How our matching works, in plain English

Two things are ranked automatically. First, the roles in your feed are ordered by how well they fit your profile. Second, when an employer posts a role, candidates who meet its requirements are ordered and presented as a shortlist.

Ranking is not a decision to hire or to reject. A human recruiter reviews the shortlist and makes every shortlisting, contact and hiring decision. But ranking materially affects who gets seen, so we treat it as substantially assisting a decision that affects your interests — which is why this page exists.

• Skills, with proficiency and how recently you used them, mapped to our AI taxonomy
• Role family and seniority band, including years of experience in AI specifically
• The text of your profile, projects and experience, compared against the role description
• Your stated day rate or salary expectation
• Your availability and notice period
• Your location and preferred work arrangement
• Your work-rights status and any security clearance

Age, sex, gender, race, ethnicity, national origin, disability, religion, marital or parental status, sexual orientation, political opinion, and trade union membership.

These are not merely excluded by policy. They are not collected at signup, there is no field for them on a profile, they are not extracted from your CV, and the ranking code has no parameter that could receive them. Employers cannot filter or search on them either — attempts are blocked and recorded.

Hard requirements are applied first and are absolute. Work rights, security clearance and the employer's rate ceiling either match or they do not; no amount of similarity elsewhere can overturn them.

What remains is scored on six dimensions — skill overlap, seniority fit, similarity between your project history and the role description, industry adjacency, rate alignment, and availability — combined using weights set at platform level. Skills the employer marked as must-haves count three times as much as nice-to-haves, and skills you have not used recently count for less.

Location and engagement type usually reduce a score rather than exclude you, and the reason is shown. An employer who genuinely cannot flex can mark a role strict, in which case a mismatch does exclude — and the role says so.

Every ranking stores what matched, what we inferred from adjacent experience and where the gaps were, along with the weight version used. That record is kept, and it is what lets us reconstruct any past decision rather than reasoning about it after the fact.

You can request an explanation from your privacy settings. The request itself is logged, and we respond with the stored reasoning for the ranking you asked about.

You can turn automated matching off entirely from your privacy settings. Your profile remains searchable by employers, existing rankings are deleted, and you can still browse and apply to every open role. You will see fewer relevant roles, because nothing is being ordered for you — that is the trade, and we would rather state it than bury it.

No offer, rejection or shortlist decision is made by the system. Recruiters mark matches good or bad, which is how we measure whether the ranking is any good — and those labels are the only source of our published precision and recall figures. We do not estimate them.

We monitor outcomes in aggregate for disparate impact. That monitoring uses non-identifying data and cannot be traced back to you.

Candidate data is retained for 24 months after your last activity unless you delete it sooner. You can download everything we hold, or delete your account, from your privacy settings. Records that a deletion occurred are retained with your identity removed — that is the evidence the deletion happened.