GPT-5.5 for Excel (2026): Formulas & Data Analysis Workflow

Short answer: Use GPT-5.5 (ChatGPT Plus, Team, or API) to draft formulas, fix broken logic, and plan dashboards when you describe columns clearly and keep sensitive cells out of the chat. Budget 25 to 40 minutes to learn a repeatable workflow. Skill level: comfortable with basic Excel functions.

We tested this flow in Microsoft 365 Excel on June 2, 2026 with GPT-5.5-class ChatGPT (Plus/Team). Pair with the latest AI models hub for when to use Claude or DeepSeek instead of OpenAI.

Last updated: June 2, 2026.

What you need

Safety rules before you paste anything

1. Check company policy on AI and customer data.
2. Remove PII and secrets; use fake labels.
3. Prefer structure over dumps: headers + 2 rows beats 10,000 rows in chat.
4. Recalculate in Excel; models do not run your full workbook.
5. Log which version of Excel you use (2019 vs 365 changes answers).

For agent-style office automation across files, OpenAI markets GPT-5.5 for spreadsheets in broader agent flows. This guide stays on analyst workflows you can do today without custom macros.

Workflow overview

Step 1: Define the outcome in plain language

Write one sentence:

• “Flag rows where OrderDate is older than 30 days and Status is Open.”
• “Sum Revenue by Region for Q2 only.”
• “Match SKU from Sheet2 to Price on Sheet1 without VLOOKUP left-limit bugs.”

Expected result: You know the success cell or column before you open ChatGPT.

Common mistake: “Fix my sheet.” The model guesses wrong columns.

Step 2: Send a schema block (not the whole file)

Paste this template into ChatGPT:

Excel 365. Help me with formulas only.

Sheet: Orders
Columns:
A OrderID (text)
B OrderDate (date)
C Region (text)
D Revenue (number)
E Status (text: Open, Closed)

Sample rows:
ORD-1 | 2026-05-01 | West | 1200 | Open
ORD-2 | 2026-06-15 | East | 800 | Closed

Goal: [your one sentence]
Return: one formula for F2 to fill down, and explain each part in plain English.

Expected result: A formula using XLOOKUP, FILTER, or IF patterns that match 365 syntax.

Common mistake: Uploading a CSV with employee emails still in column C.

Step 3: Formula patterns that work well with GPT-5.5

Lookup without VLOOKUP pain

Ask:

Use XLOOKUP to pull Price from Sheet2!A:B into Orders!F2.
Sheet2: A=SKU, B=Price. Return 0 if missing.

Typical answer shape (verify in your sheet):

=IFERROR(XLOOKUP(B2, Sheet2!$A:$A, Sheet2!$B:$B, 0), 0)

Adjust cell references to your table.

Filter open orders older than 30 days

=FILTER(A2:E100, (E2:E100="Open")*(TODAY()-B2:E100>30))

GPT-5.5 often suggests FILTER for 365. Confirm your range and table bounds.

Revenue by region (SUMIFS)

=SUMIFS(D:D, C:C, "West", B:B, ">="&DATE(2026,4,1), B:B, "<="&DATE(2026,6,30))

Dynamic unique list

=UNIQUE(FILTER(C2:C100, D2:D100>0))

Expected result: Excel calculates without #NAME? or #SPILL! if ranges are valid.

Step 4: Validate like an analyst (not like chat)

Run five checks manually:

If something fails, paste Excel error text and the formula bar contents back into ChatGPT:

Formula returns #SPILL!. Here is my formula: [paste]
Here is the occupied range in column G: G2:G50 has data.
Fix with minimal change.

Expected result: Revised formula that clears the spill error.

Step 5: Data analysis and dashboard planning

GPT-5.5 is strong at metric definitions even when you build charts yourself.

Prompt:

I need a simple dashboard for sales leadership.
Metrics: MTD revenue, QoQ growth %, top 5 regions, % open orders > 30 days.
Data: Orders sheet schema above.
Output:
1) metric definitions in Excel terms
2) suggested chart types
3) one pivot table layout (rows, columns, values)
4) risks if Status values are inconsistent

Expected result: A build checklist you execute in Excel in 20 minutes.

For heavy statistical work (regression, forecasting), ask for Power Query steps or Python in Excel only if your install supports them. GPT-5.5 will sometimes suggest Python; decline if your laptop build does not include it.

Copilot in Excel vs ChatGPT GPT-5.5

Microsoft 365 Copilot in Excel (2026) includes formula generation on the grid, the COPILOT() function for natural-language cells, and Agent mode on desktop for multi-step sheet tasks—availability depends on your tenant license and update channel.

Our June 2026 habit: Copilot for quick fills on safe internal sheets; ChatGPT for tricky logic and documentation text. If Copilot is not enabled in your tenant, ChatGPT with schema-only prompts is still the fallback—not GPT-4o, which is not the current ChatGPT Plus default in 2026.

API route (optional, for repeat jobs)

If you automate monthly reports, use OpenAI API with gpt-5.5 and send JSON schema instead of raw grids:

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

prompt = """
Given columns OrderID, OrderDate, Region, Revenue, Status (Open/Closed),
return Excel 365 formula for column Flag where Flag="Stale" when
Status=Open and OrderDate is more than 30 days before TODAY().
Return JSON: {"formula": "...", "explanation": "..."}
"""

resp = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": prompt}],
    response_format={"type": "json_object"},
)

print(resp.choices[0].message.content)

Expected result: JSON you parse into an internal tool that writes formulas into templates.

Common mistake: Sending full CSV exports with PII to the API. Send schema only.

Pair cheaper models for drafts

Not every sub-step needs GPT-5.5.

See DeepSeek vs ChatGPT vs Claude for coding adjacent workbooks (macros, Office Scripts).

Troubleshooting

Example prompts library

Conditional revenue

Excel 365. Sum Revenue where Region=H2 and OrderDate in current calendar month.
Headers in row 1, data rows 2:500. Return formula for I2.

Clean duplicates

Return formula pipeline: UNIQUE list of OrderID where Status=Open,
then count duplicates per ID in helper column. Prefer dynamic arrays.

Executive one-liner

Write two sentences explaining why open orders >30 days rose,
given these aggregate numbers: [paste only aggregates, no row-level PII].

Structured tables (Excel Tables) with GPT-5.5

Tables make formulas stable. Tell the model you use them:

Excel 365 with Table name "OrdersTable" on Sheet1.
Columns: OrderID, OrderDate, Region, Revenue, Status.
Goal: calculated column "StaleFlag" = TRUE when Status is Open and OrderDate < TODAY()-30.
Use structured references like OrdersTable[Status].

Typical structured formula shape:

=IF((OrdersTable[Status]="Open")*(OrdersTable[OrderDate]<TODAY()-30), TRUE, FALSE)

Expected result: Column fills without breaking when you add rows inside the table.

Common mistake: Mixing raw A1 refs and structured refs in one formula. Pick one style per column.

Power Query handoff (ETL before formulas)

When data is messy CSV exports, ask GPT-5.5 for Power Query steps in plain English, then build in the UI:

Source is CSV with mixed date formats in column C.
Steps needed: promote headers, parse dates, trim Region text, remove duplicate OrderID keeping latest date.
List each step as Power Query UI action names.

Expected result: A checklist you click through in Data → Get Data → From Table/Range.

GPT-5.5 is useful for the recipe. Excel still runs the query locally.

Month-end close workflow (40 minutes)

Never paste full GL detail with account names into consumer chat. Aggregates only.

Office Scripts and macros (when formulas are not enough)

GPT-5.5 can draft Office Scripts TypeScript for repetitive clicks:

Office Script for Excel 365: on Sheet1, hide rows where column E = "Closed"
and column B is older than 90 days. Return count of hidden rows.

Run in Automate tab if your tenant allows scripts. IT policies often block macros; check first.

We still prefer formulas for auditability. Scripts are for one-off cleanup.

Google Sheets variant (one section)

If your team uses Sheets, prepend every prompt:

Google Sheets, not Excel. Use FILTER, QUERY, or MAP/LAMBDA where appropriate.
No XLOOKUP.

Example QUERY shape the model may return:

=QUERY(A1:E, "select C, sum(D) where E = 'Open' group by C label sum(D) 'Revenue'", 1)

Validate column letters match your sheet.

Variations

• Claude: Better long policy memos about spreadsheet controls; paste formulas for critique.
• Gemini in Google Sheets: Native if your org is on Workspace.
• Excel Labs / Python: Use only when enabled; verify export rules.
• DeepSeek Flash: Cheap formula drafts via OpenRouter before GPT-5.5 final pass.

Verdict

Use GPT-5.5 for Excel when you want fast formula drafts, clear explanations, and dashboard planning without hiring a macro consultant for every task. Do not treat the model as a calculator on live confidential data without enterprise controls.

Workflow in one line: schema in, formula out, five tests in Excel, then ship. For model choice across the rest of your stack, start at the latest AI models hub. For research behind market sizing sheets, see Exa MCP setup. Freelancers packaging spreadsheet work as a service should read Make Money with AI Tools (2026).

Changelog

• 2026-06-02: Fact-check. Confirmed GPT-5.5 as ChatGPT Plus default (not GPT-4o/o3); updated Copilot Excel features (COPILOT(), Agent mode); verification date June 2, 2026.
• 2026-05-20: Initial publish. Added safety rules, formula patterns, validation table, Copilot comparison, API optional section, eight FAQs.