Claude in Excel: 15 Prompts That Work

Most people don't actually love formulas, pivot tables, or VBA. We love what they get us. If Claude can get you to the same answer faster, that changes how you use Excel.

Key Takeaways

• Claude works best in Excel when you prompt for an outcome, not a tool.
• Natural-language prompts can replace a surprising amount of formula writing, data cleaning, and ad hoc pivot work.
• The catch is accuracy: spreadsheet AI still needs human review, especially for finance and multi-step reasoning tasks.
• The strongest prompts specify sheets, columns, output layout, and edge cases.
• Claude is best used as a junior analyst inside Excel, not as an unreviewed autopilot.

What is Claude in Excel actually good at?

Claude in Excel is best at turning vague spreadsheet intent into structured outputs like summaries, formulas, transformations, and draft analyses. It shines when you know the result you want but don't want to remember the exact formula, pivot configuration, or macro logic needed to get there [1][2].

Here's the important nuance. Spreadsheet research is getting better fast, but it's not solved. Recent work on spreadsheet generation found Claude models are often preferred by users overall, yet their failures tend to cluster around correctness and numerical computation rather than polish alone [2]. Another March 2026 benchmark on financial spreadsheets found no standalone model was accurate enough for unsupervised professional use, especially on harder multi-step reasoning tasks [3].

That matches what I've noticed in practice. Claude is fantastic for getting you 80% of the way there in seconds. It is not a license to stop checking the workbook.

How should you prompt Claude in Excel?

The best Claude in Excel prompts define the workbook context, identify the exact columns or sheets involved, and describe the final deliverable in plain English. Prompts perform better when you ask for an auditable output, such as a formula, cleaned table, summary block, or new sheet layout [2][3].

A weak prompt says, "analyze this spreadsheet."

A strong prompt says, "Using Sheet 'Sales', group by Region and Product Category, show total revenue, average order value, and year-over-year growth in a new summary table, and flag any missing values."

That structure matters because spreadsheets are messy artifacts. Research on spreadsheet generation shows quality depends on functional correctness, structure, formatting, and domain conventions all at once [2]. In other words, Claude needs clearer instructions than you think.

Here are the 15 prompts I'd actually keep handy.

1) Replace a lookup formula

On Sheet "Orders", fill in the Customer Segment for each row by matching Customer ID against Sheet "Customers". Use the existing columns only, and return the best Excel formula for cell E2 that I can drag down. Explain the formula in one sentence.

2) Replace nested IF logic

In Sheet "Deals", write an Excel formula that labels each deal as "Hot", "Warm", or "Cold" based on these rules: ARR above 50000 and stage is Proposal = Hot; ARR above 20000 or stage is Demo = Warm; otherwise Cold.

3) Replace text-cleaning formulas

Clean the "Company Name" column in Sheet "Leads" by removing extra spaces, standardizing capitalization, and stripping legal suffixes like LLC, Inc, Ltd. Show the exact formula or transformation steps.

4) Replace date math

From the "Created At" and "Closed At" columns in Sheet "Pipeline", calculate sales cycle length in days, then bucket each row into 0-30, 31-60, 61-90, or 90+ days.

5) Replace a pivot table summary

Summarize Sheet "Transactions" by Month and Category. Show total spend, transaction count, and average transaction value in a clean output table I can paste into a new sheet.

6) Replace a multi-condition count

Write the best Excel formula to count rows in Sheet "Tickets" where Priority is High, Status is Open, and Owner is not blank.

7) Replace duplicate detection rules

Find likely duplicate rows in Sheet "Contacts" using Email as the primary key and Name + Company as secondary signals. Suggest a method to flag duplicates inside Excel.

8) Replace a manual anomaly scan

Review Sheet "Expenses" and identify possible anomalies such as duplicate reimbursements, round-number outliers, and transactions far above each employee's normal range. Return a flagged table with reasons.

9) Replace pivot-based cohort analysis

Using Sheet "Users", group users by signup month and calculate 30-day retention by cohort. Show the output as a matrix with cohort month in rows and retention percentage in columns.

10) Replace a VBA split-and-reshape script

In Sheet "Raw Export", split the combined field "City, State, Country" into separate columns, preserve rows with missing parts, and show the safest Excel-native way to do it.

11) Replace a monthly reporting macro

Create a repeatable monthly summary from Sheet "Revenue" that shows revenue by rep, by region, and by product line, with top 5 wins and biggest month-over-month changes.

12) Replace formula debugging

Audit the formulas in Sheet "Forecast" for likely broken references, inconsistent logic across rows, or hardcoded values inside formulas. Point to the cells most likely to be wrong.

13) Replace conditional formatting guesswork

Set up conditional formatting on Sheet "Targets" to highlight reps above quota in green, within 10% of quota in yellow, and below 90% in red. Give exact rule logic.

14) Replace ad hoc scenario modeling

Build a simple scenario table for Sheet "Pricing" showing Base, Upside, and Downside cases using conversion rate, average contract value, and churn as inputs.

15) Replace lightweight VBA automation

I want to automate cleaning and summarizing this workbook every week. Based on the current sheets, propose an Excel-first workflow that minimizes VBA and uses formulas, tables, and repeatable steps instead.

Which prompt patterns replace formulas, pivot tables, and VBA best?

Claude replaces spreadsheet tools best when the prompt maps to a job-to-be-done: lookup, classify, clean, summarize, audit, reshape, or model. Those jobs are easier for a language model than asking it to "do Excel stuff," which is too vague to produce reliable structure [2].

Here's a simple way to think about it:

This is also where tools like Rephrase help. If you already know what you want but your prompt is still fuzzy, a prompt optimizer can turn "summarize this sheet" into something Claude can actually execute well in one shot.

Why do Claude Excel prompts still fail sometimes?

Claude Excel prompts fail when the task requires hidden spreadsheet context, exact numeric rigor, or domain conventions that were never stated. The biggest risks are wrong formulas, wrong aggregation logic, and outputs that look polished but are structurally off [2][3].

The research here is pretty blunt. In SpreadsheetArena, formatting and visible structure often influenced preference, but expert finance review found weak adherence to professional spreadsheet standards [2]. FinSheet-Bench goes further: even the strongest models still made too many errors for unsupervised financial work, especially on aggregation and complex reasoning [3].

That means you should treat Claude like this:

First, use it to draft.
Then, inspect formulas.
Then, test edge cases.
Then, keep the repeatable parts in Excel itself.

If you do that, Claude becomes genuinely useful instead of dangerously impressive.

What does a better before-and-after Claude prompt look like?

A better Claude prompt moves from "analyze this" to "produce this exact artifact from these sheets and columns." That shift reduces ambiguity, improves auditability, and gives you something you can verify directly in Excel [2][3].

Here's a real before-and-after pattern.

Before

Can you help me with this sales spreadsheet?

After

Using Sheet "Sales_2026", create a summary I can paste into a new sheet with:
1. total revenue by month,
2. revenue by region,
3. top 10 accounts by ARR,
4. rows with missing owner or close date,
5. one Excel formula to classify each deal as New, Expansion, or Renewal based on the Type column.
Keep the output easy to audit.

That's the whole game. Specificity beats cleverness.

If you want more workflows like this, the Rephrase blog has more prompt breakdowns across writing, coding, and AI tool use. And if you're constantly rewriting rough prompts before sending them to Claude, Rephrase is a clean shortcut.

Claude in Excel won't kill formulas, pivot tables, or VBA. But it will absolutely kill a lot of the time you spend reaching for them by default. That's the real upgrade.

References

Documentation & Research

1. Announcing Claude Opus 4.6 and Claude Sonnet 4.6 on Vertex AI - Google Cloud AI Blog (link)
2. SpreadsheetArena: Decomposing Preference in LLM Generation of Spreadsheet Workbooks - arXiv cs.CL (link)
3. FinSheet-Bench: From Simple Lookups to Complex Reasoning, Where LLMs Break on Financial Spreadsheets - arXiv cs.AI (link)

Community Examples 4. A Guide to Which AI to Use in the Agentic Era - One Useful Thing (link)