Short answer: both ran the account well, and they won different rounds. ChatGPT was faster and needed less hand-holding on the quick analysis prompts. Claude took longer on everything, then pulled ahead decisively on the biggest task of the test, turning the account into full report documents at a level ChatGPT didn't match.
Here's the test behind that answer. We connected the same Jobber account to ChatGPT and Claude through the same private MCP connector, opened them side by side, and ran the same business-analysis prompts in both. Same data, same wording. The full run is in the video below, and every prompt is in this post with a copy button so you can run the exact same audit on your own account.
Watch the full side-by-side:
How we set up the test
The account belongs to Eden's Garden, a landscaping business running its operation in Jobber: requests, quotes, jobs, invoices, and payments. Both assistants connected through the same JobberToClaude connector, which means both had identical access to identical data.
One thing to know before the results: the models were not an exact match, on purpose. ChatGPT ran on its fast Instant model. Claude ran on Opus 4.8, its deepest model. That's the realistic setup, since it's what each app hands you when you open it and start working. It also explains the pattern you'll see all the way through: ChatGPT answers in seconds, Claude thinks in minutes. Keep that trade in mind as you read.
Round 1: CEO Audit
“Analyze our entire Jobber account as if you were the CEO we just hired. What are the five biggest problems you would solve first? Use specific evidence from our data.”
ChatGPT took this round. It came back quicker, organized its five problems more cleanly, and supported each one with more examples and numbers pulled from the account. Claude's answer was solid but thinner, and it took noticeably longer to arrive.
Our honest read: Claude's output here probably improves a lot with a more specific prompt. But that's exactly the point of this round. Given the same open-ended question and no coaching, ChatGPT made more of it.
Round 2: Profit Leak Detector
“Identify every area where we're likely losing profit. Look for pricing inconsistencies, underpriced jobs, inefficient workflows, callbacks, discounts, labor inefficiencies, or anything else costing us money. Quantify the impact whenever possible.”
Same story, wider gap. ChatGPT returned significantly more findings and more data in a single pass: more categories of leakage, more numbers attached. Claude sent over less. Follow-up prompts would likely have closed the distance, but on a one-shot basis, ChatGPT clearly gave the owner more to act on.
Round 3: Pattern Recognition
“What patterns do you notice across our jobs, customers, quotes, invoices, and payments that the average business owner would completely miss?”
For this round we gave Claude some extra instruction on top of the base prompt, asking it to be more descriptive and reference specific records. The extra guidance helped, and it previews the theme of this whole test: Claude rewards steering.
Then came the best moment of the test. ChatGPT flagged that a business policy in the account had recently changed. Nobody told it that. It noticed the before-and-after difference in the data and called it out. We didn't even know the change was there until it surfaced it. Claude caught the same shift, but ChatGPT laid it out in more detail. Slight edge to ChatGPT, and a genuinely impressive round for both, because spotting a quiet policy change across hundreds of records is exactly the thing a busy owner would miss.
Round 4: The executive deep dive
For the final round we rolled the remaining prompts into one big assignment: pricing, customers, funnel, growth, blind spots, and the full executive report, in a single run.
“Create an executive report on this business. Include strengths, weaknesses, risks, opportunities, operational issues, pricing issues, customer insights, and your top recommendations for increasing profit.”
This is where the two assistants showed completely different personalities. ChatGPT worked for about eleven seconds and came back with a finished, confidently written executive report, leading with its single strongest conclusion about the business. Good output, fast.
Claude did something else. It paused before writing and said it wanted to widen its data pull first, so the customer and pricing sections would rest on the full job history instead of just the most recent jobs. Then it went back into the account, pulled deeper, and produced actual report documents, loaded with data. It took far longer. It was also, clearly, the better deliverable. This was the most demanding prompt of the test, and Opus 4.8 treated it like the assignment it was.
ChatGPT did well here. Claude was on another level.
The scorecard
| Round | Winner | Why |
|---|---|---|
| 1. CEO Audit | ChatGPT | Faster, better organized, and backed its points with more examples from the data. |
| 2. Profit Leak Detector | ChatGPT | Simply returned more findings and more numbers in one pass. |
| 3. Pattern Recognition | ChatGPT, barely | Both caught a policy change in the account. ChatGPT explained it in more detail. |
| 4. The executive deep dive | Claude | Took far longer, but produced full report documents at a level ChatGPT didn't reach. |
What we think is actually going on
A pattern held across every round: ChatGPT needed no guidance, and Claude wanted some. Our working theory is that Claude followed each prompt closely as written, while ChatGPT filled in the gaps the prompts left open. These prompts were intentionally open-ended, which plays to a gap-filler's strengths on the quick rounds and to a careful reader's strengths on the big one. That's hard to prove from one test, so take it as a theory, not a finding. But it matches what we see day to day running both.
So which one should you use?
Use ChatGPT when you want fast answers with zero prompt-crafting: quick audits, daily questions, first passes. Use Claude when the output matters more than the wait: full reports, deep analysis, anything you'd hand to a partner or a banker. And if you use Claude, give it direction, because it gets noticeably better when you tell it exactly what you want.
The part that matters either way: this was one connector, one subscription, two assistants. You don't have to pick. Here's how to connect Jobber to Claude and how to connect Jobber to ChatGPT.
The full prompt list
Every prompt from the test, ready to copy. Run them one at a time, or stack the last several into one big assignment like we did in round four.
1. CEO Audit
“Analyze our entire Jobber account as if you were the CEO we just hired. What are the five biggest problems you would solve first? Use specific evidence from our data.”
2. Profit Leak Detector
“Identify every area where we're likely losing profit. Look for pricing inconsistencies, underpriced jobs, inefficient workflows, callbacks, discounts, labor inefficiencies, or anything else costing us money. Quantify the impact whenever possible.”
3. Pattern Recognition
“What patterns do you notice across our jobs, customers, quotes, invoices, and payments that the average business owner would completely miss?”
4. Pricing Intelligence
“Analyze all completed jobs and identify opportunities where we could increase prices without significantly affecting close rates. Support every recommendation with examples from our Jobber data.”
5. Customer Intelligence
“Who are our ideal customers based on historical data? Tell me exactly what characteristics our best customers share and who we should target more often.”
6. Sales Funnel Analysis
“Analyze our sales process from request → quote → approval → completed job → payment. Identify every bottleneck that's slowing revenue growth.”
7. Growth Opportunities
“If our goal is to double revenue next year, what opportunities already exist inside our Jobber account that we're not taking advantage of?”
8. Executive Business Report
“Create an executive report on this business. Include strengths, weaknesses, risks, opportunities, operational issues, pricing issues, customer insights, and your top recommendations for increasing profit.”
9. Hidden Opportunities
“What questions should I be asking about my business that I'm not asking? Use the Jobber data to uncover hidden opportunities or blind spots.”
10. The Million-Dollar Prompt
“Pretend you've spent the last five years running this company and know everything inside Jobber. What advice would you give me that could realistically make or save me over $1 million in the next five years? Be brutally honest and support your conclusions with evidence from the data.”
Frequently asked questions
Do I need separate subscriptions for ChatGPT and Claude?
Not for the Jobber connection. One JobberToClaude subscription gives you a private connector that works in ChatGPT, Claude, and any other MCP-compatible assistant. The AI plans themselves are separate: ChatGPT needs a paid plan with Developer Mode for custom connectors, and Claude needs a paid Claude plan.
Did both AIs use the same data?
Yes. Both were connected to the same Jobber account through the same private MCP connector, and the prompts were run word for word in both, in fresh chats.
Were the models an exact match?
No, and that's on purpose. We ran each assistant the way an owner actually would: ChatGPT on its fast Instant model and Claude on Opus 4.8, its deepest model. That's why ChatGPT consistently answered faster and Claude consistently dug longer. It's a real-world comparison, not a lab benchmark.
Can I run these prompts on my own Jobber account?
Yes. Every prompt in this post has a copy button. Once your Jobber account is connected, paste them into ChatGPT or Claude and you'll get the same kind of analysis on your own data.
Will these results change over time?
Probably. Both companies ship model updates constantly, so we plan to re-run this test periodically and update the results. The date on this post reflects the most recent test.
Run the audit on your own account
Everything in the video came from pasting a prompt and hitting enter. No reports, no exports, no spreadsheet work. Connect your Jobber account once and both assistants are ready to dig.
See what's hiding in your own Jobber account
We ran this on a real business and found $139K in late jobs and a $25K monthly leak. Start your 7-day free trial and find out what yours is sitting on.
- Your own dedicated MCP server, fully managed
- Secure Jobber login, no passwords stored
- Works for every crew member, no per-seat fees
- Scheduling, quotes, invoicing, clients, and reports by chat
No setup fee and no per-task charges. One predictable subscription, no matter how hard your team uses it.