ChatGPT is the tool a colleague quietly used to write the email you complimented them on. It’s also the tool that invented a court case that never existed and got two New York lawyers sanctioned in 2023. Both things are true, and together they explain the product better than any feature list.
I’ve used it most days for about two years now, for client work, admin, code snippets and the occasional awkward message to a landlord. Here’s the honest split.
What ChatGPT genuinely does well
Drafting and rewriting at a pace you can’t match
A 400-word email with three competing asks in it can become 90 clear words in about fifteen seconds. That’s not glamorous, but it’s the use case that survives contact with real work. The same goes for turning a messy list of bullet points into readable prose, shifting tone from blunt to diplomatic, or writing the fourth version of a paragraph you’ve been staring at since Tuesday.
It’s also surprisingly good at the small stuff people forget to mention: summarising a long thread, generating fifteen variations of a subject line, or converting rough notes into a structured table.
It explains things at whatever level you ask for
Ask it to explain compound interest like you’re a smart fifteen-year-old and you’ll get something useful. Ask again for the version a finance director would accept and you’ll get a different answer. That flexibility is genuinely hard to find elsewhere. In practice, it’s replaced a lot of frantic forum searching for spreadsheet formulas, regex patterns, contract boilerplate and Excel errors that make no sense.
The free tier is no longer a demo
You can do a lot without paying. Free accounts get a capable model, file uploads, image generation in limited daily quantities and voice conversation. If your needs are occasional, the ceiling is high enough that the paid plan buys you speed and volume more than raw capability. It’s worth comparing what free tiers across the big chatbots actually include before you hand over a card, and this breakdown of free AI tools and their real limits covers the fine print well.
Where it falls down, specifically
It states wrong things with total confidence
The Mata v. Avianca case is the classic example: lawyers submitted a brief citing six cases that ChatGPT had invented, complete with plausible names and citations. That wasn’t a bug from 2015. It’s the core behaviour of a system predicting likely-sounding text rather than checking facts.
In everyday use this shows up as invented statistics, misremembered dates, fake product specs and book quotes that don’t exist. Anything factual that matters needs a second source. Treat its output as a first draft from a confident intern who doesn’t flag uncertainty.
It agrees with you too easily
Push back on an answer and it will often fold, apologise and produce a new answer that matches your objection even when the original was right. That’s pleasant and terrible for decision-making. If you need a genuine critique of a plan, ask it to argue the opposite case or list the three strongest reasons you’re wrong. Framing it as a devil’s advocate gets much better results than asking “is this good?”
Long conversations drift
Context windows are generous now, in the hundreds of thousands of tokens on paid plans, but attention thins over long chats. Instructions from an hour ago get ignored, earlier restrictions get forgotten, and the writing style slowly flattens. Starting a fresh chat with a clean, well-written brief usually beats nursing a 40-message thread.
Privacy deserves more thought than most people give it
Don’t paste in client contracts, medical details, salary data or anything covered by an NDA unless your organisation has cleared it. Consumer accounts have historically fed conversations into training by default, and while there are opt-outs and enterprise tiers with stronger guarantees, the safe habit is simple: if you’d hesitate to email it to a stranger, rewrite it before you paste it.
Everything starts to sound the same
Left to its own devices, ChatGPT produces a recognisable register: tidy triads, hedged claims, the word “delve”, and a fondness for bullet points where paragraphs would do. Readers notice. Use it for structure and speed, then put your own voice back in before anything ships.
Is the $20 a month actually worth it?
Plus gives you access to the stronger models, higher message limits, file and image analysis, custom instructions, voice mode and early access to newer tools. For anyone using it more than a few times a day on real work, it pays for itself in the first week. For someone who opens it twice a month to reword an email, it’s a $240 annual subscription to a problem you don’t have.
A reasonable middle path: use the free tier for a fortnight, note every time you hit a limit or feel the model is too weak, and let that list decide. There’s a wider comparison of which paid AI subscriptions are worth the money for beginners if you want to see where ChatGPT sits against the rest.
How it stacks up against the alternatives
ChatGPT isn’t the only sensible choice anymore. Claude tends to handle very long documents and nuanced writing tone better, which matters if you’re summarising 80-page reports or editing prose that needs to sound human. Our Claude review for 2026 digs into where it wins and where it doesn’t. Gemini is tightly woven into Google Workspace, which is a genuine advantage if your whole working life lives in Docs and Gmail. Copilot makes more sense inside Microsoft 365.
The differences between the top models are narrowing faster than the marketing suggests, and most people’s results hinge on how well they brief the tool rather than which logo is on it. That gap between beginner and experienced users is explored properly in this comparison of beginners versus experienced users, and it’s a more useful thing to study than a feature matrix.
Who should pay, and who should walk away
- Worth it: writers, marketers, developers, analysts and anyone whose day involves producing text, code or summaries at volume.
- Worth it: small business owners doing their own admin, proposals and customer replies.
- Probably worth it: students using it as a tutor rather than a ghostwriter, with the caveat that you learn nothing by outsourcing the thinking.
- Skip it: anyone whose employer already provides an approved enterprise tool. Duplicating subscriptions is how IT departments develop grudges.
- Skip it: people who want reliable facts with no verification step. It is not a search engine and never quite behaves like one.
A one-afternoon test before you commit
Don’t judge ChatGPT on clever demo prompts. Give it three tasks from your actual job and score the output honestly.
First, something you know cold: a document type you’ve written a hundred times. You’ll spot invented details instantly, which tells you how much supervision everything else needs. Second, something you don’t know well: a technical question in an unfamiliar field. Check the answer against a source you trust. Third, a long input: paste in a 20-page report and ask for a one-page brief with the five decisions it implies. If the summary misses the point that matters most, the tool isn’t ready for that part of your work.
Twenty minutes of that beats a month of reading reviews, including this one. ChatGPT is a genuinely useful instrument with a specific failure mode, and the people who get the most from it are the ones who’ve worked out exactly which of their tasks sit on the safe side of that line.