03 · What You Need to Know
What AI Can Contribute to Scientific Visualization
What Counts as a Research Figure?
A research figure is a visual representation used to communicate scientific information, including data, analytical results, conceptual relationships, or methodological processes.
Figures may include scatterplots, histograms, box plots, line graphs, forest plots, flowcharts, conceptual diagrams, and other visual formats.
Not all figures make the same evidential claim. A scatterplot represents measured observations or derived statistics, while a conceptual diagram may communicate a proposed theoretical relationship.
Consequently, the appropriate use of generative AI depends on whether the figure represents empirical results, analytical procedures, or conceptual information.
What Is the Difference Between AI-Generated Graph Images and AI-Generated Plotting Code?
This distinction is fundamental to research integrity.
A general-purpose image generator may produce a picture resembling a chart without computing the graphical elements from a dataset. Such an image may contain invented values, inconsistent axes, or incorrect proportions.
By contrast, an AI coding assistant can generate instructions for statistical software to plot actual observations or verified summary statistics.
AI-rendered chart image
A visual image generated from a prompt, potentially without a reproducible numerical relationship to the research dataset.
AI-assisted data-driven graph
A visualization produced through executable code or analytical software using identifiable data and documented transformations.
Code-based plotting does not automatically guarantee accuracy. AI can write incorrect code, select the wrong variables, or calculate inappropriate statistics.
Nevertheless, a reproducible workflow makes the source of each plotted value easier to inspect and correct.
What Types of Research Figures Can AI Help Create?
AI assistance can support several forms of scientific visualization.
| Figure Type |
Potential AI Assistance |
Primary Verification Need |
| Bar chart |
Generate plotting code for group summaries. |
Verify the statistic, group labels, scale, and uncertainty indicators. |
| Scatterplot |
Plot observations and optional fitted relationships. |
Verify data coordinates, transformations, and model specifications. |
| Line graph |
Visualize repeated observations or ordered measurements. |
Check time ordering, aggregation, and whether connecting observations is justified. |
| Histogram |
Display the distribution of a numerical variable. |
Verify binning, frequencies, and data exclusions. |
| Box plot |
Summarize distributions across groups. |
Check quartile definitions, whisker conventions, and outliers. |
| Forest plot |
Display effect estimates and confidence intervals. |
Verify effect sizes, intervals, reference lines, and scales. |
| Conceptual diagram |
Arrange theoretical constructs and proposed relationships. |
Check whether arrows and labels reflect the actual framework. |
| Research workflow |
Organize procedural stages into a diagram. |
Verify the sequence and description of the actual methodology. |
AI can also assist with formatting and accessibility, including figure captions, labels, and alternative descriptions. These tasks still require verification against the intended scientific meaning.
Can ChatGPT Create Graphs From Research Data?
Some ChatGPT configurations can analyze uploaded datasets and generate visualizations using computational tools.
Researchers may also use ChatGPT to generate code for Python, R, or other statistical environments.
However, the availability of particular capabilities depends on the product configuration and tools accessible during the session.
Regardless of the interface, the researcher should verify the dataset, variable selection, calculations, and graphical output.
A chart appearing directly in a conversation should not be assumed correct merely because it was produced from an uploaded file.
Why Can AI Produce a Graph That Looks Correct but Is Numerically Wrong?
Generative AI may misunderstand variable names, select incorrect columns, aggregate observations improperly, or introduce numerical values not supported by the data.
For example, a researcher might request a graph of average student satisfaction by academic program. The model could mistakenly calculate percentages instead of means or include missing responses as zeros.
The resulting chart might look entirely plausible.
Other errors involve plotting correct values against incorrect labels or using a transformed scale without explaining it.
The possibility of AI-generated graphs plotting incorrect values means researchers should examine both the numerical calculations and the final visual representation.
Can AI Invent Data While Creating a Figure?
Yes. A model may generate values when the prompt does not provide sufficient information, particularly when asked to produce a complete example or visually convincing chart.
For instance, requesting a graph showing how an intervention improved student performance may lead a model to create illustrative scores if no actual dataset is supplied.
Such values can be appropriate in an explicitly hypothetical teaching example, but they must not be presented as empirical research findings.
Researchers should distinguish genuine observations, calculated statistics, simulations, and invented illustrative values.
The concern about AI inventing data in research visualizations is particularly serious when synthetic values become indistinguishable from authentic results.
Can AI Select the Most Appropriate Graph?
AI can recommend chart types based on the research question, variable types, and analytical purpose.
For example, a scatterplot may be suitable for examining the relationship between two continuous variables, while a box plot may help compare distributions across categories.
However, selecting an appropriate graph requires understanding the measurement scale, research design, data distribution, and intended inference.
A bar chart of group means may obscure important distributional variation. A line graph may imply continuity or ordering that the data do not support.
The separate task of using AI to choose an appropriate research graph requires methodological judgment rather than simply selecting a familiar chart type.
Can AI Generate Figures Directly From Raw Data?
Yes, when suitable computational tools are available. AI can assist with reading datasets, preparing plotting instructions, and executing code that produces visualizations.
However, researchers must inspect how the data were processed.
For example, a graph of average test scores requires decisions about missing values, group membership, weighting, and the statistic being displayed.
If these decisions are made incorrectly, the plotted graph may misrepresent the study even when the code executes successfully.
Researchers should establish the accuracy of AI-created figures derived from raw research data through reproducible calculations and source-level checks.
Can AI Make Graphs Visually Misleading Without Changing the Data?
Yes. Visual design choices influence how readers perceive differences, trends, and uncertainty.
A bar chart with a truncated vertical axis may exaggerate relatively small differences. A three-dimensional chart may distort the apparent magnitude of categories. Unequal axis intervals can misrepresent the spacing between observations.
Color intensity, line thickness, and annotation placement can also draw disproportionate attention to particular results.
These choices may be misleading even when the underlying numerical values are accurate.
The risk of AI-generated visualizations creating misleading emphasis should therefore be considered separately from numerical accuracy.
Should All Bar Charts Start at Zero?
Bar charts encode magnitude through length, so a zero baseline is generally important when the bar length represents the measured quantity.
Truncating the axis can exaggerate differences in the apparent lengths of the bars.
However, other graph types may appropriately use nonzero axis limits. For example, a line graph displaying small changes in a physiological measurement may use a restricted vertical range when that choice is clearly communicated and scientifically justified.
Researchers should evaluate how the graphical encoding affects interpretation rather than applying a single rule indiscriminately to every visualization.
Can AI Add Statistical Significance Indicators to Figures?
AI can help place significance annotations, confidence intervals, or other statistical information on graphs.
However, these elements must be derived from actual analyses.
A model should not invent p-values, confidence intervals, significance stars, or effect estimates to make a figure appear complete.
Researchers should also verify whether the displayed statistical test matches the study design and whether multiple-comparison adjustments are relevant.
Statistical annotations should communicate the analysis accurately rather than merely signal that a result is important.
How Should Researchers Verify Error Bars?
Error bars can represent different quantities, including standard deviations, standard errors, and confidence intervals.
These quantities are not interchangeable and communicate different aspects of variation or uncertainty.
AI may produce error bars without clearly identifying what they represent or may calculate them using the wrong denominator.
Researchers should verify the calculations, sample sizes, and assumptions underlying the displayed intervals.
Figure captions should explain what the error bars represent and, where relevant, how they were calculated.
For repeated-measures or clustered data, uncertainty estimates should reflect the appropriate analytical structure rather than treating all observations as independent.
Can AI Create Conceptual Research Diagrams?
Yes. AI can help arrange constructs, organize processes, and produce preliminary conceptual illustrations.
For example, an educational technology researcher might create a diagram depicting proposed relationships among perceived usefulness, self-regulated learning, and technology adoption.
However, AI may introduce unsupported arrows or imply causal relationships that the study has not established.
The resulting diagram should be checked against the theoretical framework and actual research model.
Creating AI-assisted conceptual research diagrams involves different verification requirements from plotting numerical results, but conceptual accuracy remains essential.
Can AI Improve Figure Labels and Captions?
Yes. AI can help refine titles, axis labels, legends, and captions to improve readability.
However, editing can change the meaning of a figure if terminology becomes less precise.
For example, replacing "perceived learning improvement" with "learning improvement" removes an important distinction between self-reported perception and measured performance.
Similarly, changing "association" to "effect" may introduce an unsupported causal interpretation.
Researchers should verify that AI-assisted figure labels and captions preserve the meaning and evidential scope of the results.
What Makes a Research Graph Reproducible?
A reproducible visualization can be regenerated from identifiable data using documented procedures.
For a computational graph, relevant records may include the source dataset, analysis script, software version, data transformations, statistical calculations, and plotting parameters.
Researchers should also retain the final figure and any necessary intermediate analytical outputs.
Reproducibility does not guarantee correctness. A script can repeatedly produce an incorrect graph if it uses the wrong variables or calculations.
However, reproducibility makes errors easier to investigate and correct.
How Can Researchers Verify Every Plotted Value?
Verification should connect each plotted element to the original data or validated analytical output.
For a bar chart, researchers can compare each bar height with the corresponding summary statistic. For a scatterplot, they can confirm that the coordinates represent the intended observations.
For a forest plot, they should verify each effect estimate, confidence interval, and reference line.
Where code generates the figure, reviewing the underlying plotting data and calculations is generally more reliable than inspecting the rendered image alone.
The systematic process of verifying values in AI-generated research figures should be completed before publication.
Are AI-Generated Research Figures Allowed in Journal Articles?
Acceptance depends on the publisher, journal, figure category, and method of generation.
Some publishers distinguish AI-assisted data visualizations produced through reproducible computational procedures from images created using general-purpose generative image tools.
Elsevier's generative AI policies for journals, for example, distinguish explanatory illustrations, data visualizations, and primary research images.
Other publishers may impose different restrictions on AI-generated artwork or figures.
Researchers should verify the current policies of their intended publication venue rather than assuming that permission to use AI for manuscript writing also covers every figure type.
The related question of using AI-generated images in research papers requires particular attention to these editorial distinctions.
How Should AI Assistance in Figure Creation Be Disclosed?
Disclosure depends on the applicable journal, institutional, and disciplinary requirements.
When AI materially contributes to the analytical or visual production of a research figure, researchers should document its role and the procedures used to verify the output.
For code-generated graphs, the Methods section may need to describe the computational workflow, particularly when AI assistance affects data transformations or analytical procedures.
For permitted explanatory illustrations, a publisher may require a figure-caption statement and a separate AI-use declaration.
Researchers should distinguish tool disclosure from methodological reporting. Naming a model does not establish that the plotted values are accurate.
When Could an AI-Generated Graph Constitute Research Fabrication or Falsification?
Research integrity concerns arise when a figure presents invented values as empirical observations or deliberately alters results so that the research record is misrepresented.
For example, generating fictional student performance scores and presenting them as actual experimental outcomes may raise fabrication concerns.
Changing plotted values or selectively omitting inconvenient observations to misrepresent the results may raise falsification concerns.
However, an incorrect graph does not automatically establish formal misconduct. Errors may arise from misunderstanding, programming mistakes, or inappropriate visualization choices.
Formal determinations depend on applicable research misconduct standards and the circumstances of the case.
Watch Out
A polished AI-generated graph may contain accurate-looking but invented values, incorrect labels, or misleading scales. Never accept a figure solely because it resembles a professional scientific visualization. Verify its numerical and graphical relationship to the original evidence.
