A practical framework for universities, business schools, NGOs and youth employment organisations using Virtway
A job interview can be intimidating for any candidate. For a young person with little formal experience, low confidence or a history of rejection, it can feel like an obstacle that is impossible to overcome.
Universities, business schools, NGOs, foundations and public employment services can help by creating safe environments in which young people practise interviews before facing a real employer.
Immersive technology and conversational AI make it possible to provide this practice at scale. However, building an effective interview simulator involves much more than placing an AI interviewer inside a virtual room.
The objective should not be to create a machine that decides whether someone is employable. It should be to create a learning environment that helps young people recognise their abilities, communicate them clearly and prepare for real employment opportunities.
This guide explains how to design that environment using Virtway’s immersive 3D technology and AI-powered roleplay capabilities.
1. Start with the right objective
The first decision is not technological. It is educational.
A job interview simulator should help participants learn how to:
- understand what an employer is looking for;
- identify their own relevant skills;
- provide specific examples;
- explain their motivation;
- respond to difficult questions;
- ask informed questions about a role;
- manage an unfamiliar professional situation;
- identify the next step in their employment journey.
The simulator should not attempt to determine whether a person deserves a job.
This distinction affects every subsequent design decision. A selection system tries to rank or filter people. A learning system helps them improve.
For this reason, avoid presenting participants with an overall “employability score”. A score such as 63% can appear scientific while hiding subjective assumptions about communication, personality, confidence or cultural behaviour.
Instead, show the participant what they did, why it was effective and what they can practise next.
The central message should be:
You have not yet found the best way to explain what you can do.
It should never be:
The AI has determined that you are not employable.
2. Understand the people you want to support
Young people who are not currently in employment, education or training are often discussed as a single group. In reality, the category includes people with very different experiences, motivations and barriers.
Some may be looking for their first job. Others may have left education early, lost a previous job, recently arrived in a new country, experienced health difficulties or assumed family responsibilities.
The OECD estimates that approximately 12.8% of people aged 15 to 29 across its member countries are not in employment, education or training. It also distinguishes between young people actively looking for work and those who are not currently participating in the labour market.
For this reason, “NEET” should be treated as a statistical category, not as a personality type or identity.
Before designing the scenarios, speak directly with the young people who will use them. Include youth workers, employment counsellors, recruiters, accessibility specialists and potential employers in the research.
A short onboarding assessment can help personalise the experience. It might explore:
- previous employment or volunteering experience;
- preferred language;
- literacy and digital confidence;
- sectors of interest;
- comfort with interviews;
- accessibility requirements;
- preferred interaction format;
- perceived barriers to employment.
This assessment should not feel like another test. Its purpose is to choose the right starting point.
3. Define observable learning outcomes
Concepts such as confidence, motivation and professionalism are difficult to measure fairly. They are also easily influenced by accent, disability, personality and cultural expectations.
Use observable communication behaviours instead.
For example, after completing the programme, the participant should be able to:
- answer the question that was asked;
- describe a relevant situation;
- explain their own actions;
- state the result or what they learned;
- connect the example to the available position;
- ask at least one relevant question;
- explain a period outside employment without disclosing private information;
- complete an interview with reduced external support.
These outcomes are teachable, measurable and understandable to participants.
They can also be assessed by a human coach, making it possible to compare the AI’s feedback with professional judgement.
4. Build scenarios around real jobs
A common mistake is to build one generic interview containing the same familiar questions:
- Tell me about yourself.
- What are your strengths?
- What is your greatest weakness?
- Where do you see yourself in five years?
These questions can be included, but they should not define the entire learning experience.
A useful simulator starts with a real position.
The participant could upload a job advertisement, select an occupation from a library or follow a pathway created by the organisation. The system would then identify the tasks, working conditions and competencies associated with the role.
Initial pathways could focus on accessible or entry-level employment areas, such as:
- retail;
- hospitality;
- logistics;
- customer service;
- administration;
- maintenance;
- care services;
- manufacturing;
- tourism;
- entry-level digital support.
For every scenario, create a simple competency map.
Example
Role: Retail assistant
Typical task: Supporting a dissatisfied customer
Competency: Communication and problem-solving
Interview question: “Tell me about a time you helped someone resolve a problem.”
Possible evidence: Employment, volunteering, sport, education, a family business, caring responsibilities or a personal project.
This last point is particularly important. Young people without formal employment experience often believe that they have no relevant examples.
The simulator should help them recognise transferable skills developed outside paid work.
Someone who regularly cares for a family member may have developed organisation, responsibility and communication skills. Someone who participates in a sports team may have examples of collaboration and perseverance. Someone who sells second-hand products online may have experience in customer communication, negotiation and basic digital commerce.
The simulator should help the participant translate these experiences into credible evidence without exaggerating or inventing information.
5. Teach before assessing
A simulator should not begin by placing an unprepared participant in a high-pressure interview.
Create a progressive learning journey.
Stage 1: Orientation
Explain what will happen, what the participant will practise and how feedback will be provided.
Show that mistakes are expected and that the simulation does not affect any real recruitment process.
Stage 2: Demonstration
Present an example answer and explain why it works.
It can also be useful to compare a vague answer with a more specific one.
Vague answer:
I am good with people and I always try to help.
More specific answer:
During a volunteering event, a visitor could not find the correct registration area. I listened to what they needed, checked the event map and accompanied them to the right desk. They arrived before their session started.
The second response gives the interviewer something observable.
Stage 3: Guided practice
Allow the AI coach to help the learner construct an answer through prompts such as:
- What was happening?
- What responsibility did you have?
- What did you do personally?
- What happened as a result?
- What did you learn?
Stage 4: Independent simulation
The participant completes an uninterrupted interview with an AI interviewer.
The interviewer can ask relevant follow-up questions rather than simply moving through a fixed script.
Stage 5: Reflection
Before showing the AI’s analysis, ask the participant what they think went well and what they would change.
This develops self-assessment instead of creating dependence on the system.
Stage 6: Focused feedback
Provide one or two priority improvements, not a long list of errors.
Stage 7: Immediate repetition
Ask the participant to answer the same question again.
This is essential. Feedback becomes useful when the learner can apply it immediately.
Research into virtual job interview training has found promising improvements in interview skills, self-confidence and employment-related outcomes among several groups facing barriers to employment. However, much of this research has focused on specific populations, including people with disabilities or mental health conditions. Organisations should therefore validate the approach with their own target participants instead of assuming that one model will work equally well for everyone.
6. Use immersion for a clear purpose
The value of Virtway is not simply that the interview takes place in 3D.
Immersion allows organisations to recreate the social and spatial elements of an interview that a conventional chatbot cannot reproduce.
Participants can practise:
- entering a professional building;
- finding the correct room;
- waiting to be invited inside;
- introducing themselves;
- sitting opposite an interviewer;
- speaking to an interview panel;
- managing pauses and turn-taking;
- participating in a group exercise;
- attending an online interview;
- navigating a careers fair;
- asking a receptionist for information;
- speaking informally before or after the formal interview.
These interactions may appear minor, but unfamiliarity with the environment can increase stress.
Virtway combines interactive 3D environments, avatars and AI-powered roleplay. It also supports access through desktop and mobile devices, which is important when a programme cannot assume that every participant owns a virtual reality headset.
The immersive layer should be introduced progressively.
A participant might begin with a text conversation, move to audio, then enter a virtual interview room. A more advanced participant might face a panel, an unexpected follow-up question or a group selection activity.
The purpose of immersion is not to impress the learner. It is to reproduce relevant social conditions in a controlled and repeatable way.
7. Create different AI interviewer personas
Real interviewers do not all behave in the same way. The simulator should therefore support different interviewer profiles.
Possible personas include:
- a supportive recruiter;
- a busy store manager;
- a formal human resources representative;
- a friendly small-business owner;
- a technical supervisor;
- a two-person interview panel;
- an interviewer who asks brief follow-up questions;
- an interviewer who needs clarification before accepting an answer.
Each persona should have a defined purpose, tone and difficulty level.
Virtway’s AI roleplay technology allows scenarios and AI personalities to be adapted to different learning objectives.
A useful difficulty structure could contain three levels.
Supportive practice
The interviewer speaks clearly, allows additional time and provides prompts when the participant becomes stuck.
Standard interview
The interviewer behaves realistically, asks follow-up questions and expects the candidate to structure their own responses.
Challenging interview
The interviewer introduces ambiguity, silence, interruptions or unexpected questions.
“Challenging” should not mean aggressive or humiliating. The objective is to develop readiness, not to reproduce harmful recruitment practices.
8. Design a transparent feedback model
AI feedback should be based on a small and understandable rubric.
A practical model could assess six areas:
Relevance
Did the participant answer the question?
Evidence
Did they provide a real example or specific information?
Structure
Was it possible to understand the situation, action and result?
Clarity
Was the answer understandable and reasonably concise?
Connection to the role
Did the participant explain why the example matters for this particular job?
Authenticity
Does the answer reflect information supplied by the participant, and can they expand on it when questioned?
Feedback should always refer to something specific.
Instead of:
Your answer lacked professionalism.
Use:
You explained that you helped in a family business, but you did not describe your responsibilities. Add one task you performed regularly and one example of a problem you solved.
The system can then ask:
Did you serve customers, organise products, manage payments or solve delivery problems?
Avoid generating a polished fictional answer for the learner to memorise.
The AI may help reorganise information that the participant has already provided, but it should never invent employment history, qualifications or achievements.
The objective is to make the participant’s real experience visible.
9. Make psychological safety part of the product
For many participants, fear of judgement will be a greater barrier than lack of knowledge.
Design the experience so that mistakes feel temporary and useful.
Avoid:
- public rankings;
- red warning screens;
- pass or fail labels;
- comments about intelligence or personality;
- comparisons between participants;
- humiliating interviewer behaviour;
- excessive correction after every sentence;
- compulsory camera use;
- language such as “you do not appear motivated”.
Allow the participant to:
- pause the simulation;
- repeat a question;
- change from voice to text;
- reduce the difficulty;
- practise privately;
- remove a recording;
- restart without a penalty;
- ask why feedback was given.
Feedback should recognise progress as well as deficiencies.
For example:
This time you explained your own action clearly. The next step is to add the result.
This communicates that improvement is possible and specific.
10. Avoid evaluating cultural conformity
Interview technology can easily reward the behaviour of people who already resemble the designers’ idea of a successful candidate.
Do not automatically penalise:
- regional or foreign accents;
- limited eye contact;
- a monotone voice;
- pauses before answering;
- repetitive movements;
- shyness;
- non-standard grammar;
- the use of assistive communication;
- a communication style associated with neurodivergence.
These characteristics do not demonstrate whether someone can perform a job.
The simulator may provide optional mechanical observations, such as:
- the answer lasted nearly four minutes;
- the microphone volume was low;
- the system detected repeated use of one filler word;
- parts of the response could not be transcribed.
Present these as observations, not conclusions about personality or emotional state.
The system should not claim to detect honesty, enthusiasm, nervousness, confidence or motivation from a face or voice.
This is both an ethical and regulatory concern. The EU AI Act prohibits certain forms of emotion inference in workplaces and educational institutions, except in limited medical or safety contexts. It also treats some AI systems used for recruitment and employment decisions as high-risk.
11. Design for accessibility and real-world constraints
An inclusive simulator should work through:
- voice;
- text;
- voice and text together;
- subtitles;
- simplified instructions;
- extended response times;
- keyboard-only navigation;
- desktop and mobile devices;
- different connection qualities.
Speech recognition is another important consideration. Before assessing a response, allow the participant to review and correct the transcript. Otherwise, the system may assess the transcription engine’s mistake rather than the learner’s communication.
Language pathways should also distinguish between two objectives:
- communicating effectively in an interview;
- developing proficiency in the interview language.
A participant should not receive negative feedback about their professional potential because they speak with an accent.
12. Protect participant data
Interview practice may lead participants to reveal information about:
- health;
- disability;
- immigration status;
- financial circumstances;
- family responsibilities;
- discrimination;
- previous dismissals;
- mental health;
- criminal records;
- periods outside employment.
The simulator should not encourage unnecessary disclosure.
The AI coach can intervene with guidance such as:
You do not need to share that personal detail. Let us prepare a professional answer that protects your privacy.
A responsible implementation should include:
- recording disabled by default where possible;
- clear and specific consent;
- automatic deletion periods;
- an accessible deletion process;
- minimum necessary data collection;
- separate storage for identity and learning analytics;
- role-based access controls;
- encryption;
- clear policies on whether conversations may be used to train models;
- additional safeguards for minors;
- a data protection impact assessment where required.
The European Data Protection Board recommends mapping data flows and addressing privacy risks throughout the lifecycle of systems based on large language models.
13. Keep training separate from recruitment
A clear boundary should exist between the simulator and employer selection.
The recommended principle is:
The simulator works for the young person, not for the employer.
Employers should not receive an AI-generated employability score or use practice results to reject candidates.
A participant may choose to share a completed exercise, answer or certificate of participation. That is different from giving an employer access to private practice sessions.
If an organisation later decides to use the technology to rank, recommend or filter job candidates, it is no longer simply providing training. Its legal, ethical and technical responsibilities change considerably, particularly under the EU AI Act.
14. Combine AI practice with human support
An interview simulator should be part of a broader employment pathway.
The World Bank and ILO have found that youth employment programmes often combine skills training with employment services, counselling, mentoring, job-search support and placement activities. Programme design and the combination of interventions can affect employment outcomes.
At the end of a Virtway session, the participant should receive a practical next step, such as:
- repeat one interview question;
- revise one section of their CV;
- complete a short sector-specific course;
- apply for an appropriate vacancy;
- meet an employment counsellor;
- attend a group interview workshop;
- practise with a human interviewer;
- prepare documentation;
- research a company;
- attend a virtual careers fair.
With the participant’s consent, an adviser could see:
- completed learning activities;
- competencies practised;
- progress over time;
- challenges declared by the participant;
- answers deliberately shared;
- agreed next actions.
The adviser should not receive inferred emotions, hidden personality scores or unnecessary private recordings.
15. Measure learning, not just engagement
Completion rates and satisfaction surveys are useful, but they do not show whether the simulator helps participants obtain better employment outcomes.
Evaluation should take place at four levels.
Learning
Compare interview performance before and after the programme.
Where possible, use independent human reviewers who do not know whether the interview was recorded before or after training.
Confidence and readiness
Measure changes in interview self-efficacy, willingness to apply for jobs and anxiety about recruitment situations.
Behaviour
Track relevant actions such as:
- applications submitted;
- real interviews attended;
- adviser meetings completed;
- learning sessions repeated;
- vacancies explored.
Employment and education outcomes
Where consent and programme duration allow, study access to:
- further education;
- vocational training;
- internships;
- work placements;
- job interviews;
- employment.
Do not assume that obtaining a job was caused entirely by the simulator. Employment outcomes are also influenced by training, counselling, local labour demand, personal circumstances and employer behaviour.
Equity should be measured separately. Analyse whether performance or completion differs according to:
- language;
- accent;
- gender;
- disability;
- device type;
- education level;
- age;
- employment history;
- interaction mode.
16. Build a focused first version
A minimum viable programme does not need hundreds of occupations or highly complex virtual environments.
A strong initial version could include:
- three employment sectors;
- text and voice interaction;
- optional camera use;
- three difficulty levels;
- six to eight questions per interview;
- adaptive follow-up questions;
- a five or six-dimension feedback rubric;
- no more than two recommendations after each answer;
- immediate repetition;
- a simple progress report;
- an adviser dashboard;
- configurable data deletion;
- human evaluation of a sample of sessions.
Do not include facial emotion recognition, personality scoring, public rankings, employer-facing candidate scores or automated certification in the first version.
These functions add risk without being necessary to test the educational value of the concept.
17. An example Virtway learner journey
A participant enters a Virtway employment centre through a phone or computer.
They create an avatar and select a retail interview pathway.
An AI guide explains that the session is private and that mistakes will not affect any real application.
The participant visits a preparation area where they learn how to provide a specific example. They construct one answer with support from the AI coach.
They then enter a virtual shop manager’s office.
The AI interviewer asks:
Tell me about a time you had to help someone who was frustrated.
The participant gives an answer based on helping a customer in a family business.
The interviewer asks a relevant follow-up question:
What did you personally do to calm the situation?
After the interview, the participant reviews the response.
The AI coach identifies one strength:
You explained your action clearly.
It then identifies one improvement:
Add what happened after you solved the problem.
The participant repeats the answer and includes the result.
Finally, the platform recommends a next action:
Practise explaining why you want to work in retail, then review two current retail vacancies with your employment adviser.
The participant leaves with evidence of progress and a concrete next step, not simply a score.
Technology should create opportunities to practise, not another barrier
An immersive AI interview simulator can give young people something that is often difficult to provide at scale: repeated, realistic and private practice.
Virtway can bring together AI-powered roleplay, interactive avatars, realistic spaces, individual learning and collaborative support within the same environment.
The technology alone, however, does not determine the programme’s value.
Success depends on:
- understanding the participants;
- defining teachable outcomes;
- using real employment scenarios;
- providing focused feedback;
- protecting dignity and privacy;
- designing for accessibility;
- involving human professionals;
- connecting practice with genuine opportunities;
- measuring outcomes fairly.
The best simulator will not teach young people to perform a fictional version of the perfect candidate.
It will help them recognise what they already know, identify what they still need to learn and communicate their potential more effectively when a real opportunity appears.
