoned designers in refining their visualizations. That means the buyer base is not just beginners: professional desks get value from a fast first-pass reviewer.
  • [0] states the gap explicitly: interactive editors let users author visualizations without code but offer no guidance on the craft of effective visual communication — a gap a thin plugin can close without re-architecting the editor. Integrating guidance directly into the tool, matched to the user's current purpose, is the integration pattern argued for in the visual-analytics guidance literature [6].

Written by GLM · 9 Oct 2026

AI assessment

Backed by 2 papers88

A highly focused, feasible plugin that leverages a proven research pipeline to provide non-invasive design critiques for professional charting tools.

Evidence strength
5/5
The idea directly implements the 'Visualizationary' system described in Paper [2], utilizing the exact same pipeline of perceptual filters and LLM preambles.
Market pull
4/5
Data journalists and BI analysts have a high urgency for accuracy and aesthetics, and the 'image-in/text-out' approach solves the primary enterprise hurdle of data privacy.
Novelty & moat
3/5
While similar to 'Visual Bias Guard', this focuses on aesthetic/perceptual design guidance rather than just error detection, filling a gap in craft-based feedback.
Feasibility
5/5
The MVP requires only a browser extension or API integration to send a screenshot to an existing LLM with a predefined prompt and basic image filters.
Wedge clarity
5/5
The 'Critique my chart' button is a sharp, single-purpose entry point that provides immediate value without requiring a platform shift.
Simplicity / focus
5/5
The product is strictly limited to a feedback loop (image-in, text-out), avoiding the complexity of auto-editing or data integration.

Scored by Gemma 4 · 9 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • Datawrappercompany

    Its editor is used by journalists who often lack formal design training; an embedded critique button built on the demonstrated preamble-plus-perceptual-filter pipeline extends its 'publish-ready' positioning, and the image-only design means no user data ever leaves the newsroom.

  • Flourish (Canva)company

    The tool targets non-expert storytellers; automated design feedback narrows the gap between amateur drafts and professional output, reducing rework and support load.

  • Observablecompany

    Notebook users typically iterate on charts alone; a 'critique cell' gives them a reviewer when no colleague is available, and the image-in/text-out architecture requires no access to their underlying data.

  • The New York Timescompany

    Its graphics desk is expert-heavy, and [0] found LLM guidance aided even seasoned designers, so the tool works there as a quick first-pass reviewer before human edit rather than as a beginner crutch.

Research it builds on

  1. Providing visual analytics guidance through decision support
    Wenkai Han, Hans‐Jörg Schulz · 2023 · 16 citations
    All ideas from this paper →
  2. Visualizationary: Automating Design Feedback for Visualization Designers Using Large Language Models
    Sungbok Shin, Sanghyun Hong, Niklas Elmqvist · 2025 · 7 citations
    All ideas from this paper →
Leave feedback
feasibility