Describe Your Columns Before Asking for Analysis

What it is

A setup step: give ChatGPT a short data dictionary — what each column means, its units, and any quirks — before asking your analysis question, so it doesn't guess wrong about your data.

A quick data dictionary prevents wrong assumptions about your columns.

Why it works

Column names are ambiguous: 'value' could be revenue or units, 'date' could be signup or churn, a status code could mean anything. The model guesses, and a wrong guess produces a confident, wrong analysis. A few lines describing the columns removes that entire class of error up front.

When to use it

Whenever column names are cryptic, encoded, or domain-specific, and any time the analysis depends on units or categories the model can't infer.

When not to use it

Clean, self-explanatory data with obvious headers, where describing 'name' and 'email' would just be noise.

Prompt

Here's <file>. Column guide:
- <col>: <meaning, unit, quirks>
- <col>: <meaning, unit, quirks>
Note: <any gotchas — e.g. status 3 means refunded>.

Now: <your question>. Ask if any column is unclear before analysing.

Example

Telling it 'amt is in cents, not dollars; status 0 = pending, exclude those' prevents a report that's 100× too large and includes orders that never completed.

Common mistakes

Letting the model guess what an ambiguous column means.
Forgetting to state units, so totals are off by orders of magnitude.
Not flagging which category codes to include or exclude.