Why we are writing this down
When a tool tells you that you are an 82% match for a job, the obvious question is: according to what? Most job sites will not say. We would rather you could judge the answer for yourself, and so this page walks through each step of the process in the order it happens, including the parts that can go wrong.
This matters more in a year when graduate vacancies have fallen sharply and employers are receiving record numbers of applications. We covered the numbers in why the UK job market is so hard in 2026. When every application counts, you want to know why a tool is suggesting a job.
Step 1: Reading your CV
You upload a PDF, Word document or text file. Your browser extracts the text from the file. If it cannot find enough text, for example because a PDF is a scanned picture, we ask you to try a different file rather than guess.
From that text we look for three things: your skills, your education level, and how many years of experience you describe. This step is not AI. It is a deterministic search against a fixed dictionary of more than 13,000 skills, drawn from the European ESCO skills classification plus our own list for common UK terms. Because it is a dictionary, the same CV always produces the same skills. Nothing is invented, and nothing is hidden from you.
Its weakness is that a dictionary can match words without understanding them. A skill that happens to share a name with something else can be flagged by mistake. That is what the next step is for.
Step 2: AI judges how strong each skill is
Having a skill in a list on a CV is not the same as having used it at work. A hiring manager knows that, and so we try to as well.
The AI step receives your CV text and the fixed list of skills found in step 1. It cannot add or remove any skill. For each one it picks the strongest evidence it can find in the CV:
- Real experience: used in a paid role or genuine work.
- Project: used in a project you built, shipped or delivered, personal, academic or freelance.
- Bare mention: listed with no supporting detail.
- Course or certificate only: appears only alongside a course, certificate or training programme.
- Likely false match: the word is there but there is no real support for the skill.
Each category carries a weight. A skill shown in real work or a real project counts fully. A bare mention counts less. A course counts for about half. A likely false match counts for almost nothing. If the AI step fails for any reason, every skill gets a neutral default weight instead, so the match still runs.
The honest limit is that the AI reads your words, not your actual ability. A well written CV for someone with thin experience can look stronger than a badly written CV for someone with deep experience. We cannot fix that. It is also the reason your CV text is sent to an AI service for this step, which our privacy policy explains.
Step 3: Scoring every live job
Now each live job is scored against your weighted profile. The score combines four things:
- Job coverage (35%). Of the skills this job asks for, how many do you have, weighted by how well evidenced they are?
- CV relevance (15%). Of the skills on your CV, how many are relevant to this job? This stops a short job description with two skills from scoring a perfect 100 for everyone.
- Match depth (25%). How many skills overlap in absolute terms? Gains taper after four, because a fifth overlapping skill tells you less than the second did.
- Experience fit (25%). If the job signals seniority, through phrases like "5 years experience" or words such as "senior" or "lead" in the title, your years of experience are compared to it. A candidate with two years against a three year ask is penalised less than one with none.
The result is a score from 0 to 100 and a label. A strong match needs at least two overlapping skills and a high score. A possible match scores in the middle. A stretch is a job you could reach with some development. Next to each job we list the skills you matched and the skills the job asks for that you are missing, which is more useful than the number.
Step 4: Eligibility is a filter, not a score
When you upload your CV you can tell us your right-to-work status. If your status would require sponsorship and a job states that it does not offer it, that job is removed from your results entirely instead of being given a lower score. A 70% match is not much use if you cannot legally take the job.
Here is the limit, and it is an important one. We detect statements in the advert such as "no visa sponsorship" or "sponsorship not available". We cannot tell what an employer will do if the advert says nothing. A job with no sponsorship statement is not a promise that they will sponsor you. Treat it as unknown and ask. The visa rules guide explains how sponsorship works.
Step 5: Live, checked jobs only
The jobs you are scored against come from Reed's job listings through its official API. Each job is checked with Reed that it is still open when we first load it, and checked again every few days afterwards. Jobs that Reed reports as removed or expired are dropped. We explain why that matters in dead listings and ghost jobs.
This does not make every job perfect. An employer can fill a role the day after our check, and we cannot see that. It does mean far fewer dead links than a search that never checks.
Step 6: A short written summary
Finally, the AI writes a short summary of what your CV signals, which of your top matches look strongest, and which missing skills appear most often. It is told to base your level on real work and education, and to ignore course-only or branded virtual programme credentials when describing seniority. It is told not to invent jobs, companies or skills. The summary is advice, not a decision.
What we keep
We do not keep your CV file. We keep a record of your match results for up to 90 days, linked to an anonymous identifier stored in your browser, so you can come back and see them. We do not send your CV to employers. You apply on the employer or job board site yourself. Read our privacy policy for the full detail.
The honest limits, in one place
- It can only see what is on your CV, not what you can do.
- A dictionary match can be wrong. The AI step reduces this but does not remove it.
- A match score is a guide. It is not a prediction of whether you will get the job.
- Sponsorship checks only catch what the advert states.
- Seniority detection uses wording in the advert. A role that does not state it can be misjudged.
See your own matches
Upload your CV and look at the matched and missing skills for yourself. Free, no account needed.
Match my CV