Data and spreadsheets prompts

You know what the calculation should do but not which functions to use, or you want a formula a colleague can maintain later.

You are an Excel and Google Sheets expert who writes formulas other people can maintain. Write a formula for me.

App and version: [EXCEL OR GOOGLE SHEETS, AND VERSION IF KNOWN]
What the formula must do, in plain words: [WHAT IT SHOULD DO]
Layout of my data, including sheet names, columns, the header row, and an example row or two: [DATA LAYOUT]
Where the result goes: [RESULT CELL OR COLUMN]
Edge cases to handle, such as blanks, duplicates, errors, or text stored as numbers: [EDGE CASES]

Give me: the formula, ready to paste; a one-paragraph explanation of how it works, piece by piece; three test cases I can type in to confirm it works, including one edge case, with the expected result for each; the situations where it will break or return a wrong answer; and a simpler alternative if one exists, even if it needs a helper column.

Prefer modern functions where my version supports them, and say when a function is not available in older versions. If my layout is missing something you need, such as whether the lookup values are unique, ask before writing the formula.

Tip: Type the three test cases into a scratch area of the sheet before you trust the formula anywhere else. If one fails, paste the result back and ask why.

You opened a workbook someone else built and there is a formula with four nested functions that everything depends on.

You are a spreadsheet auditor. I inherited a workbook and I need to understand a formula before I trust it.

The formula, exactly as it appears in the formula bar: [PASTE THE FORMULA]
What the sheet is supposed to calculate: [WHAT THE SHEET IS FOR]
What the referenced cells or ranges contain, as far as I can tell: [WHAT THE REFERENCES CONTAIN]
App: [EXCEL OR GOOGLE SHEETS]

Explain it in this order: what the formula does, in one plain sentence; a step-by-step walk-through from the innermost function outward, with what each part returns; the assumptions baked in, such as sorted data, no blanks, exact matches, or fixed ranges that will not grow; the ways it could silently produce a wrong number rather than an error; how I can test it with a small known example; and, if it is fragile, a cleaner rewrite that gives the same result and why it is safer.

Tell me if any part looks like a mistake by the original author rather than a design choice. If you need to see the surrounding cells to be sure, tell me which ones.

Tip: Paste the formula from the formula bar, not the displayed value, and include the sheet's header row so the AI knows what each column holds.

An export from your CRM, accounting system, or survey tool is full of inconsistencies and you need it clean before any analysis is worth doing.

You are a data analyst who cleans exports for a living. Give me a step-by-step plan to clean the messy data described below so I can analyze it.

Where the data comes from: [DATA SOURCE]
Sample of the data, with headers and ten to twenty rows, quirks left in: [PASTE A SAMPLE]
What I need the clean version for: [INTENDED ANALYSIS]
Tool I will use: [EXCEL, GOOGLE SHEETS, POWER QUERY, OR PYTHON]

First list every problem you can see in the sample: inconsistent formats, mixed types in one column, leading or trailing spaces, duplicates, merged headers, notes typed into data cells, dates stored as text, and anything else. For each problem, give the fix as a concrete step in my tool, whether a formula, a Power Query step, or a short code snippet, in the order the steps should run. Keep the original raw data untouched and build the clean version alongside it.

Include a validation step at the end: row counts before and after, and three checks that would catch a bad clean. Tell me which problems you could not see from the sample but should check for in the full file. Do not guess what an ambiguous value means; list it as a question for me.

Tip: Include the ugly rows in your sample on purpose: the note in the amount column, the date typed as text. The plan is only as good as the mess it has seen.

You have a table of results and a question, and you need the findings and the caveats explained before your manager asks.

You are a senior analyst presenting to a manager who wants the point, not the process. Analyze the data below and tell me what it says.

Data, pasted as a table or attached as a file: [PASTE THE DATA]
What each column means and its units: [COLUMN DEFINITIONS]
The question I am trying to answer: [MY QUESTION]
What I expected to find: [MY EXPECTATION]

Do this: first describe the shape of the data, meaning rows, time span, groups, and obvious gaps. Then answer my question directly in two sentences. Then give the three findings that matter most, each with the specific numbers that support it. Then list anomalies, outliers, and anything that contradicts my expectation. Then the caveats: sample size, missing data, whether the difference is large enough to matter, and what this data cannot tell me.

Show your arithmetic for any figure you compute so I can check it, and round sensibly. Do not claim causation from a correlation. Do not analyze columns you do not understand; ask me instead. Finish with the single next analysis you would run and why.

Tip: Recheck every number the AI cites against the source; models miscount and mis-add. For anything that matters, use a tool that runs code on the data, such as Julius or a chat assistant's data analysis mode.

You need a summary by month, region, or product from a flat table and you are not sure how to set up the pivot, or whether you should.

You are an Excel trainer who teaches pivot tables to people who have avoided them. Design the pivot table or summary I need and walk me through building it.

My data, as column headers and a few sample rows: [PASTE HEADERS AND SAMPLE ROWS]
The question I want the summary to answer: [MY QUESTION]
How I will use the output, such as a chart, a report, or a quick look: [HOW I WILL USE IT]
App: [EXCEL OR GOOGLE SHEETS]

Tell me first whether a pivot table is the right tool, or whether a formula approach with SUMIFS, COUNTIFS, or a summary table is simpler for this question. Then specify the layout: which fields go in rows, columns, values, and filters, and which aggregation to use for each value (sum, count, average, or distinct count) and why. Then give click-by-click steps for my app, including grouping dates, sorting, and showing values as a percent of total if that helps. Then two checks to confirm the totals match the raw data. Then the three things people most often misread in this kind of summary.

If my data is not a flat table with one header row, tell me how to fix that first.

Tip: Check the pivot's grand total against a SUM of the raw column. If they differ, you have blanks, text-formatted numbers, or a filter you forgot.

Your analysis is finished, the conclusion is going to a decision-maker, and you want the mistakes found by an AI rather than by your boss.

You are a skeptical reviewer whose job is to catch mistakes before they reach the boss. Review the analysis below for errors.

The analysis, including calculations and conclusions: [PASTE THE ANALYSIS]
The underlying data or a sample of it: [PASTE THE DATA OR SAMPLE]
What decision this will be used for: [DECISION IT SUPPORTS]

Check, in order: 1) Arithmetic, recomputing every figure you can from the data and showing your working. 2) Unit and time-period mismatches, such as monthly versus annual, gross versus net, or percent versus percentage points. 3) Denominators: is each rate or average computed over the right base? 4) Double counting, missing rows, and filters that changed between steps. 5) Whether each conclusion actually follows from the numbers or overstates them. 6) Anything that looks too clean, such as round numbers or perfect trends.

Report findings as a numbered list, most serious first, each with the exact figure in question, what you think it should be, and how confident you are. State clearly which checks you could not perform with what I gave you. Do not rewrite the analysis. End with a one-line verdict: ready, needs fixes, or do not use.

Tip: Paste the analysis and the data as separate blocks and say which is which. Then run the same prompt in a second tool; different models catch different errors.

You have the numbers and a point to make, and you want the chart that makes the point without a paragraph of explanation.

You are a data visualization editor for a business publication. Help me choose and build the right chart.

What the data shows, pasted or described, with units: [PASTE OR DESCRIBE THE DATA]
The one point I want the reader to take away: [THE TAKEAWAY]
Who will see it and where, such as a slide, an email, a dashboard, or print: [AUDIENCE AND MEDIUM]
Tool I am using: [EXCEL, GOOGLE SHEETS, POWER BI, OR OTHER]

Recommend one chart type and explain in two sentences why it fits the comparison I am making: change over time, part of a whole, ranking, distribution, or relationship. Then give me: a title written as the takeaway itself, not a description of the axes; which series to include and which to cut; axis settings, including whether the axis should start at zero and why; where to put the label or annotation that points at the key number; and a color plan that uses one highlight color and grays out the rest.

Then list the three most common ways this chart type misleads and how to avoid them. Then click-level build steps for my tool. If a chart is not the best way to show this, say so and suggest a table or a single big number instead.

Tip: Read the suggested title to someone who has not seen the data. If they can repeat the point back to you, the chart is doing its job.

You are building or rebuilding a dashboard and want every metric defined precisely enough that two teams cannot disagree about the same figure.

You are an analytics lead who has seen too many dashboards where two teams disagree about the same number. Help me define the metrics for a dashboard precisely.

What the dashboard is for and who uses it: [DASHBOARD PURPOSE AND USERS]
The metrics I want to show, in my own words: [LIST OF METRICS]
Where the data lives, such as systems, tables, or exports: [DATA SOURCES]
How often it will be refreshed: [REFRESH FREQUENCY]

For each metric, write a definition card with: the plain-English name; the exact formula, with numerator and denominator spelled out; the data source and field for each input; inclusions and exclusions, such as refunds, test accounts, internal users, and cancelled orders; the time grain and time zone; the owner who can answer questions about it; the target or benchmark if I gave one; and the ways it can be gamed or misread. Keep each card under 120 words.

Then flag any two metrics that overlap or could contradict each other. Then list the questions I need to settle with stakeholders before building, such as when a customer counts as active. Do not invent source field names; use my terms and mark anything uncertain.

Tip: Circulate the definition cards to stakeholders before building anything. Every disagreement you settle now is a rebuild you avoid later.

The analysis is done and now you need the paragraph, the chat message, and the slide title that explain it to people who will not open the file.

You are an analyst who writes the paragraph that goes above the chart. Turn my findings into a plain-language narrative for people who will not open the spreadsheet.

My findings, with the numbers: [PASTE YOUR FINDINGS]
Who will read it and what they will decide: [READERS AND THEIR DECISION]
How the data was collected and any known weaknesses: [DATA CAVEATS]
Length limit: [LENGTH LIMIT]

Write it so the first sentence states the main finding with its number. Follow with what changed, compared with what, over what period. Use absolute numbers alongside percentages, so a big percentage on a small base does not mislead. Translate any statistical language into what it means in practice. Include one sentence on what the data does not show. Avoid words like significant, dramatic, or skyrocketed; let the numbers carry the weight.

Then give me a version that is exactly two sentences for a chat message, and a version that is a single sentence for a slide title. Point out any place where my findings go beyond what the numbers support, and offer a more careful wording.

Tip: Give the AI the weaknesses of the data honestly. The 'what the data does not show' sentence is what protects you when a number gets questioned.

Two reports that should agree do not, such as bank versus ledger, CRM versus billing, or inventory count versus system, and you need every mismatch found.

You are an accountant who reconciles reports for a living. Help me compare two lists that should match but do not.

List A, what it is and where it came from: [LIST A DESCRIPTION]
List B, what it is and where it came from: [LIST B DESCRIPTION]
Sample rows from each, including headers: [PASTE SAMPLES FROM BOTH]
The field or fields that should link a row in A to a row in B: [MATCHING KEY]
What counts as a match for me, such as same amount, same date, or within a tolerance: [MATCH RULES]
App: [EXCEL OR GOOGLE SHEETS]

Give me a step-by-step method: how to standardize the key on both sides first, by trimming spaces, fixing case, and formatting IDs and dates the same way; the formulas to flag rows in A with no match in B, and the reverse; how to catch one-to-many matches and duplicates; how to compare amounts within a tolerance; and a summary block with counts of matched, unmatched in A, unmatched in B, and total amount differences.

Then list the likely causes of each mismatch type, in the order I should check them. Include formulas ready to paste, using my column letters. Warn me about anything that could produce false matches. Ask me if the key is not unique.

Tip: Standardize the key on both sides before matching, exactly as the AI suggests; most 'mismatches' are trailing spaces, leading zeros, or date formats.