ChatGPT launched at the end of November, and my timeline has not been quiet since. Half the posts say it will replace every knowledge worker by Tuesday. The other half say it is a party trick. Both are wrong, and I would rather tell you what I actually found than add to the noise.
I spent two weeks putting ChatGPT for business tasks through real work, not demos. I drafted marketing copy with it, summarized long documents, asked it to explain unfamiliar code, and tried to break it on facts. This is the sober second piece, written by someone who ships software for a living and has no product to sell you.
Short version: it is a genuinely useful drafting assistant with a lying problem. Treat it that way and it earns its keep. Wire it directly to your customers right now and it will eventually embarrass you.
Where It Genuinely Delivered
Some tasks it handled well enough that I have kept using it after the experiment ended.
- First drafts of copy. Product descriptions, email outlines, ad variations. The output is rarely final, but starting from a decent draft instead of a blank page is a real speed-up. I rewrite maybe forty percent of it, which is still faster than writing all of it.
- Summarizing long text. Paste a rambling meeting note or a dense document and ask for the key points. It is reliably good at compression, because it is working from text you gave it, not from its own memory.
- Explaining code and concepts. When I hit an unfamiliar library, asking for a plain-language explanation got me oriented faster than digging through scattered docs. I still verify, but it is a good tour guide.
- Reformatting and restructuring. Turn these bullet points into a table. Rewrite this in a more formal tone. Boring, mechanical, and it nails it every time.
The common thread: it shines when the correct answer is a matter of style or structure, not fact.
Where It Quietly Lies
Here is the part the hype skips. ChatGPT will produce a confident, fluent, completely wrong answer and give you no signal that it is wrong. In the industry people call this hallucination. In plain terms, it makes things up.
I asked it factual questions where I already knew the answer. It invented statistics. It cited a regulation detail incorrectly. It described a feature of a tool that does not exist. Every one of these was delivered in the same calm, authoritative tone as its correct answers. That is the danger. A junior employee who is unsure at least sounds unsure. This does not.
For a business, that means one hard rule for now: never let it answer factual questions on your behalf without a human checking. Not to customers, not in a contract, not in anything where being wrong costs you.
The Customer-Facing Trap
The tempting move is to bolt ChatGPT onto your customer service. Do not do that yet, and I say that as someone who builds automation for a living.
Think through the failure mode. A customer asks about your return policy. The model, having no idea what your actual policy is, confidently invents one. Now you either honor a policy you never wrote or call your own customer a liar. Both are bad. The model does not know your prices, your stock, your rules, or your promises, and it will happily pretend it does.
There are ways to constrain a language model to your real data, and those approaches will mature. But an off-the-shelf chatbot answering customers from its own imagination is a lawsuit or a viral complaint waiting to happen. If you want to understand where this is heading in a more controlled form, I sketched the direction in AI agents explained: what business owners need to know.
How I'd Actually Use It Today
My working policy after two weeks is simple:
- Internal drafts, yes. Copy, summaries, outlines, brainstorming. A human always edits before it leaves the building.
- Facts, always verify. Treat every specific claim as unverified until you check it.
- Customer-facing, not yet. Not until it is constrained to your real data and a human is in the loop.
- Sensitive data, careful. Do not paste customer records or confidential contracts into a public tool. You do not fully control where that goes.
Used inside those lines, it is a legitimate productivity tool. Your marketing person drafts faster. Your ops person digests documents faster. That is a real gain, and it is available today.
The Takeaway
The honest verdict on ChatGPT for business is unglamorous: it is a fast, tireless, slightly unreliable intern. Brilliant at first drafts, dangerous when trusted with facts, not ready to speak to your customers unsupervised.
The businesses that win with this will not be the ones that believe the hype or dismiss it. They will be the ones who figure out exactly which tasks it does well, put a human check on everything else, and quietly get faster while their competitors argue online.
Start small, keep a human in the loop, and let the tool prove itself on low-stakes work before you trust it with anything that matters.