KeyMatch vs Keyword Search — How AI Job Matching Actually Works

The keyword search problem

Most online job boards use keyword search to match your CV to roles. You upload a CV that lists "Python", and the search returns every job that contains the word "Python" in the JD. Roles that say "Pandas" or "scikit-learn" or "Django" — and would obviously be relevant to a Python developer — don't show up unless they also use the literal word "Python" somewhere.

This is fine for finding any-job-with-this-word, terrible for finding the-right-job-for-you. The misses are systematic: senior roles tend to use higher-level descriptions ("data infrastructure"), startups use frameworks instead of languages ("Next.js" instead of "JavaScript"), and entire job categories use industry shorthand a keyword index doesn't understand.

What semantic matching does differently

Semantic matching reads your CV and the job description as meanings, not character strings. Internally, both texts get embedded into a multi-dimensional space where related concepts sit close together. "Python" sits near "Pandas" sits near "scikit-learn" sits near "data analysis" — even though none of those four strings overlap character-for-character.

When KeyMatch scores a role against your CV, it's measuring the distance in that space. Closer = more relevant. The score is on a 0–100 scale: 90+ means your skills overlap heavily with what the JD demands; 50–70 means meaningful overlap with some gaps; below 40 means you're significantly under-qualified.

The practical difference shows up immediately. A keyword search for "Python" might surface 200 roles. KeyMatch surfaces those 200 plus another 80 roles labelled "data engineer", "ML engineer", "backend engineer" that didn't say the word "Python" in the JD but are obviously relevant given the work being described.

Why this matters for the application stage too

The same matching engine that surfaces relevant roles also tells you which skills your CV is missing for a specific role. So instead of generic feedback ("your CV looks fine"), you get the actual list:

• Missing: "stakeholder management" — the JD mentions this 4 times, your CV doesn't.
• Missing: "ETL pipelines" — concretely the gap between "Senior Data Analyst" and "Data Engineer".
• Strong overlap: SQL, Python, Pandas, dashboard tools.

You then have a precise list of what to add to your CV (if you have the experience) or what's genuinely missing (if you don't). CV Improver picks up that gap analysis and offers section-level rewrites that bake the missing keywords into your existing experience bullets without making the CV sound robotic.

Concrete examples of where keyword search fails

A few real cases from KeyMatch runs we've reviewed:

1. The "frontend developer" who's actually a TypeScript engineer. A candidate's CV listed React, TypeScript, Next.js. Keyword search showed 40 roles. KeyMatch showed those 40 plus another 18 labelled "product engineer", "fullstack engineer", "UI engineer" — all genuinely relevant, all missed by literal string matching on "frontend".

2. The PM who's better at growth marketing. CV listed "product manager" (their official title), but actual experience emphasised A/B testing, growth funnels and acquisition metrics. KeyMatch surfaced senior growth marketer roles at much higher scores than mid-level PM roles. The candidate had been ignoring an entire job category that suited them better than their current title.

3. The applied-ML researcher mislabelled as "data scientist". PhD background, three years applied research, CV listed "Data Scientist". Keyword search returned generic DS roles. KeyMatch surfaced ML Engineer, Research Engineer and Applied Scientist roles at top-of-stack AI labs (Cohere, Mistral, DeepMind) at higher scores. The candidate hadn't realised those titles were the natural next step.

How to read your KeyMatch score

A 0–100 score sounds simple but each band carries different strategic advice:

90–100 — apply now. Your CV is well-aligned with this role; the filtering risk is low. Apply within 48 hours and tune the CV to the JD with CV Improver before submitting. Expect a meaningful interview- rate from this band (5–15% across applicants in this score range).

75–89 — apply with care. Solid fit; one or two skill gaps. Worth applying. Read the gap list carefully — if the missing skills are specific tools you can mention from a side project or coursework, add them to the CV before applying. If they're 5+ years of experience you don't have, be realistic.

60–74 — apply if you're stretching. This is the upper bound of "reach" applications. Worth applying if you're in a hot market or if the company is one you're particularly motivated about, but don't spend the bulk of your application time here. Interview rate drops to 1–4%.

40–59 — only if you're prepared to invest. You'll need to make the case for why you can do the role despite the gaps. This works for genuinely transferable backgrounds (e.g. quant analyst applying for ML engineering) but isn't for spray-and-pray.

Below 40 — skip. Your time is better spent in higher-scoring bands. The exception: a role at a dream company where you genuinely want to wait it out for 6 months and re-apply with closer-fit experience.

Try it on a real CV

The fastest way to see the difference between keyword search and semantic matching is to run KeyMatch on your own CV. Upload it once, paste a target JD (or just a role title like "Software Engineer"), and watch the ranking come back with a per- skill gap breakdown for every live UK and EU role.

If your top 10 matches surprise you — roles you'd never have searched for, that score 90+ against your CV — that's the keyword-search blind spot we just walked through. Apply to those first.