For comparing totals across categories, bars usually need a zero baseline because their lengths encode values. For a time series, preserve the order and actual spacing of dates. An unobserved month needs an explicit treatment; it is not automatically zero.
Decide what the reader should compare
Suppose a workshop wants to show its recorded completed repairs by month. This original fictional table is a teaching example, not business evidence or an observed ChatGPT result:
| Month | Completed repairs | Data status |
|---|---|---|
| January | 80 | Recorded |
| February | 84 | Recorded |
| March | Not available | Export missing |
| April | 82 | Recorded |
January to February increased by four repairs: 84 - 80 = 4. Relative to January, that is 4 / 80 × 100 = 5%. Calling it a four-percent increase confuses an absolute count with a relative change.
The useful question is how the recorded monthly totals compare, with March unavailable. A pie chart would imply a known whole across the months; that is a poor fit for this incomplete series. A line chart can show the sequence, provided March remains a visible gap and the caption explains it. A bar chart can compare the three known totals, with March shown as missing rather than as a zero-height recorded value.
State the scale and missing-data rules
The Office for National Statistics' axes guidance explains why a zero baseline is needed when filled bars or areas represent amounts. Line charts can have a different numerical range when the purpose calls for it, but their labels and context must make that range clear. Source checked October 24, 2026.
For this example, a bar chart starting at 79 would make 84 look several times larger than 80 even though the recorded difference is 5%. Request a zero baseline for the bars. If you instead choose a line chart with a restricted vertical range, make that scale visible and avoid a title suggesting a dramatic increase.
Do not connect February directly to April as if March had been observed. Ask for a gap at March and a note saying its export is unavailable. Do not interpolate a value unless that is a separate, justified analysis clearly distinguished from observed data.
Ask for a chart specification before the image
OpenAI's data-analysis documentation, checked October 24, 2026, describes creating charts from supplied data and reviewing the analysis. Available chart tools depend on the account and working environment. You can still request a chart specification and caption when a particular rendering option is unavailable.
Help me visualize this fictional monthly completed-repair table.
January: 80; February: 84; March: unavailable because the export is missing;
April: 82.
Reader question: how do the recorded monthly totals compare?
First return a readable table with values, units and missing status.
Suggest one chart type and explain what its marks and axes encode.
Keep chronological order. If using bars, start the value axis at zero.
Show March as missing; do not replace it with zero or estimate it.
If using a line, leave a gap for March instead of connecting across it.
Then create the chart if supported, plus a text caption stating the
measure, period and missing-month qualification. Label the count unit.
Check January-to-February absolute and relative change separately.
Show the plotting code or the chart settings used where available.
Do not invent causes, targets, annual totals or evidence of a trend.
Check the exported version with its table
Read the axis labels at the size your audience will receive. Confirm that January is 80, February 84 and April 82; March must not acquire a numeric point during plotting or export. Check that the legend and caption survive outside the conversation, where your original prompt will not accompany them.
A suitable authored caption is: “Recorded completed repairs, January–April: 80 in January, 84 in February and 82 in April. March's export is unavailable. February's total is four repairs, or 5%, above January.” These values support that comparison, not a claim that business improved for a particular reason.
If the underlying table still has duplicated records or unclear denominators, use the file-analysis checks first. For this visualization task, keep the approved table, plotting settings and caption together. Open the actual exported chart on the intended screen before sending it.