Prompt Engineer Course Options, Ranked by What They Skip

Eight results sat on the front page for prompt engineer course the day I looked. A Google Cloud explainer, Udemy, a Reddit thread asking which one is best, Coursera's search results, AWS Skill Builder, Codecademy, Learn Prompting, and an IBM course. I opened every one that would let me in, which turned out to be most but not all of them.

Short version if you're deciding this afternoon. The free courses teach the same material as the paid ones, so what you're buying with money is a certificate and some structure, and both are worth roughly what you'd guess. The bigger problem is coverage. Of the courses I could load and read properly, not one of them teaches image or video prompting, and that's the part where the model vendors publish better material for free than any course I found.

Here's what each one actually contains, what it costs, and the syllabus I'd write instead.

What The Front Page Is Selling

I pulled the stated length and price off each course page myself rather than trusting a roundup. Two of them blocked me, which I've marked honestly rather than filling in from somewhere else.

Two notes on that table. The Udemy row is secondhand, because Udemy returned a 403 to me on three different course URLs today, so the price and the hours come from a TechRepublic roundup rather than from the seller, and I'd treat the $119.99 as a list price that goes on sale constantly the way everything on that platform does. The AWS row is blank because the Skill Builder page rendered as a bare heading for me and I'm not going to describe a course I couldn't open.

Coursera's own search for this phrase paginates out past 500 pages, which tells you something about how crowded this is.

Everybody Converged On The Same Twelve Ideas

Read four of these back to back and the repetition gets funny.

Zero-shot. Few-shot. Chain of thought. Assign a role. Give context. Specify the format. Break the task into steps. Iterate. Use delimiters or tags to separate your material from your instructions. Ask it to ground answers in quotes. Provide examples of the output you want. Say what to avoid.

That's it. That's the curriculum, across a three-hour course and a twenty-hour one, and the twenty-hour one is mostly more worked examples of the same twelve moves. MIT Sloan's guide, which ranks near the top for a lot of the neighbouring queries, compresses the whole thing into three strategies and one table of six prompt types, and honestly you could learn eighty percent of the value from that table plus an afternoon of actually sending prompts.

Anthropic's own documentation says something quietly damning about all of this. Their prompt engineering overview opens by assuming you already have a clear definition of success criteria, a way to empirically test against them, and a first draft prompt to improve. In other words the technique list isn't the hard part. Knowing whether the output got better is the hard part, and no course I looked at teaches that, because teaching it would mean giving you a real task with a real standard attached.

The Half That Isn't Taught Anywhere

Search volume for image prompting phrases dwarfs the course volume, and yet the courses ranking for this keyword mostly pretend that generation doesn't exist.

I find this genuinely strange. Google publishes a whole set of named image prompt templates in their Gemini image generation docs, thirteen of them by my count, covering photorealistic scenes, stylized illustrations and stickers, accurate text inside an image, product mockups, minimalist negative-space compositions, sequential comic panels, style transfer, inpainting, and a few more. Each one comes with a fill-in-the-blank skeleton. The stylized illustration one hands you four labelled slots and asks you to fill them in order, style, then the subject with its accessories and whatever it's doing, then the visual qualities such as outline weight and shading, then colour and background. That's a teaching artifact. It's free, it's from the people who trained the model, and no course on the front page for this phrase mentions it.

The one exception in my table is that Udemy bootcamp, which per the review I read does cover Midjourney and Veo alongside the text stuff across 20-plus projects. I can't confirm that firsthand because of the 403, so treat it as a lead rather than a recommendation.

If image and video is what you're here for, the vendor documentation plus a lot of rerolling is the actual course. I've written up the parts I've worked out in ai image prompt and ai video prompts.

The Syllabus I'd Write Instead

Ten exercises. Each one has a stop condition, because "learn prompting" without an acceptance criterion is how people spend six weeks feeling productive.

Exercise seven is the one I'd defend hardest and it's the one nobody assigns, because it's boring and it takes an afternoon. Generation is random, so each version needs running twice before you decide the word did anything, otherwise you're crediting your edit for a coin flip. I worked through a version of that on my own machine and came out with a shortlist of maybe fifteen terms whose behaviour I'd actually bet on, which is worth more to me than any certificate would be.

Exercise ten is there because every course teaches addition and none teach subtraction. Long prompts feel like effort. Past a certain density the descriptors stop competing and start blending, and you get a soft picture with no single word to blame for it.

The whole list is maybe eight to twelve hours of work depending on how thorough you are, which puts it between the IBM course and the Vanderbilt one on time, and ahead of both on transferable skill. That's my claim and I'd expect an argument about it.

What The Certificate Is Actually For

Nobody has ever asked me for one. That's not evidence of much, since I sell my own things rather than applying for jobs, so take it as one person's data point and not a career recommendation.

What I'd say is that the certificate market and the skill market are separate. IBM's course hands you one for free after five hours. Codecademy gates theirs behind a paid plan. Coursera puts theirs behind a subscription, and Coursera Plus runs $35 a month or $239 a year for access to their whole catalogue. If your employer reimburses training, the annual is obviously the move and this entire article is moot for you. If you're paying out of pocket to learn a skill for yourself, I'd think hard about what that $239 buys against what else it buys.

Which brings me to the arithmetic I keep coming back to.

Course Money Versus Generation Money

The thing that actually taught me image prompting was volume. Not a curriculum. Volume, and the ability to be wrong forty times in an evening without flinching at the bill.

So here's a comparison nobody in this SERP makes, because none of them sell generation and neither do the course platforms.

Those generation figures come off the published price sheets. Google lists Gemini 3.1 Flash Image at $60 per million output tokens, with a 1K image costing 1,120 tokens, which they work out to $0.067 an image, dropping to $0.034 on their batch tier. OpenAI lists gpt-image-2 at $0.211 for a high-quality 1024 by 1024, $0.053 at medium, and $0.006 at low. I divided, that's all.

Three thousand five hundred images is not a small number. It's more than enough to run exercise seven on every visual term you've ever wondered about, twice, and still have most of the budget left. My own setup generates on the machine in front of me, an M4 Pro, which puts the per-attempt cost somewhere around the price of the electricity, and that's a different position from most people's so I'd rather state it than let it quietly inflate my advice.

The counterargument, and it's fair, is that unstructured rerolling without a framework is how people acquire superstitions. You need both. I'd just note that the framework is twelve ideas long and free in six places, and the volume is the expensive part.

Questions People Ask About This

Is prompt engineering a real job? Less than it was in 2023, as a standalone title. As a component of other jobs it's everywhere, and the version that's durable is specification rather than phrasing tricks. Nobody's hiring for "knows to say take a deep breath" anymore, which is correct, because that never worked.

Which free course would you start with? The DeepLearning.AI one, purely because it's an hour and forty minutes and you'll finish it. Finishing a short course beats abandoning a good long one, and the completion rates on free online courses are famously grim. Then IBM's if you want the certificate for a profile.

Do I need to know Python? For chat window work, no. The DeepLearning.AI course is built around code examples and you can follow it without writing any, though you'll get less from it. If you're building something that calls an API, obviously yes, and at that point the course you want is about evaluation, not prompting.

Are the paid bootcamps worth it? I genuinely don't know, and I'd rather say so than guess. The 22.5-hour one is the only thing I found that covers generation properly, so there's a real argument for it if that's your gap. Udemy discounts constantly, so paying list price for anything there seems unwise.

What about certifications from the model vendors? Google, AWS, and IBM all have their own tracks and they're all fine and they all teach the same twelve ideas with their own product names attached. The vendor documentation is more useful than the vendor course, in my experience. Anthropic ships an interactive prompting tutorial as a GitHub repo and also as a Google Sheet, which is a slightly odd pairing and both work.

Is there anything a course does that self-teaching doesn't? Sequencing and finishing. Which is not nothing, if you're the kind of person who needs a syllabus to keep going. I'm not, but I've watched enough people bounce off self-directed learning to take it seriously.

Where I'd Start

Read MIT Sloan's page and the Gemini image templates. That's maybe forty minutes and it covers the concept layer for both halves.

Then run exercise one against something with your own stakes attached, not a demo request lifted off a course page, because the entire difficulty of this skill is noticing what you failed to say about your own situation and there's no way to practise that on somebody else's example.

If you want the underlying framework in one place, it's in chatgpt prompts, the deeper technique layer is in chatgpt prompt engineering, and the tooling side is covered in prompt engineering tools. None of those cost $239 either.