Comparing AI Models for Visualising Data

In this article I look at how well current AI models can visualise simple data. To make it interesting I am using real data on the most common first names of MOA Benchmarking users. The top five most common first names are summarised in Table 1. I asked a few AI chatbots to have a go at visualising the data. The results were more varied than I expected and I found that each model had its own strengths and weaknesses.

Table 1: Top five most common first names among MOA Benchmarking users as of 14th of July 2025.

FirstName	n
Michelle	153
Sarah	128
Karen	116
Lisa	100
Julie	91

Prompt

Can you visualise this data?
FirstName,n
Michelle,153
Sarah,128
Karen,116
Lisa,100
Julie,91

Claude (Sonnet 4)

Visualisation technology: Recharts (javascript).

Good

Claude visualises the data inside of the chat client, so I didn’t have to run any code.
The chart was interactive, allowing me to hover over bars to see the counts.

Bad

x-axis labels are on an angle making them harder to read.
The y-axis doesn’t have a title.
I would also avoid vertical grid lines for this bar chart as they don’t improve readability.

ChatGPT (GPT-4o)

Visualisation technology: matplotlib (Python).

Good

Sensible use of axis titles.
Clean presentation.

Bad

I had to run the Python code myself.
The plot looks a bit boring.

ChatGPT (DALL-E 3)

Visualisation technology: ChatGPT-4o image generation powered by DALL-E 3.

Good

Funny results.

Bad

One of the names is dropped.
Bars line up with incorrect grid lines, indicating we can’t trust this plot.
x-axis title is unnecessary.
y-axis title should be more descriptive than “n”.

Gemini (2.5 Pro)

Visualisation technology: matplotlib (Python).

Good

Gemini is able to run Python code, so I did not have to run any code myself.
Gemini shares the code used, which is quite readable. It was easy to modify the code to make improvements.

Bad

x-axis labels are on an angle making them harder to read.
x-axis title is unnecessary.
y-axis title should be more descriptive than “n”.

My attempt without AI

Code

library(tidyverse)
library(showtext)
font_add('Neue-Montreal', regular = here::here('fonts/NeueMontreal-Regular.otf'), 
         bold = here::here('fonts/NeueMontreal-Bold.otf'))
showtext::showtext_auto(enable = TRUE)
top5 |>
  ggplot(aes(fct_reorder(FirstName,n), n)) +
    geom_bar(stat='identity', width = .8, fill='#432585') +
    geom_text(aes(y=0,label=FirstName),nudge_y=2,hjust=0,
              fontface = 'bold',family = 'Neue-Montreal',
              size=10, colour='white') +
    labs(y='Count') +
    labs(title='The Five Most Common First Names Among MOA Benchmarking Users') +
    theme_minimal(base_size = 22, base_family = 'Neue-Montreal') +
    scale_y_continuous(breaks = scales::breaks_pretty(n=7)) +
    scale_x_discrete(labels=paste0(5:1,'.')) +
    theme(axis.title.y.left = element_blank(),
          axis.text.y = element_text(margin=margin(r=-15), family = 'Neue-Montreal'),
          panel.grid.major.y = element_blank(),
          plot.title = element_text(face='bold', size=26, family = 'Neue-Montreal',
                                    margin=margin(b=20)),
          axis.title.x.bottom = element_text(margin=margin(t=15)),
          plot.margin = margin(r=20),
          plot.title.position = 'plot') +
    coord_flip()

Visualisation technology: ggplot2 (R).

I tried to make the graph eye catching while avoiding unnecessary grid lines.

Conclusion

At the time of writing, my main takeaways are:

I would recommend Claude if you want to analyse data within the chat interface.
I would recommend Gemini for preparing shareable graphs if you don’t want to run code on your computer.
If you are able to run the code yourself then any mainstream LLM chat interface will likely work fine, but in this experiment I would say ChatGPT (GPT-4o) produced the best results.
ChatGPT-4o image generation (DALL-E 3) is unsuitable for graphing as it hallucinates and the process isn’t transparent.
Using ggplot2 or your graphing library of choice without AI is still powerful and satisfying.

These tools are developing rapidly, so my findings may not be relevant a year from now. However, I think it is worth keeping an eye on this space as there is clearly great potential.