VextronAI has AI features that help recruiters assess candidates for a job. This page explains what each one does, what it reads, what it never reads, and who makes the decision.

CV scoring

  • What it does. It compares one candidate’s CV with one job’s requirements. The recruiter writes and confirms those requirements, up to 10 of them, and marks each one must-have or preferred.
  • What it says. For each requirement, the CV either shows it, doesn’t show it, or is unclear. Where it does, the AI quotes the words it relied on, up to 15 words copied exactly from the CV.
  • The result. A fixed rule, not the AI, turns those answers into a band: Strong, Good, Possible, Weak, or Not scored when there isn’t enough to judge.
  • What it reads. The CV saved on the candidate’s record as their CV, and the job’s title, type, town or city, description and confirmed requirements.
  • What’s removed first. Before the CV is read, VextronAI uses pattern-based redaction to remove known identity and contact details and labelled personal facts: the personal details, references and interests sections; the candidate’s name, email, phone, address and links; other people’s contact details and ID numbers; and lines labelled with date of birth, age, nationality, marital status, gender, pronouns, religion, family leave, health or disability. Some personal information may remain. It then trims the CV to a set length, prioritising recent roles.
  • What it never reads. The job’s pay or notes, any other candidate, or the record numbers that identify anyone.

Candidate matching

  • What it does. It suggests which of the agency’s own candidates fit a job, in two steps.
  • Step 1, rules. VextronAI leaves out candidates who aren’t active, are already on the job, or were marked “not relevant” for it, and those clearly missing a must-have skill, or below a minimum rating if the recruiter sets one. It never leaves someone out just because information is missing. It then ranks the remaining candidates using fixed weights for skills, availability and rating.
  • Step 2, AI. The AI checks the top 20 against each requirement in the same way as CV scoring, and a fixed rule gives each a band. A recruiter can ask it to review the next 20.
  • What it reads. Each candidate’s current job title and employer, skills, summary, the date they’re available and their notice period; and the job’s title, type, sector, location, remote policy and start date, and its confirmed requirements.
  • What it never reads. Recruiter notes, the CV itself, pay expectations, where the candidate lives, their compliance records, their rank, or any other candidate.

A person decides

  • Both features give a suggestion with its reasons. A recruiter at the agency decides what to do.
  • VextronAI never rejects a candidate or moves them on automatically.
  • A recruiter can mark a suggestion “good” or “not relevant”, with a reason. “Not relevant” hides that candidate for that job only.

Fairness

  • The AI is told never to mention or infer age, sex or gender, race or ethnicity, nationality, religion or belief, disability or health, pregnancy or maternity, marital status or sexual orientation, even if the text mentions them.
  • The AI is instructed to treat CV, profile and job text as data, and to ignore any instructions within it.

Who runs the AI, and what happens to the data

  • With the Standard option, AI requests go directly to Anthropic (the maker of Claude) and may be processed anywhere in the world, including in the USA. With UK/EU only, they go to Amazon Web Services (AWS) through Amazon Bedrock and are processed in the UK or EU.
  • Anthropic deletes what we send and what it sends back within 30 days. It keeps them longer only if its automated systems flag a request as breaking its usage policy (up to 2 years), or where the law requires.
  • VextronAI never trains AI on your data, and Anthropic’s commercial terms say it “may not train models on Customer Content from Services”.

Known limits

  • The AI can be wrong. It may miss experience that’s worded differently, or read too much into a line. That’s why it quotes its evidence, and why a person checks.
  • A suggestion is only as good as the requirements and the information on the record. When there isn’t enough to judge, the result is “Not scored”, not a guess.
  • CVs, profiles and job requirements can carry bias. Removing personal details and telling the AI to ignore protected characteristics reduces it, but can’t remove every source, so recruiters should review each suggestion with that in mind.
  • Under the EU AI Act, AI used to assess job candidates is treated as high-risk, and we’re preparing for the duties that come with that.

Reporting a bad result

If you’re a recruiter, mark the suggestion “not relevant” and give a reason, and tell us at [email protected] if something looks systematically wrong. If you’re a candidate, ask the agency that holds your information, or email [email protected] and we’ll pass your message to the agency.

What agencies tell their candidates

The agency decides whether to use these features, and it is responsible for telling its candidates about them in its own privacy notice. Our candidate notice wording gives agencies model text to adapt. Our privacy policy explains our part.