Most people who sign up for a ChatGPT course already know how to type a question into the box. They have the account, they’ve asked it to tighten up an email, and then the novelty wears off. The tool sits there doing roughly what a search engine did in 2019 while a colleague in the next cubicle turns around a competitor analysis in forty minutes.
The gap between those two people isn’t talent. One of them has habits: prompts they’ve saved, a way of feeding in their own documents, and a decent sense of when the model is bluffing. A good course installs those habits. A bad one sells you a PDF of prompt templates you’ll open once and forget.
What a ChatGPT course should actually teach
Strip out the branding and the intro slides and there are four skills that matter. If a course doesn’t touch all four, it’s a tutorial, not training.
- Prompt structure. Getting specific about task, format and constraints instead of hoping one sentence does the work.
- Context loading. Pasting in the right background (a style guide, three sample emails, a spreadsheet extract) so the output sounds like your company rather than a chatbot.
- Verification. Knowing which outputs to check line by line, namely anything with numbers, dates, citations or names, and which to skim.
- Repeatability. Turning a one-off win into something you can run again next Tuesday without rebuilding it.
Notice what’s missing: neural network diagrams, the history of transformer architecture, long detours into how training data gets scraped. Useful context occasionally, but nobody’s job got faster because they could sketch an attention mechanism on a napkin.
Prompting is a writing skill in a tech costume
The people who get the most out of ChatGPT tend to be the ones who already write a clear brief. “Summarise this” gets you mush. “Summarise this for a client who has already read the proposal and wants to know what changed since Tuesday, under 150 words, no bullet points” gets you something you can send. Courses that teach prompting as a stack of magic phrases miss this. Courses that teach it as clear instruction-writing stick.
Context beats cleverness
A one-line prompt with no background will lose to a plain-English prompt with three paragraphs of context, every single time. Good instructors demonstrate that live, running the same task twice so you can see the difference. Watch for it in any preview video. If every example is a clever one-liner, the course is teaching parlour tricks.
Three tiers of course, and what the price tells you
Free. There’s genuinely good material out there alongside plenty of dross. An afternoon spent with the good stuff is a fine way to work out what you don’t know yet, and separating free ChatGPT courses worth your weekend from the email traps takes less time than a single paid module.
$20 to $80, self-paced. Where most useful courses sit. Recorded lessons, worked examples, maybe a template pack. Fine value when the syllabus matches your job. The risk is that half of it repeats documentation you could read free.
$500 and up, often cohort-based. Live sessions, a chat channel, homework that gets reviewed. Worth it in exactly one situation: you need accountability and a deadline, and you’ll actually show up. Otherwise you’re paying five hundred dollars for a discussion forum.
The ten-minute test before you hand over a card
You can usually tell from a course page and one preview lesson. Run these checks:
- Does the syllabus name specific tasks (“turning a sales call transcript into a CRM note”) or just “module 3: advanced prompting”?
- Are lessons dated? Anything recorded before mid-2023 is teaching around features that have since been replaced.
- Is there something to do? An exercise, a project, a graded assignment. Watching someone prompt isn’t prompting.
- Does the instructor show a bad output and explain why it failed? Anyone who only shows wins is selling.
- Is there a refund window you’d realistically use? Seven days is reasonable. “All sales final” on a $400 course is a warning sign.
What almost every course skips
The gap in most curricula isn’t technique, it’s judgment. Confidentiality comes up first: pasting a client contract or an HR complaint into a consumer chatbot is a data incident waiting to happen, and a responsible course covers workspace settings, retention and what must stay off the tool. Cost and rate limits come second, because free tiers, paid tiers and API pricing behave completely differently under load. Evaluation comes third, and it’s the one people skip. How do you know the summary is accurate? You check a sample against the source, then keep checking. Courses that dodge this produce confident people shipping unreliable work.
Match the course to your actual job
General ChatGPT training is a decent start, but the payoff usually comes from specialised material. In finance, the interesting part is assumption modelling and variance commentary, not haiku writing, so a generative AI for finance course beats a generic one for the same eight hours. If you write code, look for something that treats the model as a pair programmer with a dodgy memory, which is roughly what a solid generative AI for developers course should do. Starting from zero? Don’t spend much yet. A beginner-level generative AI course will tell you within a week whether this is a tool you’ll use or a tab you’ll close.
Build one thing before you buy another
The best predictor of whether training sticks is whether you have a project. Not a certificate. A thing you actually use: a weekly report you now draft in twenty minutes, a folder of customer feedback the model tags for you, a reusable prompt that turns messy meeting notes into an action list.
That’s also the right lens for comparing courses. Before you buy, ask what you’ll walk out holding. A six-hour course that ends with a working workflow beats a thirty-hour one that ends with a quiz, which is the logic behind ranking generative AI courses by what you’ll actually build rather than by lesson count. Set a deadline too: two weeks, one project, an hour a day. Most people who stall finish the videos and then have nothing to point the skills at.
The week after the last lesson
Courses end with a certificate and a vague sense of momentum, and that is usually where the value leaks away. Schedule the follow-through before you finish. Book thirty minutes on Friday to take the prompt you liked most and put it in a document your team can find. Write two sentences on what the course got wrong, because you’ll want them in six months when someone recommends the same instructor.
Then pick the next boring, repetitive, text-heavy task on your list and run it through properly: real context, output checked against the source, result saved. That loop of choosing a real task, loading good context, verifying and keeping what worked is the entire skill. A ChatGPT course can shorten the learning curve by months. It can’t run the loop for you. The people getting genuine leverage aren’t the ones who finished the most modules; they’re the ones with a growing folder of prompts that earn their keep.

