31 July 2026

 

Universities are teaching artificial intelligence, but employers say many graduates still lack the practical AI skills needed at work.

New research from Pearson and Amazon Web Services (AWS) found that 53% of surveyed employers struggle to find graduates with the right skills for AI-enabled roles. Yet 78% of higher education leaders believe their graduates meet employers’ expectations.

The findings reveal a graduate AI skills gap centred not on access to technology, but on its practical application. Students may understand AI and use generative AI tools for academic work without being able to evaluate outputs, manage risks or incorporate the technology into professional workflows.

The study, entitled AI Readiness: Building the Bridge from Higher Education to Work, is based on a survey of 2,711 respondents in the US, UK, Brazil, Saudi Arabia, Vietnam and Malaysia. The sample comprised 1,955 learners, 452 higher education leaders and 304 employers, supplemented by interviews with university leaders.

Its findings suggest universities are not necessarily failing to expose students to AI. The harder problem is turning that exposure into demonstrable workplace capability.

 

What does it mean to be AI-ready?

 

Pearson and AWS define an AI-ready graduate through four interconnected capabilities:

• Functional AI proficiency
• Strategic intelligence
• Ethical stewardship
• Critical human skills

Functional proficiency includes using AI tools effectively, instructing or prompting them and understanding how the technology works. But AI readiness extends beyond operating a chatbot or generating content. It includes evaluating outputs, applying human judgement, adapting to changing situations and collaborating with other people while working alongside AI.

This exposes a weakness in some early institutional responses to generative AI. Providing access to tools or adding an introductory AI module does not necessarily prepare students to incorporate the technology into a professional workflow.

Across the six markets, 53% of learners said they frequently use AI for core academic tasks. Yet only 34% were highly confident that their use complied with institutional rules. High adoption is therefore developing alongside substantial uncertainty about acceptable and responsible AI use.

 

UK students report low confidence in their AI readiness

 

The report highlights particularly low confidence among UK students. Only 14% rated their personal readiness for an AI-enabled workplace as high, compared with a cross-market average of 32%.

This was a self-assessment rather than a test of demonstrated ability. It should not be read as proof that 86% of UK students cannot use AI effectively.

Nor does lower confidence necessarily indicate weaker performance. It's interesting, however, that UK learners gave themselves the lowest AI readiness ratings in the study, while recent UK graduates received some of the higher capability ratings from employers.

 

Employers want practical AI skills

 

Employers’ concerns help explain that lack of confidence. Among the 304 employers surveyed, 42% cited insufficient hands-on experience with workplace AI tools as a barrier to hiring or developing graduates for AI-enabled roles. Another 41% identified the gap between academic knowledge and workplace application.

The shortage is not simply about prompting skills. Employers across the participating countries placed communication and collaboration, adaptability and human judgement among the most important competencies for working alongside AI.

These requirements reflect how businesses are adopting the technology. Most graduates will not be employed to develop foundation models or build machine-learning infrastructure. They are more likely to encounter AI inside marketing platforms, financial systems, cloud software, customer-service applications, research tools or sector-specific workflows.

In these environments, the valuable skill is not merely producing an answer. Graduates must be able to:

• Recognise when an AI-generated output is unreliable
• Decide what data can safely be entered into a system
• Understand how an output affects a business process
• Verify information against authoritative sources
• Recognise when human intervention is required

The report found that 58% of employers rated graduates’ ability to critically verify AI outputs as their weakest competency. This is harder to teach and assess than basic tool use because it depends on context, subject knowledge and practical experience.

 

Why universities and employers disagree about AI skills

 

The gap between institutional confidence and employer dissatisfaction may partly reflect the absence of a shared definition of AI readiness.

Universities tend to evaluate progress against curricula, learning objectives and academic assessments. Employers judge readiness against graduates’ ability to contribute inside an organisation. A student can therefore meet the requirements of an AI-related course without having applied the technology in a realistic workplace setting.

The speed of change makes alignment harder. Across all respondents, 67% described AI-driven workplace change as very or extremely fast. Seventy per cent of higher education leaders expect the pace to accelerate over the next two years.

Yet only 24% of all respondents believed universities were keeping pace with most or all AI-related developments. Among employers, the figure was 28%.

Higher education cannot redesign degree programmes whenever a new model or enterprise tool appears. Training students on one product would not necessarily provide durable preparation either. AI tools change, while the ability to question outputs, learn new systems and exercise judgement is more transferable.

 

Faculty AI training varies sharply by country

 

The research suggests that institutional capacity remains uneven. Only 13% of higher education leaders across the six markets rated their faculty’s AI knowledge and skills as “very strong”.

The national differences were substantial. Fifty-eight per cent of Saudi higher education leaders selected this rating, compared with 6% in the UK and 3% in the US.

Training provision displayed a similar divide. Eighty-six per cent of higher education leaders in Saudi Arabia and 80% in Vietnam reported comprehensive, continuing professional development. In both the UK and US, only 5% described faculty access to AI training as extensive.

Industry partnerships also ranked last among higher education’s AI investment priorities across the survey. Without confident teaching staff and regular input from employers, universities risk teaching AI as a subject without adequately reflecting how it is changing work.

 

Who is responsible for closing the graduate AI skills gap?

 

Universities cannot provide every student with role-specific knowledge of employers’ proprietary systems. Companies are closest to the tools, data controls and workflows graduates will encounter, giving them responsibility for part of the transition.

A more realistic division of responsibility would see universities develop foundational AI proficiency, critical thinking and responsible-use practices. Employers would contribute current use cases, access to workplace tools and role-specific training.

This could involve co-designed assignments, industry projects and assessments that require students to use AI under realistic constraints. Instead of asking only whether a student can generate an output, an assessment could examine whether they can verify it, document their process, protect sensitive information and explain when the technology should not be used.

Pearson and AWS organise their recommendations around an “AI Readiness Friction Framework”. It identifies six areas where progress can stall:

• Pace: technology changes faster than curricula and institutional processes
• Connection: universities and employers lack effective feedback loops
• Capability: faculty AI knowledge and confidence are uneven
• Governance: rules do not always translate into responsible practice
• Experience: students lack structured opportunities to apply AI
• Skills: taught capabilities do not fully match workplace requirements

The framework is a diagnostic and action framework rather than a validated measurement of employment outcomes. The report does not establish whether institutions adopting it produce graduates who find work more quickly or perform better once hired. Nor did the survey directly measure whether AI is reducing the number of entry-level positions available.

 

What the research does—and does not—show

 

The findings have methodological limits. Survey responses were self-reported, and the samples were not reweighted to account for differences among the six national education systems.

With 304 employers participating, country-level employer findings are also based on relatively small subsamples. The results are therefore best understood as a comparison of stakeholder perceptions rather than an objective audit of graduate AI skills worldwide.

Even with those qualifications, the central message is difficult to dismiss. The graduate AI skills gap is not principally a question of whether students have encountered artificial intelligence. Many already use it routinely.

The unresolved question is whether universities and employers can jointly turn widespread experimentation into reliable professional judgement. If they cannot, students may graduate surrounded by AI yet remain unprepared for the work it is changing.