Legal

AI transparency

Where we use artificial intelligence, how the test works and what no algorithm decides for you.

Version
Version 1.1-UK
In force from
16 November 2026

The summary is written so that a student can understand it. Only the full text below counts legally.

  1. 1The test result is calculated by an algorithm we built, based on the Holland model (RIASEC) and your answers. It isn't a grade and doesn't measure intelligence.
  2. 2Artificial intelligence helps us classify courses and check questions; people review what it suggests.
  3. 3The AI adviser (when available) is a program, clearly labelled as such. It can make mistakes. It isn't a psychologist, a doctor or a lawyer.
  4. 4No university, UCAS or employer uses edumatch to decide whether to give you a place. The result is just for you.
  5. 5You can ask a person at any time to explain your result or review it: office@edumatch.ro.

1.Why this page exists

We believe a student has the right to know how what we recommend is worked out. This page also meets the transparency obligations of the UK GDPR (Article 13(2)(f) and Article 22, on information about the logic of automated processing) and, because edumatch is run from the European Union, of Regulation (EU) 2024/1689 on artificial intelligence (Article 50, applicable from 2 August 2026).

2.Where we use artificial intelligence

  • Calculating the test result: this isn't generative AI but a deterministic algorithm, written by us and described below. The same answers always give the same result.
  • Classifying courses: to give each course a profile of interests, a field and an estimated level of demand, we start from the profiles of equivalent courses in Romania, which were proposed by a language model (Claude, provided by Anthropic) and then checked by a person. For the UK, these profiles are matched to the subject codes in the official data and recalibrated as students give feedback.
  • Maintaining the question bank: an automatic process regularly analyses the questions statistically (for example, if every student answers a question in the same way, it no longer tells students apart) and may suggest new wording. No question is changed without a person's approval.
  • The AI adviser: when available, a chat assistant based on a language model (Claude, Anthropic), which receives your results and public course information and answers your questions about them.

3.How the test works, in short

There are two versions of the test: for Years 7–11 (a shorter exploration) and for sixth form (the full test). The questions are statements you answer on a scale from 1 to 5 (how much it sounds like you) and pairs of situations you choose between. Each question counts towards one or more of the six interest types in the Holland model (RIASEC): Realistic, Investigative, Artistic, Social, Enterprising and Conventional, as well as a few extra preferences (for example working with people or with data, routine or novelty, how much effort you're willing to put in).

Your answers give you a profile: a score for each type. Every course, post-16 route and job in the catalogue has a profile of the same kind. The match score measures how close your profile is to each option's, adjusted for the extra preferences. We show you the options that fit you best, with the explanation: which interest types counted most.

The test doesn't measure intelligence, knowledge, grades or your chances of getting in. It isn't a psychological diagnostic tool. Your result may change from year to year, as you change; that's why we encourage you to retake it every year.

4.What no edumatch algorithm does

  • It doesn't decide or influence admission to a school, college, university or job. edumatch isn't designed for or made available to education providers to select, assess or admit students; it isn't linked to UCAS, and the result is visible only in your account.
  • It has no legal or similarly significant effect on you (Article 22 UK GDPR). It's a recommendation you can ignore.
  • It doesn't use sensitive data (health, religion, ethnic origin, family situation) and doesn't infer it.
  • It doesn't compare you publicly with other students or give you any kind of “score” as a person.
  • It isn't used for advertising or to sell you anything.

In our assessment, edumatch is not a high-risk AI system under Annex III of Regulation (EU) 2024/1689: it doesn't determine access or admission to education providers, doesn't assess learning outcomes and isn't used in an employment context. If we ever develop a feature that falls into that category, we will meet the corresponding requirements before launching it and update this page.

5.Limits and possible errors

Every model has its limits. Course profiles are estimates; a course with the same title can be taught differently at different universities. Answers given too quickly, or “as I should be” rather than “as I am”, give less accurate results. The Holland model describes interests, not talents, and was calibrated on samples that don't perfectly resemble every student. Because the UK version is new, its profiles will be refined as students in the UK take the test and give feedback.

We work to reduce systematic bias: questions are worded neutrally, with no reference to gender, background or town; we regularly check whether a question gives different results for different groups for no good reason; statistics are only published for large groups. If your result seems unfair or wrong to you, tell us.

6.The AI adviser's rules

  • It is clearly labelled as artificial intelligence in every conversation. It never presents itself as a person.
  • Its answers are generated automatically and may be incomplete, out of date or wrong. Check important information at the source, especially on UCAS and university websites.
  • It doesn't give psychological, medical, legal or financial advice. If you tell it about a difficult situation, it will point you to an adult you trust or to Childline (0800 1111, free and confidential); if you or someone else is in danger, 999.
  • Don't give it information about your health, beliefs, family or other people. Conversations are kept in your account and processed by Anthropic only to generate the answer; under the contract, they are not used to train models.
  • It is instructed to stay on the subject of school and career choices and to refuse inappropriate requests. If it still gives a problematic answer, report it to office@edumatch.ro.

7.The data we use to improve the system

We do not train artificial intelligence models on students' personal data and do not pass such data to providers for that purpose. To calibrate the questions and profiles, we use aggregated, anonymised statistics (for example how answers to a question are spread across thousands of students), as set out in the Privacy policy.

8.Human oversight and contact

A person reviews how courses are classified, approves every change to the questions and regularly checks the AI adviser's answers. You have the right to ask for an explanation of your result, to challenge it and to ask for a person to review it: write to office@edumatch.ro from your account's email. We reply within one month.

Questions or requests

Write to us and we'll reply within a month. For requests about your data, the reply can take up to three months if the request is complex; if so, we'll let you know.

office@edumatch.ro

edumatch.ro is run by Călin Spiridon, a private individual, Bucharest, Romania.