First Look at Spreadsheet Data: Quality Checks, Patterns, Outliers and Next Questions

Paste a spreadsheet and get a data quality check, findings ranked by how solid they are, outliers sorted into errors vs real, and questions to investigate next.

You getData snapshot, quality issues table and findings ranked strongest first

Prompt

Prompt
Act as a careful data analyst doing a first look at a spreadsheet. I will paste my data (CSV, copied cells or a column summary) right after this prompt, plus one line on what it tracks. Tell me what the data shows, what looks wrong and what to check next.

Rules:
1. Do the data quality check first: missing values per column, duplicate rows, mixed units or formats (kg vs lb, dates stored as text), and impossible values. Make no trend claims until this is done.
2. Every finding cites the specific rows, columns or numbers it rests on.
3. Rate each finding Solid, Likely or Shaky, with a one-line reason (sample size, quality issue, one outlier driving it).
4. Separate outliers that look like entry errors from outliers that look real.
5. Do not state causes as facts. Phrase them as questions to investigate.

Output format:
1. Data snapshot: row count, columns, date range, what each column appears to measure.
2. Quality issues table: Issue | Where | Impact on analysis | Suggested fix.
3. Findings ranked strongest first: table with Finding | Evidence | Rating | Why.
4. Outliers: likely errors vs likely real.
5. Top 5 questions to investigate, each with the check that would answer it.

End with one line on what this data cannot tell me and what extra data would help.

Example output

At a glanceAverage daily revenue by segmentCafé sales, Jan 1 to Mar 31, from the findings table
  • Saturdays: $1,310
  • Weekdays: $860
  • March: $1,020
  • January: $880
  • Rainy days: $790
  • Other days with weather: $940

Example for: a small café owner pasting 92 rows of daily sales (Jan 1 to Mar 31) with columns Date, Revenue, Transactions, Avg_Ticket, Weather.

1. Data snapshot
  • 92 rows, 5 columns, Jan 1 to Mar 31 (90 days expected).
  • Date: day of sale. Revenue: daily takings in dollars. Transactions: number of sales. Avg_Ticket: revenue per sale. Weather: free-text notes.
2. Quality issues
IssueWhereImpact on analysisSuggested fix
Duplicate rowsFeb 14 and Mar 3 each appear twiceInflates totals by about $1,740Delete the second copy
Missing valuesWeather blank in 31 rowsWeather comparisons rest on 61 daysFill from a weather archive or drop the column
Mixed formatsRevenue in rows 40-46 stored as text with "$"Excluded from sumsConvert to numbers
Impossible valueJan 22 Avg_Ticket = 0 with 118 transactionsBreaks averagesRecalculate as Revenue / Transactions
3. Findings, strongest first
FindingEvidenceRatingWhy
Saturdays bring the most revenueAvg $1,310 vs $860 on weekdays, 13 SaturdaysSolidConsistent every week, enough days
Revenue rose through MarchMarch avg $1,020 vs January $880LikelyClear gap, but only 3 months and no prior year
Average ticket is flat around $7.40Range $6.90-$7.90 on clean daysLikelyGrowth comes from more transactions, not bigger orders
Rainy days sell less14 rainy days avg $790 vs $940ShakyWeather missing on 31 days, small sample
4. Outliers

Likely errors: Jan 22 (Avg_Ticket 0); Mar 9 Revenue $9,400, ten times normal with ordinary transactions, probably an extra zero.
Likely real: Feb 14 revenue $1,890 (Valentine's Day, transactions also high); Feb 2 revenue $210 (note says "closed early, power cut").

5. Questions to investigate
  1. Is the March rise real? Compare with last year's March sales if you have them.
  2. Was Mar 9 a data entry slip? Check the till report for that day.
  3. Do Saturdays need more staff? Break Saturday sales down by hour.
  4. Does rain matter? Fill the 31 missing weather days, then compare again.
  5. Can ticket size grow? Check item-level sales for add-on items.

This data cannot show why customers come or what they buy; item-level sales and last year's figures would help most.

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • Forcing quality checks before trend claims stops the model from building conclusions on duplicates, text-formatted numbers or typos.
  • Solid/Likely/Shaky ratings with reasons show which findings you can act on and which need more data.
  • Turning causes into questions with a named check gives you a concrete next step instead of guesses dressed as facts.

Use when you open an unfamiliar export and want a sober first pass. Paste up to a few hundred rows into ChatGPT, Claude or Gemini, or upload the file, and add one line on what it tracks.

Common edits

  • Add a goal line, such as 'I want to decide staffing,' so findings are ranked by relevance to that decision.
  • Ask for Excel or Google Sheets formulas to fix each quality issue listed.

Make it repeatable

Keep using this prompt without pasting it in every time.

Prompt details

By
Prompts-GPT editors
Published
How prompts are written and checked

Did this prompt work?

Keep exploring

More in Research