Office workflows
Use ChatGPT to Turn a Spreadsheet into a Data Dictionary
Cihan's view: Upload an approved sample spreadsheet, ask ChatGPT to describe each column with evidence from the headers and rows, then check every definition against the source before sharing it.
A spreadsheet is easier to use when the next person does not have to guess what each column means. A small data dictionary can explain the fields, expected values, and questions that still need an owner.
This documentation-based guide uses a fictional spreadsheet and an illustrative output. It does not report a ChatGPT run or a measured time saving.
What you will create
You will create a short Markdown data dictionary for a sample operations spreadsheet. It will describe each column, show one example value, flag unclear fields, and avoid changing the original file.
Level: Beginner. You need ChatGPT on the web with file upload available and a spreadsheet you are allowed to share. OpenAI documents file uploads and data analysis, but feature availability, limits, and retention rules can vary by account. Check the current help pages before using a work file.
The fictional input
Create a CSV named service-requests.csv with this content:
request_id,opened_on,team,priority,status,first_response_hours,closed_on
SR-104,2026-10-06,North,High,Open,3.5,
SR-105,2026-10-06,West,Medium,Closed,8,2026-10-08
SR-106,2026-10-07,North,,Pending,,
SR-107,2026-10-08,South,Low,Closed,22,2026-10-10The blank priority and response time are deliberate. They give the review something concrete to flag.
Upload and ask for the dictionary
- Open a new ChatGPT conversation.
- Attach
service-requests.csv. Do not use real customer names, ticket text, or account identifiers in this practice file. - Paste the prompt below.
- Compare the result with the CSV. Save the dictionary only after you correct unsupported assumptions.
You are documenting a small operations spreadsheet for a teammate.
Use only the uploaded CSV. Do not calculate performance claims and do not invent definitions that the headers and values do not support. Do not edit the file.
Return a Markdown table with these columns:
- column name
- plain-English description
- example value copied from the file
- observed values or format
- ambiguity or question
Rules:
1. Include every column exactly once, in the original order.
2. Copy example values exactly. Use "blank in sample" when a value is empty.
3. Distinguish an observed value from a proposed business definition.
4. If the sample is too small to define a field, write "not established by sample".
5. Add a final section called "Questions to confirm" with only questions supported by the file.
After the table, list the file name and the row count you observed. If you cannot verify either, say so instead of guessing.Illustrative output
The output should mention all seven columns. For priority, it can say that the sample contains High, Medium, Low, and one blank. It should not claim that these are the complete allowed values. For first_response_hours, it can describe the values as numbers observed in the sample and flag the two blank cells. Those are editor-written examples, not captured ChatGPT output.
Acceptance checks
The dictionary passes this review when:
- all seven headers appear once and keep their original order
SR-106is not silently treated as a high, medium, or low priority request- blank
first_response_hoursandclosed_oncells remain visible - the row count matches the four data rows in the CSV
- no definition claims to be a complete policy unless the source file says so
- the original CSV is unchanged
Troubleshooting
ChatGPT skips a column. Ask it to return the header list first, then compare that list with the CSV before requesting the table again.
It turns a guess into a rule. Add: Label every unsupported business meaning as a question. Do not infer policy from a few rows.
The upload is rejected or the table looks incomplete. Check the current file-upload help, reduce the practice file, or paste the small sample as text. Do not assume a partial answer covered the whole file.
Limits and safer use
A data dictionary explains what a sample appears to contain. It does not establish a data policy, validate every row, or approve a metric. Remove personal and confidential data before upload, and review your organisation’s retention and workspace rules.
For a next step, see Use ChatGPT to Find Anomalies in a Sample Sales CSV.
Want one useful workflow each week? Join The Weekly Verdict.
Sources
Evidence label: Documentation-based guide with fictional input and an illustrative output. No ChatGPT execution or personal test is claimed.