ChatGPT vs Claude vs Gemini for Work: 6 Office Tasks Tested

The three-tool question, honestly

You're paying $20 a month for one of these, or you're staring at three browser tabs and switching between them without a plan. Most head-to-head comparisons of ChatGPT vs Claude vs Gemini for work grade them on coding benchmarks and reasoning tests. That's useful if you write code. It's less useful if your job is writing emails, reading contracts, cleaning up a spreadsheet, or building a deck by Thursday. If the question is whether to pay at all, start with Which AI Subscription Is Worth Paying For? Decide by Workload.

We took six ordinary office tasks and sent the same prompt to each of the three flagship chat products as of September 2026: ChatGPT (GPT-5.6), Claude (Opus 5 on Pro), and Gemini (3.1 Pro on Google AI Pro). No tuning per model, no plugins, no clever tricks. The point is to see what a working professional gets when they paste the same request into three different tools.

Here's what we found, task by task, and where each one fell apart.

The setup: same prompt, three tools

For each task we used the paid consumer tier of each product, since that's where most people at work land. Versions and prices as of September 2026:

• ChatGPT Plus — $20/month, flagship model GPT-5.6, ChatGPT Work agent, file analysis, Deep Research.
• Claude Pro — $20/month (or $17/month billed annually), Opus 5 as the strongest model on this tier, Projects, and Cowork for multi-step tasks.
• Google AI Pro — $19.99/month, Gemini 3.1 Pro with a 1M-token context window, Deep Research, and Workspace integration. (Formerly branded Gemini Advanced.)

We used the default model on the paid tier, no system prompt beyond the task itself, and no follow-up questions. Prompts are printed as code blocks below — copy them into your own account if you want to run the same test.

Task 1: Business writing (cold email that isn't obviously AI)

Write a 120-word cold email to a marketing director at a mid-size B2B
SaaS company. I sell a two-week AI training program for professional
teams. Goal: get a 20-minute call. No exclamation marks. No "I hope
this finds you well." No em dashes. Sound like a person, not a template.

• ChatGPT produced a tight, well-structured email with a clear CTA, but reached for the "I noticed…" opener even after being told not to sound templated, and slipped one em dash in.
• Claude wrote the most natural-sounding draft. Rhythm felt human, the opener was specific to the role rather than a fake compliment, and it stuck to the constraints.
• Gemini was competent but the closest to a template. It over-explained the offer and reached for a two-part sentence structure that reads as AI immediately.

Winner: Claude, by a step.

Task 2: Contract review (finding what's missing)

We pasted a 4-page Master Services Agreement into each chat and ran:

Review this MSA. List:
1. Clauses that are unusually vendor-friendly.
2. Anything a buyer's counsel would push back on.
3. Missing clauses that a mid-market SaaS buyer would expect.
Cite the section number for each point. Do not summarize the contract.

• Claude returned the most useful review. It flagged the missing data-processing terms, questioned an uncapped indemnity clause, and cited section numbers correctly.
• ChatGPT covered similar ground but hallucinated one section reference and skipped the DPA gap.
• Gemini was thorough on the summary side (which we told it not to do) and its citations were the least reliable.

Winner: Claude. Nothing here replaces a lawyer. A 2026 benchmark of AI contract-review tools found the strongest one scored 65% on redlining accuracy against 79.7% for the lawyer control group, even while beating lawyers on structured data extraction. Use these outputs as a first pass, not a signature.

Task 3: Spreadsheet formula (a real one, not a demo)

I have a Google Sheet with columns A (date), B (channel), C (spend),
D (leads). I need a formula in column E that shows cost per lead for
the last 30 days per row, and a summary formula in G2 that shows total
cost per lead by channel for the last 30 days. Give me two formulas
and explain what each one does in one sentence.

• Gemini delivered here. It suggested a clean QUERY formula for the summary and got Sheets syntax right the first time. Workspace context helps: it knows where you are.
• ChatGPT produced a working SUMIFS version but mixed in Excel-only functions that Sheets doesn't accept.
• Claude wrote correct formulas but was more conservative, splitting the summary across two cells when one would have done.

Winner: Gemini for Google Sheets. If your team lives in Excel, ChatGPT edges ahead, and Microsoft Copilot inside Excel is a separate conversation.

Task 4: Market research (30 minutes, not 30 hours)

Using each tool's Deep Research mode where available:

Research the market for AI-assisted contract review tools as of
September 2026. Cover: main vendors, typical pricing bands, the gap
between vendor claims and independent benchmarks, and the top three
objections buyers raise. Cite every claim with a link.

• ChatGPT Deep Research returned a 12-page report with a strong bibliography and a coherent structure. Two citations pointed to pages that didn't say what the report claimed. Read every source.
• Gemini Deep Research was faster and pulled in more sources, but was thinner on the objections section.
• Claude doesn't have a native Deep Research equivalent on Pro at the same depth, but its answer with web search was the most honest about what it didn't know.

Winner: ChatGPT for depth, Gemini for speed, Claude for calibration. This is a task where running two tools in parallel and comparing is worth the extra ten minutes.

Task 5: Presentation outline (that a manager would actually use)

Draft a 10-slide outline for an internal presentation to a leadership
team. Topic: our team's plan for adopting AI in the next quarter.
Audience: skeptical. Each slide: title, three bullets, one speaker
note. Include one slide with risks and how we'll handle them. No
filler slides.

• ChatGPT produced the most immediately usable outline. The structure felt like something a manager would present without heavy editing.
• Claude was more thoughtful on the risks slide and wrote better speaker notes, but the structure was slightly academic.
• Gemini exported cleanly to Google Slides, but the copy needed the most rewriting.

Winner: ChatGPT for the draft, Gemini if you want it in Slides in three clicks.

Task 6: The "AI smell" test

Same paragraph, same prompt, all three tools:

Rewrite this paragraph to sound like a human wrote it in one sitting.
Remove: "In our modern workplace," any sentence starting with
"Moreover," any "It's not just X, it's Y" construction, and any
three-item list in a row. Keep the meaning. Length within plus or
minus 10 percent.

[paragraph pasted here]

• Claude produced the least AI-flavored rewrite. It broke the rhythm of the original and used shorter sentences without sounding choppy.
• ChatGPT followed the rules but reintroduced its own tells: heavy use of "That said," and a habit of ending paragraphs on a rhetorical question.
• Gemini was the most literal. It stripped the banned phrases but left the underlying cadence intact, so the paragraph still read as generated.

Winner: Claude. For a longer take on editing AI writing without paid "humanizers," see How to make AI writing sound human.

Quick comparison

Task	Best tool	Runner-up
Business writing	Claude	ChatGPT
Contract review (first pass)	Claude	ChatGPT
Google Sheets formulas	Gemini	ChatGPT
Deep research report	ChatGPT	Gemini
Presentation outline	ChatGPT	Gemini
Removing AI smell	Claude	ChatGPT

Two patterns show up. Claude wins on writing and judgment tasks. ChatGPT is the strongest generalist and the best all-round default. Gemini is the tool of choice when your work lives inside Google Workspace or when you need speed over depth.

Where each one breaks

ChatGPT hallucinates confidently in research mode. It also has the strongest voice tics — the pivot sentence, the rhetorical question, the three-part list — so anything you publish needs an editing pass. Sites and the ChatGPT Work agent, both Plus features as of September 2026, are useful but not what most professionals need first.

Claude doesn't ship image generation on Pro, its tool ecosystem is smaller than the other two, and its Deep Research equivalent isn't at parity. Cowork, Anthropic's agentic feature opened up to Pro users in January 2026, helps with multi-step tasks but still needs supervision.

Gemini writes the most template-like copy without heavy prompting. Its Workspace integration is the strongest reason to pay for it, and also why it's the safest choice if your company already runs on Workspace. As of September 2026, Gemini is bundled into paid Workspace plans from Business Standard up, not sold as a separate add-on.

None of the three should see personal data, NDA material, or client-confidential documents unless you're on an enterprise plan with the right agreements in place. That's a policy question, not a product one.

What this costs (September 2026)

• ChatGPT Plus: $20/month.
• Claude Pro: $20/month monthly, $17/month annual.
• Google AI Pro: $19.99/month, bundled with 5 TB of Google One storage.
• Google AI Plus: $4.99/month, the cheapest paid tier of the three.

If you're weighing whether any of these is worth paying for at all, the honest answer is: yes, if you use it daily; otherwise the free tiers are enough.

The realistic answer: don't pick one

The professionals we work with who get real value from AI at work don't run one tool. They route different tasks to different tools: writing and judgment to Claude, deep research and general-purpose work to ChatGPT, anything inside Google Workspace to Gemini. If you can only afford one paid subscription this quarter, Claude Pro or ChatGPT Plus give you the widest range of professional tasks well-covered; add Gemini's free tier for spreadsheet and Workspace work.

The bigger difference isn't the model. It's how you write the prompt. A well-scoped prompt gets a usable answer from any of the three; a lazy prompt gets a mediocre answer from all three. If prompt-writing is where you keep stalling, we walk through the four elements that make prompts work in How to write better AI prompts for work.

Where the practicum picks this up

Our two-week practicum spends the first session on choosing the right tool and the anatomy of a working prompt, the second on writing and business correspondence without the AI smell, and the third on documents, numbers, and research — the same six tasks above, with your own materials in your own account. You leave with a personal AI stack you actually use, not a tour of features. Cohort dates at /#program.