AI candidate scoring tools use whatever data their category can capture: parsers read application documents, assessment platforms score test and work-sample responses, interview tools analyse what candidates say and how they say it, and sourcing tools score profile and network data. The question matters because a score is only as trustworthy as its inputs, and the data a tool leaves out often counts as much as the data it uses. Tools built for AI candidate evaluation tie these inputs to defined, job-related criteria rather than a general impression.
Resume screening and parsing tools
These read the application itself. Typical inputs are parsed resume and profile fields: job titles, skills, years and relevance of experience, education, certifications, and employment dates. Better systems look past keyword presence to context, using natural language processing to interpret what a resume says against the job requirements, and some learn from past hiring patterns. Some platforms also enrich a profile from public sources like LinkedIn and GitHub. The weakness is that self-reported documents reward phrasing and polish, which is why blind screening (hiding names and photos) is a common safeguard.
Skills-assessment platforms
These score evidence a candidate generates for the role. Inputs are responses to structured tests: coding challenges, SQL or spreadsheet exercises, multiple-choice questions, and written or video answers, each rated against a rubric. Many platforms let teams build custom assessments across those question types and score open-text and video responses automatically, alongside integrity signals such as copy-paste tracking, tab and device monitoring, and webcam snapshots. Game-based and situational-judgment assessments capture traits like judgment and problem-solving that a resume cannot show.
AI interview and video scoring tools
These score spoken answers. The core input is a transcript, analysed with natural language processing for semantic content, sentence structure, how directly a question was answered, and competency markers. Some systems also weigh vocal prosody — pace, pauses, and tone — and some vendors combine video interviews with assessments. Scoring usually compares responses against a competency model or a baseline built from strong performers, so the baseline is itself a data choice worth scrutinising.
Sourcing and matching tools
These score prospects before they apply. Inputs lean on public professional data: profile fields, career history over time, skills adjacency, and relationship or network signals. Some tools structure person and company data across time so a candidate's trajectory and tenure are visible, not just their latest title, and use relationship signals to surface warm introductions. The weakness is data quality — public profiles can be thin or stale, and inferred or self-reported attributes can quietly reintroduce bias.
Questions to ask a vendor about its data
- What inputs does it score, and how is each weighted?
- What is explicitly excluded — protected traits such as race, gender, age, and disability, and proxies like names, photos, and graduation years?
- Does the vendor test for proxy variables rather than assuming they are absent?
- Can it explain, per candidate, why a score came out the way it did?
- Does it run adverse-impact or bias audits, and how often?
- How long is candidate data retained, and how is it deleted?
For how these inputs combine into an order, see how AI ranks candidates. Teams comparing categories can start with automated candidate screening tools.