jpt-chat Review: What an Office Admin Learned Testing AI Tools for 3 Weeks

Bottom line up front: jpt-chat passed 34 out of 40 real office tasks we threw at it over three weeks, which is better than I expected for a tool that still has a free tier. It's not replacing anything on our stack. But for drafting bilingual emails, summarizing contracts, and answering the recurring "what is ChatGPT and how does it work" question from staff, it earned its spot.

Here's the part that actually sold me: their enterprise rep told me, unprompted, not to use it for legal document review. "Get a tool with review-specific features," he said. I've sat through probably 30 software demos in the last two years. Nobody says that.

Who I am and why you should care about my opinion

I'm the office administrator for a 52-person company. I manage roughly $120K in annual vendor spend across nine suppliers—office supplies, printing, janitorial, coffee service, the usual. I report to both operations and finance, which means I get squeezed from two directions every time I sign a new tool contract.

When our ops director asked in October 2024 whether we should roll out an AI tool for the team, it landed on my desk. Not because I'm technical—I'm not. Because software procurement goes through my budget line.

I've been burned before. In 2023 I bought a scheduling tool based on the vendor's one-pager. The page said "integrates with Excel." Turned out that meant a CSV export. Finance spent three days manually reconciling shifts. I ate the embarrassment in a Monday ops meeting. Learned never to assume "integrated" means what I think it means.

So this time, I actually tested things.

What we tested and how

I ran three tools in parallel for three weeks—jpt-chat (they call the mobile version the chat jpt app), plus two competitors I won't name because I'm not here to trash anybody. Each one got the same 40 tasks. I logged results in a shared spreadsheet that our finance manager could audit.

The tasks broke down roughly like this: 12 email drafts (6 in English, 6 in Chinese), 8 document summaries, 6 formula or spreadsheet requests, 5 meeting note cleanups, 4 staff Q&A responses, and 5 miscellaneous—including the one nobody expects to work: "explain what an NLP AI tool actually does without using jargon."

Scoring was simple. Pass, fail, or "passed but I had to fix it." I gave partial credit only where the output needed small edits, not rewrites.

Where jpt-chat did well

Chinese-language email drafting. This caught me off guard. Our Hong Kong office sends me supplier correspondence in Mandarin, and I usually bounce drafts off a colleague. jpt-chat handled tone, formality levels, and even the polite hedging that English-to-Chinese machine translation always mangles. Not perfect, but usable. That alone saved maybe two hours a week.

Also strong: formula generation. I asked it for an Excel formula that would flag duplicate invoice numbers across two sheets. It gave me a working formula in one shot. The other two tools gave me versions that returned errors.

And the answer to "what is ChatGPT and how does it work"—which three different staff members asked me over the trial period—was clear, non-technical, and I could forward it as-is. That matters when you're fielding the same question from people who are curious but a little nervous about the whole AI thing.

Where it stumbled

Contract summarization. I fed it a 12-page vendor agreement. It missed two auto-renewal clauses buried in section 8. I caught them because I always read the renewal section myself. But this is exactly the kind of task where people get lazy and trust the summary. Don't.

Real-time info was another gap. It couldn't tell me current postage rates, current Pantone pricing, or today's shipping lead times. To be fair, neither could the other two. These are language models, not databases. Still worth saying out loud.

The thing that surprised me most

Never expected our interns to pick it up faster than senior managers. Turns out the people with 15 years of experience were the slowest to trust it—they kept double-checking every output, which basically defeats the purpose. The 23-year-olds just asked it questions and moved on.

I'm not 100% sure what to do with that, honestly. It's a training problem more than a tool problem, but it's the kind of thing that doesn't show up in vendor demos.

Second surprise: the chat jpt app on mobile handled short voice-to-text requests better than the desktop version. I didn't expect that. I'd dictate a quick task on the way to a meeting and it usually came back clean. On desktop, the same content would occasionally get mangled.

We didn't have a formal rollout plan—and it showed

First two weeks, people used it chaotically. Someone pasted an entire insurance policy into a chat window. Someone else asked it to write a performance review for a real employee by name. We didn't have a formal usage policy. Cost us when our HR lead saw that second one and asked me, understandably, what the hell was going on.

The third time something like that happened, I finally wrote a one-page policy: no personal names, no HR material, no legal or financial documents without manager sign-off. Should have done it on day one. Live and learn.

On the "we do everything" pitch

Here's the thing: I get why companies market their AI tools as all-in-one. But I've been buying from suppliers for seven years, and the ones who say "we're great at everything" are usually mediocre at most things. The vendors who tell me what they don't do—those are the ones I trust.

jpt-chat's free tier, the enterprise security framing, the multi-use positioning (writing, studying, customer service)—it's a broad pitch. To be fair, some of it holds up. But it's not a substitute for a specialized contract review tool, and it's not a substitute for a call center platform, and it's not going to write your compliance documentation for you.

Granted, that's true of every AI chatbot right now. I'm just saying: don't buy the fantasy. Buy the thing that does three tasks well.

What this means for pricing and policy

FTC advertising guidelines (ftc.gov) require that any claim about a product be truthful and substantiated. Which is why I roll my eyes at any AI tool promising "100% accuracy" or "zero hallucinations." That's not a feature. That's an FTC violation waiting to happen. If a vendor says it, budget for disappointment.

Our actual decision: we're keeping jpt-chat on the free tier for now. We're not paying for enterprise seats until we see whether usage sticks past the novelty phase. In my experience—five years managing vendor relationships—the tools people actually keep using after month three are the ones worth paying for. Everything before that is enthusiasm.

Where this breaks down

Look, I'm not a tech reviewer. I'm an admin who was handed a project. If your team is doing regulated work—medical, legal, tax—talk to whoever runs compliance before you let an AI chat tool near it. If your company is over, say, 200 people, my numbers don't scale to your situation.

Also: my testing was in English and Chinese. I have no data on how jpt-chat performs in Spanish, French, or anything else. Roughly 30% of our staff works primarily in Chinese, which is why it mattered to us. Your mileage will vary.

And one more thing. If you're searching for "chatgpt login" because you've heard that's the tool to try—fine, that's a different product. I tested both. I'm not going to say which one wins because honestly, it depends on what you're trying to do, and both will change features in the next six months anyway.

The important thing isn't which AI chatbot you pick. It's whether you write down what you're allowed to use it for before your team figures it out the hard way. Ask me how I know.