How to Scale AI Public Speaking Training in UK Universities

On a Tuesday evening in November, a first-year engineering student stands in her kitchen rehearsing a two-minute pitch to a phone propped against a mug. Tomorrow she'll deliver it live, in a seminar, to twelve classmates and a tutor who has forty other students to see that week. She has had exactly one chance to practise out loud before it counts towards her grade.

This is not an unusual scenario. Oral presentations are built into almost every UK degree programme, from engineering pitches to nursing handovers to business case studies, yet the practice time behind them is often a single seminar slot, a peer group, and whatever nerve a student can find on the day.

For programme teams trying to build genuine, repeatable speaking practice into a curriculum of hundreds or thousands of students, the maths rarely works. There are not enough staff hours, seminar rooms, or marking slots to give every student what that engineering student needed: private, low-stakes, repeated practice with useful feedback attached.

This is where AI-powered public speaking simulation is starting to change the equation, not by replacing tutors, but by giving every student somewhere to rehearse before the moment that counts.

Why public speaking is a bigger problem than it looks

Ask most curriculum teams which skill causes students the most distress and the answer rarely shows up in a strategy document, because presentations are treated as a normal part of university life. The evidence says otherwise.

A survey of students at two UK universities found that 80% reported anxiety linked to oral presentations, compared with 45–52% for seminar discussions and 14–26% for lectures, making presentations the single most anxiety-inducing situation these students faced (Grieve et al., 2021, citing Russell & Topham, 2012). A separate, larger study of 1,745 university students confirms the pattern is widespread rather than confined to one type of student, describing public speaking anxiety as among the most common anxieties reported in higher education (Lintner & Belovecová, 2024).

The consequences reach beyond the seminar room. In a survey of UK employers, communication skills mattered to 55% of respondents, second only to enthusiasm for the role and ahead of subject knowledge itself (University Alliance / CBI Economics, 2024). Globally, the picture is similar: the World Economic Forum's Future of Jobs Report 2025 found that 63% of employers see skills gaps as the single biggest barrier to business transformation over the next five years, based on responses from over 1,000 employers representing 14 million workers (WEF, 2025).

Yet there is a persistent gap between how proficient graduates think they are and how proficient employers find them to be. NACE's 2024 Student Survey and Job Outlook 2025 found that 78.1% of US students rated themselves proficient in communication, while only 53.5% of employers agreed, a 24.6-point gap.

On professionalism the gap widens further: 84.6% of students rated themselves proficient against just 50% of employers (NACE, 2025). UK data tells a similar story in motion: employer concern about graduates' verbal communication climbed from 22% in 2023 to 42% in 2025, according to the Institute of Student Employers (ISE, 2025).

The appetite for AI-assisted learning is already there, which matters for how easily any new practice format gets adopted. In UK universities, 95% of undergraduates now use AI tools in some form for their studies, up from 66% just two years earlier (HEPI / Kortext, 2026). Globally, 91% of learning and development professionals say human skills, including presentation and interpersonal skills, matter more as AI reshapes other parts of work (LinkedIn, 2024).

Universities introducing AI-supported speaking practice are not asking students to adopt an unfamiliar technology; they are pointing an existing habit at a skill gap employers are already flagging. Separate US research found that only 48% of 2025 graduates felt prepared to apply for entry-level roles at all (Cengage Group, 2025).

Put together, these figures describe two ends of the same problem: a well-documented source of student distress at one end, and a well-documented cost to graduate employability at the other, with very little connective tissue between them in how most curricula are built.

Most curricula treat presentation skill as something students either arrive with or pick up on their own. The evidence says it needs to be taught and practised like any other skill.

Why the traditional model can't scale

The standard way to teach presentation skills, a seminar slot, a rehearsal in front of peers, feedback from a tutor with limited time, was designed for cohorts of twenty, not two hundred. Three structural constraints make it hard to scale.

Staff time doesn't grow with student numbers. A tutor giving five minutes of spoken feedback per student needs over sixteen hours to get through a cohort of two hundred, on top of marking, research, and pastoral duties.

Practice happens once, if at all. Most students get a single attempt at a presentation before it is assessed. There is no equivalent of running through a job interview a dozen times before the real one.

Feedback is inconsistent. Two tutors marking the same presentation against the same criteria will weight tone, structure, and delivery differently, and students rarely get a second data point to check whether feedback reflects a real pattern or one person's impression.

There's an equity dimension too. A fixed date and a live audience are especially difficult for students with speech or language differences, social anxiety, or English as an additional language. A model that gives every student exactly one shot, in front of the same audience, at the same difficulty level, works reasonably well for confident native speakers and poorly for almost everyone else.

There's a reporting dimension too. Professional, statutory and regulatory bodies increasingly expect programme teams to evidence how employability and communication skills are developed and assessed, not simply asserted in a module handbook. A seminar-based model that produces no consistent record of practice or improvement makes that evidence difficult to compile, on top of everything else it struggles to deliver.

None of this reflects a lack of care from teaching staff. It reflects an under-resourced format being asked to do a job it was never built for: giving every student enough repetition, in a low-stakes setting, to build a skill that only improves with practice. Scaling public speaking training is less a curriculum design problem than a capacity problem, and capacity problems need a different kind of solution than a better lesson plan.

What the evidence says works

The research on AI and virtual reality practice for public speaking is more developed than most people assume, and it points in a consistent direction: repeated, low-stakes rehearsal with a responsive virtual audience measurably reduces anxiety and, in at least one study, carries through to real-world performance.

A review of recent meta-analyses found that virtual reality exposure practice reduces public speaking anxiety with a large effect size, Hedges' g between −1.39 and −1.46 compared with waitlist controls, broadly comparable to in-person exposure therapy (meta-analytic review, PMC, 2024).

In a study that went a step further, 73 participants practised a presentation in VR before delivering it for real in a university course; those who rehearsed in front of a supportive virtual audience were rated by instructors, blind to condition, as more confident in their real presentation than those who practised with an unsupportive or no audience (Kroczek & Mühlberger, 2023).

Newer research also points to AI conversational roleplay, not only VR headsets, as an effective format. A 2025 randomised controlled trial of 61 university students found significant reductions in social anxiety symptoms after self-guided virtual reality practice, with effect sizes (Cohen's d) of 0.83 to 1.10 across two anxiety measures, holding steady a month later (Kan et al., 2025).

And in a UK study, 27 medical students across three universities used AI voice conversation to practise difficult clinical conversations; their self-reported confidence rose significantly across every scenario tested, including breaking bad news and counselling anxious patients (Mukadam et al., 2025), though the study's small sample and lack of a control group are worth keeping in mind.

A separate 2024 study of 28 university students found a significant reduction in public speaking anxiety following a VR intervention aimed specifically at glossophobia, the fear of public speaking, adding further weight to the pattern across different research groups and methods (Balachandran, Vohra & Srivastava, 2024).

The common thread across this research isn't the headset. It's the format: private enough that failing doesn't cost anything, repeatable enough that students can practise more than once, and specific enough that the feedback maps onto a skill rather than a general impression. Universities don't need every student in VR to capture this.

The same underlying practice, structured rehearsal with a responsive AI conversation partner and feedback tied to specific behaviours, works whether a student logs in on a headset, a laptop in the library, or a phone on the bus, which matters considerably for a rollout that has to reach an entire cohort rather than a lab's worth of volunteers.

At Bodyswaps, this is the model we've built our own pedagogy around: structured learning pathways that combine a short guided lesson, an AI roleplay attempt with live AI coaching, and a reflection step, so a practice session produces both a rehearsal and a data point a programme team can use. It's a way of taking the research above, repetition, low-stakes practice, feedback tied to specific behaviours, and turning it into something a curriculum team can actually timetable.

How to scale it without waiting for a bigger budget

Universities that have moved past small pilots tend to follow a similar pattern.

1. Start where the anxiety is highest, not where it's easiest

Look for the modules where presentations are assessed and consequential, final-year vivas, nursing OSCEs, employability placements, rather than starting with a low-stakes elective. Early wins in high-anxiety modules build the case for wider rollout faster than pilots in courses where the problem was never that acute, and they give a programme team a concrete, gradeable outcome to point to when the next department asks whether it's worth the effort.

2. Make practice part of the module, not an optional add-on

Tools that sit outside the curriculum get used by the students who need them least. Embedding a practice session into the assessment timeline, with a completion requirement or a small number of participation marks, changes uptake dramatically compared with an optional resource buried in a reading list.

3. Give students more than one attempt before the mark counts

The research above works because it involves repetition. A single AI-supported practice run simply replicates the one-shot problem it's meant to solve. Two or three low-stakes attempts, with feedback between them, is where the anxiety reduction and skill transfer actually happen.

4. Track it at cohort level, not just individual level

Programme leads gathering completion rates, confidence ratings, and common feedback themes across a cohort can spot where a whole year group is struggling with the same thing, structuring an argument, pacing, eye contact, and adjust teaching accordingly, which a one-to-one seminar can't easily surface. That same cohort-level data is usually what turns a single champion's pilot into a department-wide requirement.

5. Plan for every device a student might use

A rollout that assumes every student has VR headset access will stall at the equity question before it starts. Practice that works equally well on a laptop or phone removes the biggest practical barrier to scaling past a pilot cohort.

6. Report the outcome employers care about, not just usage numbers

Completion rates matter to a programme team. Confidence gains and communication proficiency matter to accreditation panels and, eventually, to the employers reading a graduate's CV. Universities already running structured employability programmes for higher education institutions tend to build the connection between practice data and employability outcomes for students in from the start, which makes the case for continued investment far easier at renewal time.

None of this requires an all-or-nothing switch from seminar-based teaching to AI practice. The universities getting the most out of it are running both, using in-person teaching for structure and judgement, and AI-based practice for repetition, the one input that seminars were never built to supply at scale.

The practical takeaway

Public speaking anxiety in UK universities is well documented, and so is the employability cost of leaving it untreated. What's changed recently is the evidence base for a specific fix: repeated, low-stakes AI-supported practice, available on whatever device a student has, that gives every student more than one attempt before the mark counts. It won't replace a good tutor. It will give that tutor's students the practice reps they were never going to get in a single seminar slot.

For programme teams weighing where to start, the smallest viable pilot is usually a single high-anxiety module with a clear assessment point attached, tracked at cohort level from day one. That's usually enough evidence to make the case for the next one.

See it before you scale it

If you're weighing up where AI-supported speaking practice fits in your employability curriculum, we're happy to talk through what a pilot could look like for your cohort.

→ Try Bodyswaps

→ Book a demo with our team

Further reading

→ VR helps colleges tackle the soft skills crisis

→ What is VR soft skills training?

→ Bodyswaps library: ready-made pathways for employability teams

Frequently asked questions

What is AI public speaking simulation training?

It is a way for students to practise a speech or presentation with a computer program that talks back, listens, and gives feedback, so they can rehearse as many times as they need before speaking in front of real people.

Does virtual reality actually reduce public speaking anxiety?

Yes. Reviews of many research trials found that VR practice leads to a large drop in public speaking anxiety, similar in size to other proven treatments, and one study showed the improvement carried over into a real, graded presentation.

How can a university scale public speaking training across a whole cohort?

Build the practice into the course itself instead of making it optional, let students try more than once before they are graded, and make sure the practice works on a laptop or phone as well as a VR headset, so cost and access don't limit who can use it.

Why does public speaking anxiety matter for employability?

Employers rate communication as one of the top skills they look for in new graduates, but research shows a large gap between how good students think their communication skills are and how good employers find them to be. Practising more before graduation helps close that gap.

References

Balachandran, A., Vohra, P., & Srivastava, A. (2024). A virtual reality approach to overcome glossophobia among university students. Proceedings of the ACM on Human-Computer Interaction (ISS 2024). dl.acm.org/doi/10.1145/3698141

Cengage Group. (2025). 2025 Graduate Employability Report. www.cengagegroup.com/news/press-releases/2025/cengage-group-2025-employability-report/

Grieve, R., Woodley, J., Hunt, S. E., & McKay, A. (2021). Student fears of oral presentations and public speaking in higher education: a qualitative survey. Journal of Further and Higher Education, 45(9), 1281–1293. www.tandfonline.com/doi/full/10.1080/0309877X.2021.1948509

HEPI & Kortext. (2026). Student Generative AI Survey 2026. HEPI Report 199. www.hepi.ac.uk/reports/student-generative-ai-survey-2026/

Institute of Student Employers. (2025). Early Talent Development Trends to Watch in 2025. luminate.prospects.ac.uk/early-talent-development-trends-to-watch-in-2025

Kan, C., Wang, Y., Hu, R., Chen, K., & Zhang, Y. (2025). Smartphone-based self-help virtual reality exposure therapy for college students' social anxiety: a randomized controlled study. Virtual Reality, 29, Article 113. link.springer.com/article/10.1007/s10055-025-01195-0

Kroczek, L. O. H., & Mühlberger, A. (2023). Public speaking training in front of a supportive audience in Virtual Reality improves performance in real-life. Scientific Reports, 13, 13968. www.nature.com/articles/s41598-023-41155-9

LinkedIn. (2024). 2024 Workplace Learning Report. learning.linkedin.com/resources/workplace-learning-report-2024

Lintner, T., & Belovecová, B. (2024). Demographic predictors of public speaking anxiety among university students. Current Psychology, 43, 25215–25223. link.springer.com/article/10.1007/s12144-024-06216-w

Mukadam, A., Suresh, S., & Jacobs, C. (2025). Beyond traditional simulation: an exploratory study on the effectiveness and acceptability of ChatGPT-4o advanced voice mode for communication skills practice among medical students. Cureus, 17(5), e84381. www.ncbi.nlm.nih.gov/pmc/articles/PMC12175028/

NACE (National Association of Colleges and Employers). (2025). The gap in perceptions of new grads' competency proficiency and resources to shrink it, drawing on the 2024 Student Survey and Job Outlook 2025. www.naceweb.org/career-readiness/competencies/the-gap-in-perceptions-of-new-grads-competency-proficiency-and-resources-to-shrink-it

University Alliance & CBI Economics. (2024). To what degree: understanding what UK businesses look for in graduates. www.unialliance.ac.uk/2024/08/14/towhatdegree/

Virtual Reality exposure therapy in the treatment of public speaking anxiety and social anxiety disorder. (2024). Meta-analytic review. pmc.ncbi.nlm.nih.gov/articles/PMC12244110/

World Economic Forum. (2025). The Future of Jobs Report 2025. www.weforum.org/publications/the-future-of-jobs-report-2025/