Listing AI skills increases interview invitations by about 8–15 percentage points, while formal certificates add only a modest extra boost; AI credentials can partly or fully offset age and education disadvantages, especially for office-assistant roles, but have smaller effects in creative occupations.

Main Finding

AI skills on résumés causally increase the likelihood of interview invitations. Across three occupations (office assistant, graphic designer, software engineer) and certification types, listing AI skills raises interview probabilities by roughly 8–15 percentage points on average. Effects are largest for technical roles, weaker for creative work, and certifications matter most when compensating for conventional disadvantages (older age, lower formal education).

Key Points

• Magnitude: AI skill signals increase interview invitation probabilities by ~8–15 percentage points overall.
• Occupational heterogeneity:
  • Software engineers: strongest positive effects; interview probabilities for AI-skilled candidates approach ~72–75% in high-usage recruiter subsamples.
  • Office assistants: AI skills strongly compensatory, especially for candidates without tertiary education; university-issued AI micro‑credentials can increase interview probabilities by up to ~25 percentage points versus no AI skills.
  • Graphic designers: smallest or muted effects; recruiter skepticism about AI in creative work lowers valuation of AI skills.
• Certification vs. self-report:
  • Self-declared AI skills already carry substantial value.
  • Company- and university-issued certificates yield modest additional gains over self-report on average; differences are often small and not always statistically significant.
  • Formal credentials matter most when they substitute for missing traditional signals (e.g., a bachelor’s degree).
• Recruiter heterogeneity:
  • Recruiters’ own AI usage and attitudes strongly moderate effects. High AI-usage recruiters reward AI-skilled candidates much more than low-usage recruiters.
  • Skeptical attitudes (notably among creative-role recruiters) reduce the premium for AI skills.
• Signaling interpretation:
  • AI skills act as a market signal of employability and can partially or fully offset traditional disadvantages, consistent with signaling theory while interacting with "muddled information" dynamics in some contexts.

Data & Methods

• Design: Paired-comparison conjoint (forced-choice) experiment presenting recruiters with side-by-side synthetic CVs; respondents chose which candidate they would invite to interview.
• Sample: 1,725 hiring professionals recruited via Prolific (sample described as drawn from major labor markets; analyses focus on hiring-experienced respondents). Each participant completed 10–15 choice tasks, yielding 22,195 pairwise comparisons.
• Roles tested: Office Assistant (administrative), Graphic Designer (creative), Software Engineer (technical).
• Manipulated attributes:
  • Candidate characteristics: age (younger ≈32 vs older ≈60), education level (e.g., associate vs bachelor), gender.
  • Contract type: 6-month fixed-term vs permanent.
  • AI skill treatments (5 levels): no AI, self-reported AI skills, LinkedIn certification, university micro-credential, company-issued certificate.
  • Counter-certificate for advantaged candidate in some pairs to control for certification effects in general (role-specific non-AI certs).
• Stimuli construction:
  • CVs were programmatically generated using an LLM (Google Gemini 2.5 Pro) with careful prompts to ensure internal consistency and role relevance; universities specified were non-elite to avoid prestige confounds.
  • AI skill text was tailored to role-relevant tasks (e.g., automation and Power Automate for office assistants; Midjourney/Adobe Firefly for designers; LLM APIs, PyTorch for engineers).
• Identification & validity:
  • Forced-choice paired design reduces scale and social desirability biases and allows clean estimation of attribute effects.
  • Heterogeneity analyses by recruiter AI engagement and attitudes help probe mechanisms and external validity.

Implications for AI Economics

• For signaling theory and labor economics:
  • AI skills function as a salient labor-market signal with measurable causal returns; however, the informativeness of AI signals varies by context and gatekeeper attitudes, highlighting a nuanced interplay between credential proliferation and signal quality ("muddled information").
  • Formal micro‑credentials can substitute for degrees in specific cases, supporting the increasing feasibility of skills-based hiring.
• For workers and reskilling strategy:
  • Acquiring and advertising AI skills (even via self-report) can improve job prospects, particularly for disadvantaged workers (older, lower-educated).
  • Micro-credentials from universities or recognized companies can be especially valuable when formal degrees are absent.
• For firms and hiring practice:
  • Recruiter AI literacy matters: firms seeking AI capabilities should upskill hiring staff to avoid under‑ or over‑valuing AI signals and to reduce heterogeneity-driven hiring biases.
  • In creative roles, firms should clarify how they value AI‑augmented work and design screening practices that assess genuine creative contribution versus mere AI-produced artifacts.
• For policy:
  • Evidence supports policies that facilitate affordable, verifiable AI micro‑credentials and broader access to AI training, as these can improve employability and act as degree substitutes in some contexts.
  • Monitor distributional risks: differential AI adoption and credential access (e.g., by gender, socio‑economic status) could create new inequalities; complementary policies should target equitable access to AI training and credential platforms.
• Cautions and research directions:
  • The use of LLM-generated CVs is realistic but raises questions about how signal perception interacts with AI-mediated presentation; follow-up work should examine real-world hiring outcomes and long-run employer fit/performance of AI‑credentialed hires.
  • Further research should explore cross-country variation, sectoral spillovers, and the interplay between AI skill signaling and actual task performance.