Is ChatGPT good for Excel?

The Excel Bottleneck: Why Spreadsheets Still Eat Your Time

It's 4 p.m. on a Thursday, and you're staring at a spreadsheet you've opened a hundred times before — trying to remember whether it's XLOOKUP or INDEX-MATCH, wishing you could write the VBA macro that would collapse an hour of copy-paste into a single click. If that sounds familiar, you're not alone, and it's exactly why AI assistance for Excel has shifted from novelty to necessity.

The bottleneck isn't laziness — it's the gap between what spreadsheets can do and what most of us can actually remember how to do:

• Formulas you learned once and forgot, so every lookup becomes a search-engine detour
• VBA and macro automation you know would save hours, but can't write yourself
• Repetitive work — cleaning data, reformatting reports, building the same summary tables week after week
• Scenario models and sensitivity tables that take an afternoon to rebuild every time an assumption changes

What's changed is that the help now exists, and people are actually using it — increasingly, all day. OpenAI's own adoption research found that six months after signup, users send 50% more messages per day and have doubled the number of distinct tasks they try (OpenAI). Usage isn't a phase people grow out of; it deepens as they discover more the tool can handle.

Workplace adoption tells the same story. A Gallup survey found frequent AI use at work nearly doubled from 11% to 19% between 2023 and mid-2025, and OpenAI's research signals show ChatGPT shifting from general content creation toward more specialized, recurring tasks — the kind that live in spreadsheets. AI diffusion overall has been faster than almost any technology in history.

The evidence that this translates to real Excel gains is already here. CPA Don Tomoff of Invenio Advisors used ChatGPT to produce working VBA code for work previously outsourced at "$150 per hour, or up to $2,000 for a typical project." In another case study, an accounting memo drafting process was cut from four hours to 30 minutes with human review. OpenAI even launched an official ChatGPT for Excel add-in in 2026 that writes formulas, traces dependency errors, and runs scenarios in plain English — though it's worth noting that Microsoft Copilot still offers the most native Excel integration for Microsoft 365 users, and both tools run on the same underlying models.

The catch, which every credible source emphasizes: ChatGPT is "accurate enough to accelerate finance work, but not accurate enough to run finance work unsupervised." Think of it as a junior analyst with infinite typing speed — brilliant at drafts, dangerous without verification. And as agent builders know, skills that automate spreadsheet workflows are everywhere but vary wildly in quality; with research suggesting ~26% of published agent skills carry at least one vulnerability, SkillzBench is building a free directory that security-scans, benchmarks, and verifies skills so you can see which ones actually lift performance before you connect them.

The question is no longer whether AI can help with Excel. It's how to use it well — starting with formulas.

What ChatGPT Actually Does for Excel: Formulas, VBA, and the New Add-In

ChatGPT's Excel capabilities are no longer hypothetical — as of 2026, there's an official add-in, and the productivity numbers from real practitioners are hard to ignore.

The official ChatGPT for Excel add-in launched March 5, 2026, reached general availability May 5, and became available across all plans by June 4. According to Coursiv's coverage of the launch, it "writes formulas, traces dependency errors, and runs scenarios in plain English." The framing matters: Excel still does the math — ChatGPT just writes the formula you would have typed. That's an accelerator, not an autopilot.

The most striking evidence comes from VBA automation. Don Tomoff, a CPA at Invenio Advisors, used ChatGPT to produce working VBA code that he previously outsourced at $150 per hour, or up to $2,000 per project. For finance teams, ChatGPT also drafts first-pass scenario models — DCFs and sensitivity tables — though OpenAI's own internal banking benchmark (43.7% on GPT-5, rising to 87.3% on GPT-5.4 Thinking) is vendor-reported and worth treating with caution.

So what does ChatGPT actually do well in Excel?

• Writes formulas from plain-English descriptions via the add-in
• Traces dependency errors that would otherwise take hours to hunt down manually
• Generates working VBA code, replacing expensive developer outsourcing
• Drafts scenario models and sensitivity tables as a starting point for human review

The Copilot comparison deserves honesty. Copilot integrates natively with Microsoft 365 — it can create formulas, apply formatting, and work directly inside your spreadsheet. Zapier's hands-on comparison calls that integration "a killer feature" for people who live in Excel, Word, and Outlook all day. But the same review rates ChatGPT higher on both power (5/5 vs. 4/5) and reliability (4/5 vs. 3/5) — and notes both tools run on the same underlying OpenAI models, so raw formula quality is likely comparable.

The practical takeaway: if you're a Microsoft 365 power user, Copilot's native integration wins. If your work spans many tools, ChatGPT's broader capability set makes it the stronger choice. And if you want to push beyond formulas into full workflow automation, agent skills can handle the rest of the pipeline — SkillzBench is building a free directory where you can browse, benchmark, and connect them via MCP.

The 'Generate, Then Verify' Workflow: Using ChatGPT Without Getting Burned

The single most useful mental model for ChatGPT in Excel is this: treat it as a junior analyst with infinite typing speed and a tendency to invent footnotes when it isn't sure. That framing, popularized in recent finance-team guidance, captures both the upside and the risk — ChatGPT is "accurate enough to accelerate finance work, but not accurate enough to run finance work unsupervised," according to 2026 practitioner coverage.

The workflow that gets results is "generate, then verify." It starts with better prompts, not better outputs:

• Paste the actual data. If you describe your spreadsheet vaguely, ChatGPT will often invent a confident-sounding number to fill the gap. Real rows, real column headers, real values.
• Specify structure and constraints. State the baseline data, time horizon, assumptions, and any limits — the same brief you'd give a new hire.
• Break complex tasks into sequential prompts. Build a DCF or sensitivity table in stages, checking each step before moving on.
• Sanity-check every number. The classic catch: a gross margin that comes back at 1.8% when it should be 18%. Excel still does the math; ChatGPT writes the formula you would have typed — so re-run the math yourself.

This discipline matters because the productivity stakes are real. One CPA reports ChatGPT produced working VBA code that previously cost $150 per hour to outsource — up to $2,000 for a typical project — while an accounting consultant cut memo drafting from 4 hours to 30 minutes, always with human review on top (source). Those gains came from supervised workflows, not blind trust.

Verification is also where the "junior analyst" metaphor earns its keep. You would never ship a junior's model unchecked, and the same standard applies here — especially for financial outputs, where a plausible-looking but wrong number is more dangerous than an obvious error.

One more guardrail: confidentiality. Free, Plus, and Pro plans may use your content for training by default, so anything sensitive belongs on Business or Enterprise plans (or stays out of the prompt entirely), per the same guidance.

Once you've mastered this generate-then-verify loop, agent skills can automate the rest of the workflow — connecting AI agents to your spreadsheets and tools. SkillzBench is building a free directory of vetted agent skills and a free MCP connector, with security scans and benchmark results, so you can see which skills actually lift performance before you rely on them. Join the waitlist to be first in line when it launches.

Privacy, Pricing, and Picking Your Setup

Before you paste that budget model into a chat window, stop and check which plan you're on — the difference between a productivity win and a data-leak headache comes down to your subscription tier.

The privacy rules matter more than most Excel users realize. On Free, Plus, and Pro plans, ChatGPT trains on your content by default (with an opt-out), while Business and Enterprise plans do not, according to practitioner guidance for finance teams. If your spreadsheet contains anything confidential, that's a dealbreaker for personal accounts.

For regulated industries, the constraint is often absolute. Baridhi Malakar, a senior model risk analyst at Western Alliance Bank, puts it bluntly: uploading research reports, investment memos, or draft offering documents to a cloud-based AI tool is "usually not an option," as reported by Coursiv. In banking, healthcare, and legal environments, confidential files stay off-limits entirely.

On pricing, the math is straightforward. ChatGPT Plus runs $20 per month, while Microsoft 365 Business Standard costs $150 per user per year plus a $252 per user per year add-on for Copilot, per Zapier's comparison testing. ChatGPT Pro sits at $200/month, and the lighter ChatGPT Go tier runs about $8/month. Notably, only about 5% of ChatGPT users pay for subscriptions at all, while roughly 40% of US businesses pay for AI products, according to Ramp credit card data analyzed by Epoch AI.

So which setup should you pick? A simple framework:

• Microsoft 365-centric worker: Copilot's native Excel, Word, and Outlook integration is what Zapier calls a "killer feature" if you live in those apps all day — worth the add-on premium.
• Tool-agnostic user: If you bounce between spreadsheets, browsers, and other platforms, ChatGPT rates higher on power and reliability (5/5 and 4/5 versus Copilot's 4/5 and 3/5) and costs less.
• Confidential-data environments: Stick to Business or Enterprise plans, or keep sensitive files out of the chat entirely and work with sanitized, dummy data.

Whichever route you choose, the generate-then-verify workflow stays the same. And once you've settled your toolchain, agent skills can automate the rest of an Excel workflow — SkillzBench is building a free directory and MCP connector so you can browse, security-scan, and benchmark those skills before loading them onto your agent.

Going Further: Automating the Rest of Your Excel Workflow with Agent Skills

Getting ChatGPT to draft your formulas and VBA is only half the workflow. The other half — pulling in fresh data, cleaning it, applying formatting, and generating the weekly report — still eats hours every week. That's where agent skills come in.

An agent skill is a packaged set of instructions that teaches an AI agent to perform a specific task reliably. Instead of pasting a spreadsheet into a chat window and asking for help, an agent with the right skills can handle the surrounding steps: fetching data, transforming it, and dropping results into your workbook. The gains compound. One practitioner documented by finance-focused analysis used ChatGPT to produce working VBA code that previously required outsourcing at $150 per hour — and skills extend that same logic to everything around the spreadsheet.

The catch is quality. Agent skills are proliferating across GitHub and community repositories, but published research found roughly 26% of skills carry at least one vulnerability. Installing an unverified skill into a workflow that touches your financial data is a risk most teams shouldn't take blind.

That's the problem SkillzBench is being built to solve. It's a free, independent directory of agent skills designed to tell you which ones actually work before you trust them with your Excel workflow. Its approach rests on a few pillars:

• Verification: every skill is pinned to a specific version, license-checked, and security-scanned before it's listed.
• Benchmarking: each skill is tested in a sandbox with the same model, with and without the skill, so you can see the measured performance lift in percentage points.
• Trust scores: a 0-100 score combines security, quality, license, maintenance, and community signals — with lift shown separately and money never touching the numbers.
• Free MCP connector: a single tool being built to connect any agent to the skills you choose, with per-agent loadouts so your agent only sees what you've enabled.

Everything on the platform — the directory, the connector, benchmarks, and change alerts — is free, funded by advertising rather than user fees.

This matters now because AI usage in spreadsheets is accelerating. OpenAI's adoption research found that six months after signing up, users send 50% more messages per day and double the number of distinct tasks they attempt. As your Excel-and-AI workflow grows, manual glue work becomes the bottleneck — and verified skills are how you remove it without introducing new risk.

If you want first access when the directory and MCP connector launch, join the SkillzBench waitlist today. Skills, proven — that's the whole promise.