AI Transparency Notice

1. Introduction

Svayz Ltd (“Svayz”) uses artificial intelligence to assist with recruitment processes. This AI Transparency Notice explains which AI systems we use, what they do, how they affect you, and your rights in relation to AI-driven decisions.

This notice is provided in accordance with the EU AI Act (Regulation 2024/1689), UK Government AI transparency guidance, and our commitment to ethical AI in hiring.

2. AI Systems in Use

2.1 CV Parsing and Structuring

2.2 Job Matching and Compatibility Scoring

2.3 Application Screening Assistance

2.4 AI-Assisted Interviews

2.5 Search and Recommendation

3. Application Quality Intelligence

Svayz operates an Application Quality Intelligence (“AQI”) system that assesses the quality and authenticity of job applications. This section describes what the system analyses, how it works, and your rights in relation to it.

3.1 Overview

3.2 Currently Active Signals

The AQI system currently analyses the following signals. This section is updated as new signals are activated.

3.2.1 Email Domain Reputation

What we check: Whether the email address used to apply comes from a disposable or temporary email service (e.g., Guerrilla Mail, Temp Mail).

What we do not check: We do not penalise personal email providers (Gmail, Outlook, Yahoo). Using a personal email address has no effect on your quality score.

How it affects your score: Applications from disposable email services receive a lower quality confidence indicator. This is one factor among several — a single signal does not determine your overall quality score.

3.2.2 Writing Analysis

What we check: AI analyses cover letter and screening answers for specificity density — counting specific details (company names, metrics, project references) per claim.

What we do not check: We do not assess grammar quality, vocabulary sophistication, or prose style. A non-native speaker citing specific company projects scores higher than a native speaker writing generically.

How it affects your score: Generic, non-specific text may reduce quality confidence. Text with specific, verifiable details improves it. If no cover letter is provided, this signal is skipped (not penalised).

AI involvement: Uses Google Gemini (generative AI model) for structured text analysis.

3.2.3 Content Relevance

What we check: How specifically the application references this particular role, company, and job requirements versus generic content that could apply to any job.

What we do not check: We do not require keyword-stuffing. Natural, genuine references to the role are sufficient.

How it affects your score: Generic, copy-paste applications that show no awareness of the specific role may receive lower quality confidence. Tailored applications improve it.

AI involvement: Uses Google Gemini (generative AI model) alongside the writing analysis in a single API call.

3.2.4 Behaviour Tracking

What we check: Application submission patterns such as time spent on the job page, scroll depth, form completion time, and application velocity (how many applications submitted in a short period). This signal requires analytics consent.

What we do not check: We do not track mouse movements, keystrokes, or any biometric data. We do not monitor browsing activity outside the application page.

How it affects your score: Only physically impossible patterns affect your score — for example, submitting hundreds of applications per hour, or completing a form in under 2 seconds. Normal human application patterns have no negative effect.

Data protection: This signal only activates when you have given analytics consent. If you decline analytics cookies, no behaviour data is collected and this signal is skipped (not penalised).

3.2.5 LinkedIn Profile Verification

What we check: When candidates optionally connect their LinkedIn profile, we compare resume claims against LinkedIn data. We use AI to semantically compare four data points: name, current role and company, employment history dates, and skills. The AI handles legitimate variations like title synonyms (“Senior Developer” vs “Sr. Software Engineer”) and company rebrands.

How it affects your score: Each data point is compared and scored. Matching data strengthens the application; differences are flagged for recruiter review with a neutral side-by-side display. The recruiter always interprets the results.

Optional: LinkedIn verification is entirely optional. Not connecting LinkedIn never reduces your application quality score.

Data handling: Only comparison results are retained permanently. Raw LinkedIn profile data can be deleted via your candidate dashboard.

AI involvement: Uses Google Gemini (generative AI model) for structured semantic comparison.

3.2.6 Video Introductions

Video intros are optional recordings that candidates can provide during the application process.

Your video is viewed ONLY by the hiring team. We do NOT use any form of artificial intelligence, machine learning, or automated analysis on your video. Specifically:

• No facial recognition
• No emotion detection
• No voice analysis
• No biometric analysis of any kind

Your video functions as a binary signal — the system records whether you submitted a video (yes/no) but does not score, analyse, or process the video content in any way.

This information is used to indicate genuine interest to recruiters, similar to how a cover letter shows effort.

3.2.7 Application Effort Indicators

As part of the text quality review, the system analyses several linguistic signals that may indicate the level of personal effort invested in an application. This analysis isinformational only and does not affect the Application Quality Score.

What is analysed

The system examines six categories of textual indicators:

• Sentence structure variety — whether sentence lengths and patterns are naturally varied or unnaturally uniform across the full text
• Vocabulary range — the diversity of word choice throughout the application
• Hedging patterns — the frequency of qualifier phrases such as “it could be argued” or “perhaps”
• Structural originality — departure from formulaic or template-like structures
• Personal details — presence of specific anecdotes, names, dates, or project references
• Authentic and suspicious markers — patterns that may suggest genuine effort or lack thereof

How it works

• Analysis is performed as part of the existing text quality review (same AI system, same API call) — no additional data is collected.
• Results are probabilistic indicators, not definitive classifications. The system reports that patterns “may indicate” or “suggest” a particular level of effort — it never declares certainty.
• Effort indicators are informational only and have no impact on the Application Quality Score. They are shown to recruiters as a separate, advisory data point.

Bias mitigation for non-native English speakers

The system is explicitly calibrated to avoid penalising non-native English speakers and ESL applicants:

• Limited vocabulary range alone is not treated as a suspicious indicator
• Grammatical patterns and simple sentence structure are not flagged
• Only structural patterns across the entire text (e.g., unnaturally uniform sentence lengths) are considered
• Personal anecdotes with specific details are weighted as strong indicators of genuine effort regardless of language proficiency

Important limitations

• Texts shorter than 100 words cannot be reliably analysed for effort indicators
• No effort analysis system is definitive — results should inform, not determine, hiring decisions
• Employers are encouraged to use these indicators as one data point among many

AI involvement: Uses Google Gemini (generative AI model) alongside the writing analysis in a single API call.

3.3 How Scores Are Computed

• Quality confidence is expressed as Low, Medium, or High.
• Scores are computed from available signals only — if a signal has not been analysed, it is excluded from the calculation. Missing signals never reduce your score.
• The system currently analyses 6 signals. As the system evolves, your score may be recalculated.
• Quality confidence is separate from job-fit matching. A high quality confidence means your application appears genuine, not that you are a strong fit for the role.

3.4 How to Improve Your Quality Score

The following actions may positively influence your application quality score:

• Use a professional or personal email address (not a disposable or temporary email service).
• Reference specific projects, metrics, or requirements from the job description in your cover letter.
• Tailor your application to the specific role and company rather than using a generic template.
• Use your professional email if available (though personal email such as Gmail carries no penalty).

3.5 Your Right to Contest

If you believe your quality score is inaccurate or unfair, you may contest it.

• Send an email to infra-admin@svayz.com including your Application ID (shown in your application confirmation email).
• A human reviewer will examine your application, the audit log, and the signal data within 10 business days.
• The reviewer may override the quality score if they determine it was incorrect.

This right is provided in accordance with EU AI Act Article 86 (right to explanation) and GDPR Article 22 (right not to be subject to solely automated decisions).

3.6 Audit and Record-Keeping

• Every quality score computation is logged in a dedicated audit trail.
• Audit records include: which signals were analysed, the score for each signal, the final confidence level, and a human-readable explanation.
• Audit records are retained for 5 years in accordance with EU AI Act Article 12 (record-keeping) and Article 26(6) (deployer log retention).
• If you exercise your GDPR right to erasure, personal identifiers are removed from audit records, but the anonymised decision record is retained as required by the EU AI Act.

4. EU AI Act Compliance

Svayz's AI systems used in recruitment (job matching, application screening, AI interviews) are classified as high-risk AI systems under Annex III of the EU AI Act (employment, workers management, and access to self-employment).

Our compliance measures include:

4.1 Risk Management

• Written risk management procedures for all high-risk AI systems
• Regular bias testing across protected characteristics
• Incident logging and escalation procedures

4.2 Data Governance

• Training data documentation for models we fine-tune
• Data quality controls and outlier handling
• Data minimisation — we only use data necessary for the stated AI purpose

4.3 Technical Documentation

• Architecture documentation for all AI systems
• Performance benchmarks and accuracy metrics
• Change management records

4.4 Transparency and Logging

• All AI interactions are logged with timestamps and model version
• AI decision audit trails are retained for 5 years (anonymised on GDPR erasure; anonymised records retained for EU AI Act compliance)
• Candidates are informed when AI is used in their assessment

4.5 Human Oversight

• No fully automated hiring decisions
• All AI outputs are presented as advisory to human decision-makers
• Employers are contractually required to exercise independent judgment

4.6 Accuracy and Robustness

• Regular performance monitoring against fairness metrics
• Mechanisms to flag anomalous or low-confidence AI outputs
• Fall-back processes when AI systems are unavailable

5. Bias and Fairness

We take the risk of algorithmic bias seriously in recruitment contexts.

5.1 Protected Characteristics

Our AI systems are designed and tested to minimise differential outcomes based on:

• Age
• Disability
• Gender reassignment
• Marriage and civil partnership
• Pregnancy and maternity
• Race and ethnicity
• Religion or belief
• Sex
• Sexual orientation

5.2 What We Do

• Score distribution tracking is implemented to monitor for potential demographic skew in AI-generated scores, with systematic bias monitoring planned for expansion as part of our ongoing AI governance programme
• Test datasets include representative samples across protected characteristics
• Anomalous scoring patterns trigger human review
• Employers are instructed not to use AI scores as the sole basis for exclusion

5.3 Known Limitations

AI language models may reflect biases present in their training data. While we work to mitigate this, no AI system is bias-free. We encourage employers to:

• Review AI assessments critically
• Consider the full context of each application
• Never reject candidates based solely on AI outputs

6. Your Rights

6.1 Right to Be Informed

You have the right to know when AI is being used to assess your application. We will:

• Display a clear notice on any AI-assisted application or interview process
• Provide this AI Transparency Notice on our Platform

6.2 Right to Explanation

If you receive an adverse decision (rejection) after AI was used in your assessment, you can request:

• A general explanation of how the AI tool works
• Information about which data was used as inputs

To request an explanation: Email infra-admin@svayz.com with subject line “AI Decision Explanation Request”.

6.3 Right to Human Review

Where AI has contributed to a significant decision about you, you have the right to request that a human reviews the AI's output. This is always available — contact the employer directly or email infra-admin@svayz.com.

6.4 Right to Object

You may object to AI-based processing of your data. If you do, your application will be reviewed without AI assistance (where technically feasible). This may result in longer processing times.

To object: Email infra-admin@svayz.com with subject line “AI Processing Objection”.

7. AI Providers

We use the following third-party AI providers:

We have Data Processing Agreements with all AI providers. No provider retains your personal data for training their models beyond our contracted processing purposes.

8. AI-Generated Content Disclosure (EU AI Act Article 50)

In accordance with Article 50(2) of the EU AI Act, we disclose that the following content on the Platform may be generated or substantially modified by artificial intelligence:

• Job descriptions: Employers may use AI to generate or optimize job posting content. Where AI has been used to generate a job description, this is indicated on the posting.
• Candidate summaries: AI-generated summaries of candidate applications are produced for employer review. These summaries are clearly labelled as AI-generated.
• Interview transcripts and assessments: AI produces transcripts of interview sessions and structured assessment reports. These are labelled as AI-generated output.
• Compatibility scores and match explanations: AI-generated numerical scores and text explanations of candidate-job compatibility are produced for employer review.
• Application quality assessments: AI-generated quality confidence indicators and supporting explanations.

All AI-generated content on the Platform is clearly identified as such in the user interface. No AI-generated content is presented as human-authored.

9. Contact Us

For questions about AI use, bias concerns, or to exercise your rights:

Email: infra-admin@svayz.com
Privacy email: infra-admin@svayz.com
Address: Svayz Ltd, 124-128 City Road, London, England EC1V 2NX

For complaints about automated decision-making, you may also contact the UK ICO (AI and automated decision-making guidance).