I've taken online courses since Coursera launched in 2012. I've completed about 20 and dropped another 15. Here is what I've learned about when they work and when they don't.
Technical skills with clear right/wrong feedback loops: programming, data analysis, accounting, math-heavy subjects. These translate exceptionally well to online formats because the student gets immediate feedback from working exercises and the knowledge is testable. Video content for visual or demonstrational subjects — cooking techniques, physical procedures, design software. And self-contained conceptual knowledge where the goal is understanding rather than practicing a skill (history, philosophy, science literacy).
Anything that requires real-time feedback on your performance — writing, public speaking, clinical skills, most interpersonal skills. The absence of synchronous critique from a skilled instructor limits how much you can actually improve in these areas online. Motivation-dependent learning: completion rates for MOOCs hover around 5–15% because self-directed asynchronous learning requires discipline that most people realistically don't maintain without external structure. And cohort-based learning where the peer interaction is the product — you can't replicate that asynchronously.
The single biggest predictor of completion and retention in my experience: applying the material immediately to a real project rather than just consuming the content. I've learned far more from "build this thing as I learn the concepts" than from "watch all the lectures then try to do something." Spaced repetition for memorization-heavy content (vocabulary, formulas, historical facts). Teaching what you're learning to someone else, even in informal conversation, dramatically increases retention.
Coursera and edX have the strongest university-affiliated certificates, which carry more employer weight than platform-specific certificates. Udemy has the best value for practical skill courses taught by practitioners rather than academics. LinkedIn Learning integrates with professional profiles and is adequate for quick skill top-ups. Brilliant.org is genuinely excellent for math, science, and logic — the interactive problem-solving format is significantly better than passive video for these subjects.
Here's where I land: Online learning works for specific types of knowledge. Build something with what you're learning or expect to forget 80% of it.
Re-reading highlighted notes — the most common study technique — is one of the least effective methods by research standards. It produces familiarity without producing durable memory. The discomfort of self-testing is precisely the signal that genuine learning is occurring, which is why students consistently underuse retrieval practice even when they know it works better. Feeling productive and being productive are different things in learning contexts.

Rachel Foster is an education researcher, former high school teacher, and learning science writer who covers how people learn, what education systems do well and poorly, and the evidence behind effective teaching and stu...