How it works, and what the numbers do not mean

The pipeline

1. 01

   You name a role

   A UK job title is all we need. A CV is optional. Every result is in the UK, so we do not show roles you could not take on a UK visa.

2. 02

   We pull live ads

   We search Reed and Adzuna for that title and keep only UK locations. An ad that just says "Remote" is left out because it could be anywhere, but "Remote, UK" is kept.

3. 03

   We read the full descriptions

   For Reed ads we fetch the full job description. Adzuna only gives us a short snippet, so we can read fewer skills from that source.

4. 04

   Employers are matched to the register

   We compare each employer name with the Home Office Skilled Worker register, allowing for small spelling differences. Then we check the match in both directions, so two names that only share one word cannot count as a match. Recruitment agencies are never treated as confirmed sponsors.

5. 05

   Company job boards are added

   Some employers publish jobs on a hiring system we can read, such as Greenhouse, Ashby, Workable or Recruitee. When we find one, we fetch that board and add its roles. For those roles we are certain who the employer is. Employers we do not know yet are checked in the background after your results appear, so you never wait for it.

6. 06

   Skills are counted

   We count skills across every ad we could read in full, mostly Reed descriptions, plus Adzuna snippets where they help. Tools that clearly belong to a skill count for it, so GitHub Actions counts as CI/CD.

7. 07

   Results are ordered

   Verified sponsors come first. After that we order by how long the company has held its licence, then by how recent the ad is. The licence length comes from our own archive of past registers, which only goes back to 2023, so it shows how long we have seen a licence and not its true age.

8. 08

   Your CV is compared if you upload one

   We compare your CV with the skills asked for in this role's ads. An AI service then reads your CV like a hiring manager. For each top skill it decides whether you have really shown it in your work, even if you used different words, and it must quote the line from your CV that proves it. We check that the quote is really in your CV. The AI also decides which missing skills would actually stop you getting the role and which are only nice to have. It also works out your level, such as graduate or senior, so you are not shown jobs far above you. Your CV text is sent to the AI service to do this.

What each confidence tier means

Honest accuracy

For roles taken from a company's own hiring board, we are certain who the employer is. For roles from job sites we work it out from the employer name. When we checked 100 of those by hand, the name match was right about 59 times in 100. An earlier check of 50 scored 68%. We show that number instead of hiding it, because a matching name is not proof.

Skills in job ads are found by looking for the skill names and the tools that belong to them. This works well on full Reed descriptions and poorly on short Adzuna snippets, so a skill can be undercounted if it only appears deep in a cut-off description. The AI review of your CV can also be wrong about whether a skill is blocking, and it can give a slightly different answer if you run the same CV twice. Full figures are in our accuracy notes, called ACCURACY.md.

Where verified coverage is thin

We can only verify an employer when it publishes roles on a hiring system we can read. That is common in technology, finance and fast-growing companies. It is rare in healthcare, hospitality, retail and the public sector, where employers use systems with no public job feed. Roles in those areas show up with a name match instead. So if your search has few verified roles, that may say more about the sector than about the sponsors in it.