Critiquing AI-Suggested Visualizations

This chapter develops the next AI-collaboration skill: critiquing AI-suggested visualizations. The previous skill was working with libraries you are still learning.

The risk is slightly different now. With pandas, the AI might invent a method name or use the wrong column. With charts, the code might be valid and the image might look polished, while the chart still answers the wrong question or suggests a misleading conclusion.

Your job is to critique the chart idea, not only the syntax.

Four Checks

When an AI assistant suggests a chart, check four things.

First, does the chart match the question?

If the question is “How has study time changed over days?”, a subject-total bar chart does not answer it.

Second, does the code use the real columns?

Your study-log DataFrame has:

subject, minutes, date

Code that uses "topic", "duration", or "day" is written for a different table unless you adapt it.

Third, did the code prepare the right summary?

For total minutes by subject, the code should group by subject and sum minutes:

summary = df.groupby("subject", as_index=False)["minutes"].sum()

Plotting raw session rows may answer a different question.

Fourth, can you explain the result honestly?

A chart can show that one subject has more logged minutes than another. It cannot prove that one subject is more important, harder, or better. Those are interpretations beyond the chart.

How has my study time changed over days?

sns.barplot(data=df, x="subject", y="minutes")

A Worked Critique

Suppose you ask:

How has my study time changed over days?

Use this complete six-row sample so the suggestion and the resulting image can be checked against the same data:

import pandas as pd

records = [
    {"subject": "math", "minutes": 45, "date": "2026-05-10"},
    {"subject": "history", "minutes": 30, "date": "2026-05-11"},
    {"subject": "programming", "minutes": 60, "date": "2026-05-11"},
    {"subject": "math", "minutes": 25, "date": "2026-05-12"},
    {"subject": "programming", "minutes": 90, "date": "2026-05-13"},
    {"subject": "history", "minutes": 50, "date": "2026-05-13"},
]

df = pd.DataFrame(records, columns=["subject", "minutes", "date"])

An assistant suggests:

import seaborn as sns

sns.barplot(data=df, x="subject", y="minutes")

This code may run. It uses real columns. But it does not answer the question. It compares subjects, not days.

Here is what the suggestion actually draws for a small sample log:

There is another issue, and the picture shows it. sns.barplot groups the raw rows by the category on the x-axis. By default, each bar is the mean of the minutes values for that subject, and seaborn adds uncertainty intervals. Those bars are not subject totals and they do not show change over time. For a beginner project, it is clearer to prepare the daily total yourself:

daily = df.copy()
daily["date"] = pd.to_datetime(daily["date"])
daily = daily.groupby("date", as_index=False)["minutes"].sum()
daily = daily.sort_values("date")

Then plot and label the corrected chart:

import matplotlib.pyplot as plt

fig, ax = plt.subplots()
sns.lineplot(data=daily, x="date", y="minutes", marker="o", ax=ax)
ax.set_title("Study minutes over time")
ax.set_xlabel("Date")
ax.set_ylabel("Minutes")
plt.show()

Now the chart matches the question: one total per day, in date order.

Rebuild that fix yourself: arrange the preparation and plotting lines that answer the daily-minutes question.

Better Prompts for Chart Help

A vague prompt:

Make a cool chart for my study log.

This invites the assistant to guess the question, the data shape, and the chart type.

A better prompt:

I have a pandas DataFrame named df with columns subject,
minutes, and date. The date column can be converted with
pd.to_datetime. I want one chart that answers this question:
how many total study minutes were logged on each date?
Please prepare the daily totals first, sort by date, then
make a simple line chart with matplotlib or seaborn. Use
readable intermediate variables and return fig, ax instead
of calling plt.show().

This prompt gives the assistant less room to drift:

• it names the DataFrame columns,
• it names the question,
• it names the summary step,
• it asks for date sorting,
• and it preserves the project boundary by returning fig, ax.

Where You Are in the Arc

The course’s AI-collaboration arc now includes:

The next chapter applies the skill directly. You will add charts to the study log, but you will keep the same boundaries you have built over the course: load data, create the DataFrame, prepare summaries, plot, and check the result.

Check Your Understanding

1. What four checks should an AI-suggested chart pass?
2. Why can chart code be valid and the image polished while the chart is still the wrong chart?
3. The question is about change over days, but the suggestion is sns.barplot(data=df, x="subject", y="minutes"). What is wrong, and what should the code do instead?
4. What should a good visualization prompt include?