AI ATS Tools in Recruiting: What HR Sees

What "AI" means inside an ATS

An applicant tracking system is first a database and a workflow: it stores applications, moves candidates through stages and records who did what. The AI features added on top fall into a few groups, and most of them help a recruiter sort and search rather than decide. Which features an employer actually uses depends on the product and on how it has been set up, so treat the table below as a map of common features, not a description of any one system.

Feature	What the recruiter sees	What it means for your CV
CV parsing	Your CV split into fields: name, job titles, dates, skills	Text lost during extraction is missing from every later step.
Match or fit score	Applicants ordered against the job's requirements	The skills and titles from the ad need to be visible as plain text.
Semantic search	Searches such as "data analyst with SQL" across the whole talent pool, including past applicants	Related terms can be found, but the exact term from the ad is still the safest.
Skill inference	Skills suggested from your job history even when you did not list them	Show skills in context, such as "built sales dashboards in Power BI".
Summaries and drafts	Short profile summaries, draft emails or job ads	A clear first third of the CV gives a more accurate summary.
Screening questions	Applicants who answered "no" to a must-have are filtered or flagged	Answer form questions accurately; they are often read before the CV.

Screening questions are not AI at all, but they sit in the same workflow and often matter more than any score.

Ranking is not the same as rejecting

Most match features sort the list; they do not send rejections on their own. A low-ranked application usually stays in the system, but a recruiter with two hundred applications may never scroll that far. Whether anything is filtered out automatically depends on how the employer configured the system, typically through screening questions about work permits, licences or location.

What has not changed

Everything above runs on the text the system extracted from your file. A two-column layout that scrambles the reading order, dates placed in a page header, or skills inside an image leave the matching step with less to work with, however capable the model. Parsing is still the first thing to check: the parser view shows, field by field, what a parser reads from your CV.

Semantic matching does not make keywords irrelevant either. It can link "customer success" with "account management", but recruiters still filter and search by the terms in the ad, and a model's guess is less reliable than the word itself. Our matching modes make the difference visible: strict, normalised and semantic matching run side by side, and a semantic hit is labelled a possible match, never a confirmed one.

Rules that keep a person in the loop

In the EU, the AI Act classes AI systems used to screen, filter or rank job applicants as high-risk. Providers of these systems, and employers who use them, have duties around human oversight, record keeping and informing the people affected. When these duties apply has been the subject of later amendment proposals, so check the current position if a date matters to you. Separately, the GDPR gives applicants rights where a decision with significant effects is based solely on automated processing.

In practice, these rules push AI ATS tools in recruiting toward recommending and ranking, with the decision left to a person.

What this means when you apply

Use a single-column layout and standard section headings so parsing keeps all of your text.
Repeat the job title and the must-have skills in the wording the ad uses, where they are true for you.
Show skills in your experience bullets, not only in a skills list.
Answer screening questions accurately; a wrong "no" can remove you before anyone reads the CV.
Write for a person as well: a ranked list still ends with a recruiter reading the top of your CV.

For the mechanics behind parsing, matching and ranking, see how ATS works.