AI transparency notice

It reads a job description into a list of requirements and searches a company’s own applicant history for people who may match. Where the company turns it on, it also searches a licensed database of professional profiles of people who have not applied. It produces a match score for each person from how many requirements their CV or profile evidences and what each requirement is worth. Every score is accompanied by the exact lines that earned it.

A language model reads job descriptions and CVs. A second check may suggest that a requirement is evidenced, given a verbatim quote, but a suggestion never changes a status or a score. The model does not produce the number: the score is fixed arithmetic over the requirement coverage, so the same inputs give the same result and justification text cannot move a score or an order.

A model may also suggest clearer requirements for a job description. A person accepts or rejects each suggestion.

A requirement read between the lines rather than stated reaches “Partly meets” and never “Meets”, because “Meets” is the status that has to survive being challenged.

The interview is a spoken conversation with an AI interviewer. Only audio is used — the camera is never requested — and the conversation is recorded and transcribed on our side. Only the questions that trace to a requirement of the job are marked, each against a written description of a good answer; the warm-up, anything about the company and the candidate’s own questions are never marked.

A separate check reads the text of each answer — only the words, never how they were said — to notice when answers are off-topic, abusive or an attempt to instruct the interviewer. It can lead the interviewer to steer back to the question and can mark the interview for a person to review. It ends the interview after a second abusive answer, following one warning, unless the company chooses only to mark it. The candidate is told this beforehand. Off-topic answers never end it. It never produces or changes a score and never declines anyone. An interview it ended early waits for a person.

It does not infer gender, ethnicity, religion, nationality or age, and no filter may rest on anything other than a fact about the work. No filter, score, ranking or automatic selection uses gender, nationality or national origin, race or ethnicity, religion, age or marital status. It does not score emotion, affect, tone or facial expression in an interview, and it does not score accent, fluency or pace unless the language is a genuine requirement of the role.

A company can let candidates who reach its pass mark move on and be invited to the next round without a person. By default, those invitations and every decline email wait for a person to send them. An interview in which a must-have was marked 1 of 5, or not covered, waits for a person whatever its score. Nobody is declined automatically unless the company turns that on, with a person reviewing a sample and time to reverse it. Rounds marked as needing approval never proceed without a person. A person decides who is hired.

Ask the company you applied to, or the one that contacted you. Because the reasoning is recorded at the time rather than reconstructed afterwards, they can show the requirements, the status of each, and the lines from your CV or profile behind each one. How data requests are handled is on the privacy page.