The state of AI-assisted data visualization

The short version

Capability is moving quickly. Complete evidence is not.

Systems increasingly generate, read, and revise complex visual artifacts. The strongest evidence still depends on context: the exact task, data, grader, human authority, delivery surface, and what happens after release. Use this site to follow those distinctions instead of collapsing them into one score.

Explore the material

Three designed ways into the evidence.

See every section and focused analysis →

Interactive experience

The state of AI-assisted data visualization research

An explanatory field guide to goals, use contexts, techniques, convergence, disagreement, failure evidence, benchmarks, baseline readiness, and the 36-month capability timeline.

Open

Interactive experience

The practitioner and reader experience

What people are trying to accomplish, which tools they reach for, what feels newly possible, and where correction, prepared change, maintenance, trust, and reader experience still break down.

Open

Interactive experience

How close are we to the ideal?

Context-specific scorecards for generation and understanding, reconstructed backward through time and connected to explicit forecasts.

Open

Choose by decision

Start with the work you need to do.

Compare all audience routes →

Visualization practitioners

I make visualizations

For analysts, designers, developers, and graphics practitioners deciding what to delegate, what to inspect, and when to stop or revert.

Open

Analytics and BI leaders

I govern analytics

For BI authors, platform owners, analytics leaders, and governance teams evaluating assistants inside real semantic, permission, and operating environments.

Open

Researchers and evaluators

I study visualization

For visualization, HCI, and AI researchers choosing benchmarks, locating gaps, and designing studies that connect technical output to human consequence.

Open

Newsrooms and public explainers

I publish for readers

For data journalists, graphics editors, newsroom developers, and public-interest communicators evaluating AI-assisted analysis and visual explanation.

Open

Product and system builders

I build AI systems

For product, engineering, and tool teams deciding between prompts, skills, structured intermediates, deterministic components, critics, and browser-level evaluation.

Open

Educators and accessibility leads

I teach or evaluate access

For teachers, curriculum designers, accessibility researchers, and teams responsible for whether intended people can understand and use the result.

Open

Focused questions

Standalone analysis where the subject warrants its own treatment.

These pieces add a distinct argument. Source-form versions of the three designed experiences remain available from those experiences, but are not promoted here as separate destinations.

What the benchmarks actually measureA crosswalk from benchmark scores to the capability demonstrated—and the claims each benchmark cannot support.
Data-visualization skills for coding agentsA comparative reading of skill packages, their techniques, public acquisition signals, paired testing evidence, audience implications, and likely durability as base models improve.
Specialized vision modelsWhere chart-specific perception, OCR, structured reconstruction, and visual critics outperform general models—and where they do not.
Human skills and banked gainsWhat novices, intermediate practitioners, and experts gain today, what remains unproven, and which human capabilities become more important.
Learning when AI can make the chartWhat becomes easier to learn, what can be lost when learners skip construction, and how curricula should rebalance.
What would unlock the next capabilities?Measurable gates, evidence-based timeframes, and the distinction between gains from stronger models and gains from better technique.

Entity library

Know what a benchmark or system actually is.

Use stable reference pages for definitions, primary links, supported claims, nonclaims, related entities, and every place the item appears in this research.

Browse benchmarks Browse systems and methods