How does AI job matching work?

What signals does AI job matching use?

A matching system needs to read both sides of the equation: what the job requires and what you bring. The richest signals fall into two groups. Hard eligibility signals are binary. You either meet them or you do not. Soft fit signals are continuous. You can meet them partially and still be a strong candidate.

Hard signals include work authorisation, physical location or willingness to relocate, and minimum years of experience stated as a threshold. Soft signals include specific tools, programming languages, domain knowledge, seniority indicators in the job title, and cultural cues buried in the responsibilities section.

• Work authorisation: citizen, permanent resident, visa sponsorship required, or open to all
• Location: on-site city, hybrid commute radius, or fully remote with country restrictions
• Time zone overlap: a posting may say remote but require core hours in EST or AEST
• Experience level: entry, mid, senior, staff, or principal, sometimes stated as years
• Skills: required versus preferred, often separated by a heading or phrasing like 'nice to have'
• Industry or domain: fintech, healthcare, defence, each of which can carry its own eligibility rules

Why eligibility filtering matters more than ranking

Ranking tells you which jobs look most relevant. Filtering tells you which jobs you can actually take. Without filtering, a US-based system might surface a role that requires active security clearance, or a UK role that requires the right to work without sponsorship, even if your profile says you need a visa. You waste time, and the employer wastes time.

Filtering also protects you from a subtler problem: applying to roles where you are technically eligible but practically excluded. A posting might say remote but list a required time zone overlap of 9 am to 5 pm Pacific. If you are in Berlin, that is a 3 am start. A matching system that reads the full posting catches this. One that only scans the job title does not.

In 2026, work authorisation rules vary significantly across the four main English-speaking markets. In the US, H-1B sponsorship is a separate employer decision and many postings now state 'no sponsorship available'. In Canada, the LMIA process means most employers prefer candidates with existing work permits. In the UK, a Skilled Worker visa requires the employer to hold a sponsor licence. In Australia, employer-sponsored visas like the TSS 482 carry their own salary thresholds. A matching system that ignores these distinctions will send you into dead ends.

How skills matching works in practice

Skills matching is not a simple keyword count. A well-designed system distinguishes between a skill mentioned once in a bullet and a skill that appears in the job title, the summary, and three separate responsibilities. Frequency and position both signal importance.

The system also needs to handle synonyms and adjacent skills. A posting that asks for 'data visualisation' and one that asks for 'Tableau and Power BI' are asking for overlapping but not identical things. A candidate who lists 'Looker' may be a strong match for one and a weaker match for the other. The matching layer has to make that call rather than treating all three as identical.

Required versus preferred is another distinction that matters. If a posting lists Python as required and R as preferred, a candidate who knows only R should score lower than one who knows both, even if the keyword overlap looks similar on the surface.

• Exact match: the skill appears in your profile and in the required section of the posting
• Adjacent match: your skill is closely related but not identical, for example Kubernetes versus Docker
• Missing required: the skill is in the required section and absent from your profile
• Missing preferred: the skill is in the preferred section and absent, which lowers the score but does not disqualify

Location, time zone and remote work signals

Remote work has made location matching more complex, not simpler. A posting marked remote in the US might mean remote within the US only, remote within a specific state for tax reasons, or genuinely global. The difference matters if you are in Canada looking at a US-listed role, or in Australia looking at a UK company.

Time zone overlap is a separate signal from location. Some postings state it explicitly: 'must overlap with CET business hours for at least four hours'. Others imply it through phrases like 'collaborate daily with our London team'. A matching system that reads the full text can surface this requirement even when it is not in a structured field.

Hybrid roles add a commute radius question. A posting that says hybrid in Austin, Texas is not relevant to someone in Seattle unless they are willing to relocate. The matching system needs to know your location, your relocation intent, and the office location, not just whether the word hybrid appears.

How Break-In Bad uses these signals

Omar, Break-In Bad's AI assistant, reads the full job posting rather than just the title and summary. She extracts location requirements, visa sponsorship language, time zone expectations, and experience thresholds, then compares them against what you have confirmed in your profile.

Jobs are scored on a 0 to 100 scale. Roles scoring 75 or above are submitted automatically. Roles between 45 and 74 are flagged for your review. Roles below 45 are held back with a note on what to address first. This scoring reflects eligibility signals, not just keyword density, so a role that looks relevant but requires sponsorship you cannot get will score lower than one where you meet every hard requirement.

What makes a matching score go up or down?

A score rises when your profile explicitly confirms the hard eligibility requirements and covers the required skills with specific, verifiable detail. Vague entries like 'experience with cloud platforms' score lower than 'three years building on AWS, specifically EC2, S3 and Lambda'.

A score falls when required fields are missing, when your stated location conflicts with the posting's requirements, or when the experience level implied by your titles does not match the seniority the posting describes. A candidate with two years of experience applying to a role that says 'eight or more years required' will score low regardless of skill overlap.

You can improve a score by updating your profile with more specific information, by confirming your visa status clearly, or by adding skills you have but did not list. The matching system can only work with what you give it.

• Add specific tool names rather than category names
• State your work authorisation status explicitly, for example 'US citizen' or 'UK Skilled Worker visa, no sponsorship needed'
• List the cities or regions you are willing to work in or commute to
• Specify your available time zone overlap if you work remotely across regions
• Include years of experience per skill where the difference between two and five years is meaningful

What AI job matching cannot do

Matching systems work on the information in the posting and the information in your profile. They cannot read an employer's internal preferences, the hiring manager's unstated priorities, or the fact that an internal candidate is already lined up. A high match score means you meet the stated requirements well. It does not guarantee an interview.

Matching also cannot compensate for a weak resume or a cover letter that does not address the role. The score tells you whether to apply. The quality of your application materials determines what happens after you do.

Finally, matching is only as good as the data it reads. Postings that are vague, outdated, or copied from a template without editing will produce less reliable scores. If a posting lists fifteen required skills with no distinction between required and preferred, the matching system has to make assumptions. Treat scores on poorly written postings as estimates, not verdicts.