Preply Study Finds AI Is Expanding Learning—But Not Real-World Mastery

Preply says artificial intelligence has made professional learning faster and easier to access, but its first annual Progress Report suggests that access is not consistently translating into usable skills.

The company surveyed 5,050 working professionals across nine countries and found that 88% use AI as part of their learning. Seventy-six percent reported spending more time learning than they did two years ago, while the average respondent used nearly three learning methods.

Yet greater activity did not necessarily produce greater confidence. Nearly half of respondents—49%—said they had maintained a learning routine for months but still struggled to apply the skill in real life. Another 52% said they were spending more time learning without always seeing better results, and 63% described themselves as busy learning without truly improving.

Preply calls the disconnect the “Progress Gap”: the distance between progress experienced during learning and confidence using the skill in a real situation.

AI is becoming the starting point

The survey shows how quickly AI has become part of the learning stack. Thirty-seven percent named AI tools as a primary learning method, ahead of employer training at 34% and online courses at 33%.

Respondents most often used AI to obtain explanations or definitions, summarize material, translate content, receive corrections, and structure their learning. Daily AI users were also nearly twice as likely to say they were learning faster than expected as people who never used AI—60% compared with 31%.

Those results help explain AI’s appeal. It can answer a question immediately, simplify difficult material, and make practice available between formal lessons. These capabilities are also driving the broader shift toward AI agents that personalize education while assisting teachers.

But speed and availability address only part of the learning process. In Preply’s survey, 66% agreed that access to information does not automatically lead to progress. More than half—55%—said it is harder than ever to identify what genuinely helps them improve. That echoes a broader concern that frictionless access to answers is not the same as deeper learning.

The distinction is important for both individual learners and employers. AI can make it easier to acquire information, but workplace performance depends on recall, judgment, practice, feedback, and the ability to act under real conditions. The same principle underpins the growing focus on AI fluency as a practical workforce capability, rather than a tally of training completions or tool usage.

Learners want AI to support—not lead

The report does not show a rejection of AI. Instead, it points to a preference for combining automation with human expertise—an approach Preply has also described in its own account of pairing human tutors with AI-powered learning tools.

Ninety percent of respondents wanted AI to play a supporting role rather than lead their learning. Seventy-five percent trusted a blend of AI and human expertise for long-term progress, compared with 14% who preferred a traditional approach without AI and 11% who preferred a fully self-directed AI experience.

Human feedback or accountability was the most frequently cited driver of learning consistency, selected by 46% of respondents. Clear goals followed at 44%, with the ability to use learning in real situations at 42%. AI reminders or tools ranked considerably lower at 18%.

“People value AI because it helps them learn more easily,” Preply CEO Kirill Bigai said in the report. “They value people because they make learning stick, and translate knowledge into real world ability.”

That division of labor suggests a practical model for AI-enabled learning. AI can reduce friction at the beginning of the process by explaining, summarizing, translating, and generating exercises. Human instructors, managers, and peers can create accountability, notice misconceptions, adapt practice to context, and evaluate whether a learner can perform independently.

Gen Z reports the widest gap

The strongest adoption figures did not always correspond with the strongest outcomes.

Gen Z recorded the highest AI adoption of any generation at 92%, but also reported the largest difficulty converting learning into action. Seventy-one percent of Gen Z respondents said they felt busy learning without improving, compared with 56% of Gen X. Fifty-seven percent said they could not apply what they had learned in real life, compared with 40% of Gen X, while 68% had abandoned learning during the previous year, versus 38% of Gen X.

The report does not establish that AI caused these differences. Preply explicitly notes that its survey cannot explain why the relationship exists. The findings instead show that the group most engaged with modern learning tools is also the group most aware of the gap between acquiring knowledge and performing confidently.

Adoption also varied by seniority and country. Forty-seven percent of senior leaders used AI for learning daily, compared with 35% of junior or middle managers and 31% of non-managerial employees. Argentina and Poland reported the highest overall adoption at 93%, while the United States and Japan were lowest among the nine markets at 85%—still a large majority.

What this means for workplace learning

For employers, the report argues against treating time spent, course completions, or learning streaks as sufficient measures of progress. If the goal is better performance, learning programs need to test whether employees can use a skill in realistic situations.

That could mean pairing AI-generated explanations with simulations, supervised practice, manager feedback, or live coaching. It also means evaluating outcomes closer to the work itself: whether someone can solve a problem, communicate clearly, make a decision, or complete a task with less assistance.

The study is based on an online survey fielded by Dynata and completed on June 26, 2026. Respondents were employed or self-employed adults ages 18 to 55 who had actively tried to learn a career-related skill during the previous 12 months. The sample covered the United States, United Kingdom, Spain, France, Germany, Italy, Poland, Japan, and Argentina. Because the findings are self-reported, they describe respondents’ perceptions and experiences rather than independently measured skill gains.

Even with that limitation, the central result is difficult to ignore: AI has reduced the cost of reaching information, but learning still has to survive contact with the real world. The next stage of AI-enabled education will be judged less by how much content it can deliver and more by whether learners can confidently use what it teaches.