Something odd is happening to certificates. Since last year, AI agents have been able to work through entire courses on their own: clicking through the modules, answering the quiz questions, and collecting a certificate at the end, all without a human in the loop. For anyone who builds and sells training, that's not a side note. It cuts straight into the thing you're actually selling: proof that someone learned something.
AI can finish a course without you ever knowing
Picture this: a learner signs up for your online course, hands the keyboard to an AI tool, and shows back up three days later with a hundred percent completion and a shiny certificate. Nobody lied, exactly. There just wasn't a person on the other end doing any learning. This is already happening, especially with courses built around watch-a-video-then-take-a-quiz.
The uncomfortable part is that this scenario only makes an old problem more visible. Completion was never that strong a signal to begin with, AI just removed the last bit of doubt.
Completion was already a weak signal
There's a frequently cited example from compliance training: employees regularly scored above ninety percent on their tests, yet in practice only about a third of them actually followed the protocols they'd just been tested on. People are remarkably good at learning how to pass a test. Whether that translates into doing things differently afterward is a separate question entirely.
The same logic applies to your own courses. A progress bar sitting at a hundred percent tells you about attendance, not ability. For years that gap got quietly ignored because there wasn't an easy alternative. That excuse is gone now.
The rise of skill mastery and micro-credentials
The wider e-learning world is shifting from "completion" to "skill mastery" for exactly this reason: not whether someone clicked through the modules, but whether they can actually do the thing. Micro-credentials, small certificates tied to one specific, demonstrable skill, are growing fast. They tend to be narrow rather than broad, evidence-backed rather than attendance-based, and verifiable rather than self-reported. Established bodies like IEEE have started rolling out their own micro-credential programs to recognize specific technical skills without requiring a full degree behind them.
For an independent trainer or a small training business, that's actually good news. You don't need a university-sized budget to look credible here, you just need an assessment that proves something real.
Building assessments that are ai-resistant and meaningful
A few changes you could make this week:
Swap out multiple-choice questions, where you can, for a task that produces something real: a document, a recording, a plan worked out for the learner's own situation. That's far harder to outsource to a tool without it showing, and it gives you a much richer picture of what someone can actually do.
Build in some form of human review, even a short one. A quick peer check or a few lines of feedback from you changes the equation completely. The moment someone might ask "walk me through why you made that choice," cutting corners stops being worth it.
Ask learners to apply ideas to their own specific situation rather than answering a generic question. "Write an onboarding plan for your team" produces a completely different kind of work than "list the four steps of onboarding from module three."
Add one small live or human touchpoint, a group call, a short video submission, a brief verbal walkthrough. These are hard to fake, and learners tend to appreciate them anyway, especially since so many courses these days feel pretty solitary.
Consider a short follow-up check two to four weeks after the course ends. Just ask: what have you actually used since? That single question often tells you more about your training's real impact than any completion rate ever will.
What this means for your platform and workflow
This does ask something of your tools. Check whether your learning environment supports file-based assignment submissions, not just auto-graded quizzes. See if you can issue separate, skill-specific badges instead of one big diploma at the very end. And look closely at what your reports actually show you: if the only metric your platform surfaces is a completion percentage, you're missing the most important part of the story.
None of this means rebuilding your entire course from scratch. It does mean retiring the old assumption that clicking through plus passing a quiz automatically equals learning.
Start small, with one course
You don't have to overhaul everything at once. Pick one course, ideally your most popular or most valuable one, and swap the final quiz for a single practical assignment with a short human check attached. See what it does to the quality of submissions, and to how learners talk about the course afterward. Chances are you'll find people actually prefer it. Let's be honest, nobody ever got excited about a multiple-choice quiz.