iaExcel
Pricing
Docs

Idioma / Language

País

Blog/Benchmark

iaExcel vs Microsoft Copilot vs ChatGPT: 5 real tests on a messy spreadsheet

We took a spreadsheet of 1,500 sales with mixed dates, broken IDs and dirty prices, and asked the 3 tools the same things. Here's what each one did — no filters.

Ei

Equipo iaExcel

Investigación de producto

May 20, 2026

6 min read

5Real tests

A spreadsheet with dirty data. Three AI tools. The same question to each. These are the results.

iaExcelCopilotChatGPT

Instead of opining on which AI tool is best for Excel, we took a realistic spreadsheet (1,500 sales with mixed dates, broken IDs and dirty prices) and asked iaExcel, Microsoft Copilot and ChatGPT the same things. Below, every answer side by side.

Tests

5

Tools

3

Real rows

1.500

Outputs compared

15

Results at a glance

Five tests, three tools, one clear pattern: iaExcel and ChatGPT tie quantitatively; Copilot lost 4 of 5.

T01
T02
T03
T04
T05
iaExcel
~
Copilot
ChatGPT
~
Initial EDA
Cleanup
Calculation
Rollback
Plan
GANA~PARCIALFALLA
iaExcelTie
92/100
ChatGPTTie
92/100
Copilot
36/100
Final score · 20 pts win · 12 pts partial · 4 pts fail
RESULTADO

TL;DR

  • iaExcel and ChatGPT tie: they win in different areas. iaExcel in live formulas, versioning and architecture. ChatGPT in fast delivery and detailed plans.
  • Copilot finished last: three consecutive hallucinations. Claimed to have generated files that didn't exist in three tests.
  • The conceptual finding: only iaExcel returns VLOOKUP as a live formula. The others return frozen numbers that break if data changes.

The 3 tools

iaExcel

Built specifically for Excel + AI. File versioning integrated in the UI.

PlanPaid
ModeloCore 1.3
Microsoft Copilot

Assistant embedded in Excel 365. Native access to the open workbook.

PlanMicrosoft 365
ModeloGPT-5 family
ChatGPT

Web client with Code Interpreter enabled. Processes files in sandbox.

PlanPlus
ModeloGPT-5

All three market themselves as "AI for Excel," but they have very different out-of-the-box capabilities. This is the real difference before the tests begin:

Capacidad
iaExcel
Copilot
ChatGPT
Reads full multi-sheet files
Returns downloadable Excel from chat
Writes live formulas (not just values)
~
Preserves Excel cell formatting
~
File version history
Internet access for live data
Natively embedded in Excel
Live multi-user co-editing
Macros / VBA
Capabilities of each tool before the benchmark begins
NOTA

The test file

ventas_pyme_cl_2025.xlsx: 3 sheets, 1,500 sales with intentionally seeded defects (dates in 3 formats, IDs in variants, prices mixing number and text, negative discounts and names with typos).

The 5 tests, one by one

01
EDA · INITIAL READ

Describe the file structure and quality issues

iaExcel
GANA

Detected all 3 sheets and returned two structured tables with metrics per sheet and per check.

  • ·187 invalid IDs
  • ·228 non-numeric values
  • ·34 out-of-range discounts
✦ Only one with chart
📊con gráfico
Copilot
FALLA

Only processed the first 186 rows and reported a single sheet.

  • ·Saw 186 of 1,500 rows
  • ·Detected 1 of 3 sheets
✕ Partial read
ChatGPT
GANA

Complete read with concrete numbers and remediation recommendations at the end.

  • ·Same numbers as iaExcel
  • ·More narrative, no chart
✦ Just as accurate, no visual
Copilot only processed 186 of 1,500 rows. If your file is bigger than its window, it won't warn you.
Finding · Test 01
02
CLEANUP · EXCEL OUTPUT✦✦

Clean dates, prices and names, and return the corrected Excel

iaExcel
GANA

Generated the file in a single step. Consolidated names by ID (not by text similarity).

  • ·Dates as real Date type
  • ·Prices with currency format
  • ·Automatic versioning
✦ Only one with Excel format
Copilot
FALLA

Four exchanges claiming the file was ready. Never actually generated anything.

  • ·4 consecutive hallucinations
  • ·0 files generated
✕ Cannot deliver files
ChatGPT
GANA

Generated visible code with fuzzy matching and saved the file correctly.

  • ·File delivered on first try
  • ·No Excel cell formatting
✦ Correct delivery, no format
Copilot said the file was ready to download four times. All four were false.
Finding · Test 02
03
MULTI-TURN · CALCULATION✦✦

Add a margin column joining with the Productos sheet

iaExcel
GANA

Returned the column as a live Excel formula (VLOOKUP + IFERROR), not as a calculated value.

  • ·Live formula: recalculates if price changes
  • ·Percentage format 0.0%
  • ·Preserves header styles
✦ Only one with live formula
Copilot
FALLA

Second consecutive hallucination. No executed code, no file delivered.

  • ·2nd hallucination in a row
  • ·Only describes the formula
✕ 2nd hallucinated download
ChatGPT
~PARCIAL

Delivered the file with margin calculated, but as a static value. Won't recalculate if data changes.

  • ·1,500 sales with correct margin
  • ·Static value, not formula
~ Correct but frozen
iaExcel returns a live formula. The others return frozen numbers. That's the difference between a spreadsheet and a table.
Finding · Test 03
04
ROLLBACK · ARCHITECTURE✦✦

Roll back to the version without margin and add ranking by branch

iaExcel
GANA

One-click rollback in the version selector. Then placed the ranking in Sucursales sheet (not Ventas) and used SUMPRODUCT + RANK.EQ as live formulas.

  • ·Rollback in 1 click, 0 prompts
  • ·Ranking in Sucursales (5 rows, not 1,500)
  • ·Weights by applied discount
✦ Structural rollback + architecture
Copilot
FALLA

Third consecutive hallucination. Claims to have generated a new file; doesn't exist.

  • ·3rd consecutive hallucination
  • ·0 files in entire benchmark
✕ 3rd hallucinated download
ChatGPT
GANA

Remembered the previous file from its sandbox and added the ranking. One-line response, no verifiable code.

  • ·Conversational versioning worked
  • ·Ranking as value, not formula
✦ Worked but opaque
In iaExcel, rolling back is one click. In ChatGPT, it's hoping the model remembers correctly. In Copilot, it's a concept that doesn't exist.
Finding · Test 04
05
TRANSPARENCY · PLAN

Show the step-by-step plan before executing

iaExcel
~PARCIAL

Correct six-step plan, waited for approval. But the least detailed of the three: no concrete examples.

  • ·6-step plan
  • ·No examples from the file
~ Meets the bar, no detail
Copilot
GANA

Clear five-step plan with real examples from the file. Copilot's first win.

  • ·Examples from the actual file
  • ·Anticipates edge cases
✦ Copilot's first win
ChatGPT
GANA

Thorough seven-step plan + five optional improvements. Most useful for iteration.

  • ·7 steps + 5 optional extras
  • ·Specific extraction rules
✦ Most useful plan
When the task is just text, Copilot is competitive. The problem starts when a file needs to be delivered.
Finding · Test 05

The Copilot pattern

Three consecutive hallucinations aren't coincidence: they're a documentable pattern. In tests requiring file delivery, Copilot claimed to have done so four times before admitting, only on the fifth turn of test 02, that it can't deliver downloadable files via chat.

4
Files promised
T02, T03, T04, T02-bis
0
Files delivered
In entire benchmark
0%
Delivery rate
0 of 4
What Copilot promised vs what it delivered
T01
Said
"Processed the entire file"
Did
Read 186 of 1,500 rows
T02
Said
"File ready for download"
Did
Nothing (hallucination)
T03
Said
"Saved as new version"
Did
Nothing (hallucination)
T04
Said
"Separate file with ranking"
Did
Nothing (hallucination)
T05
Said
"Clear plan with examples"
Did
Delivered: the plan was good

This doesn't disqualify Copilot. Where it works, it works well: in test 05 it was competitive. The issue is positioning: it's marketed as "AI for Excel" but in standard chat it can't deliver modified Excel files. To do that you have to use Copilot inside Excel, not in chat.

Where iaExcel loses by design

There are tasks where Copilot or ChatGPT are genuinely better. Saying otherwise would be selling smoke.

VBA
Macros and VBA

iaExcel runs Python, not Visual Basic. To automate Excel with macros, use Copilot or ChatGPT.

Mejor opciónCopilot
👥
Live co-editing

No simultaneous multi-user editing yet. If the team edits the same file in parallel, Office 365 is better.

Mejor opciónCopilot
GB
Files > 2 GB

Sandbox has a ~2 GB memory limit. For millions of rows, use a specialized tool.

Mejor opciónOther tools
VEREDICTO

Which one to choose?

There's no absolute winner — there's a best fit for your case. The most important thing is to test with your real Excel, not a benchmark's.
01

To work all day inside Office 365

Copilot inside Excel, not in chat. Native integration outweighs any gap.

02

For real spreadsheets (messy, multi-sheet, Spanish)

iaExcel. Better defect handling, versioning and traceability.

03

For one-off answers if already paying for ChatGPT Plus

ChatGPT with Code Interpreter. No point adding another tool.

En una frase

ChatGPT knows how to write Excel files. iaExcel knows how to use Excel.

It's the difference between a generalist tool that understands a format and a product designed end-to-end for spreadsheets.

Try iaExcel for free

Upload your Excel and ask in plain language. No credit card required.

Get started
iaExcel vs Microsoft Copilot vs ChatGPT: 5 real tests on a messy spreadsheet | iaExcel