How to Spot AI-Written or Overly Perfect CVs with HCIA

The Rise of AI-Generated Résumés

Job seekers now have one-click a``ccess to tools that rewrite, embellish, or entirely fabricate CVs. The result isn't just cleaner formatting — it 's a flood of résumés engineered to beat keyword scanners and look flawless at a glance. For recruiters, the question is no longer "does this ca ndidate look good on paper?" but "is this paper real?"

Why "Too Good to Be True" Is a Measurable Signal

HCIA's compliance engine includes a dedicated analysis called Too Good To Be True (TGTBT) — a 0–100 score that quantifies how suspiciously p erfect a profile appears. It doesn't rely on vague impressions. The LLM evaluates concrete signals:

• All match scores exceeding 95 across every category
• 100% impact density — every bullet point quantified with a metric, no soft statements
• Zero detectable gaps or weaknesses in work history
• Career progression faster than the 95th percentile for the role
• Template detection confidence above 80

When TGTBT crosses 70, recruiters see a red warning banner with the specific indicators listed — not a`` black-box "rejected" stamp, but an eviden ce-backed prompt to look closer.

Beyond Keywords — What HCIA Actually Analyses

HCIA's screening pipeline runs in two tiers.

Core Tier

The core tier extracts a structured candidate profile and scores it against the job description across 11 dimensions: skills match, experien ce match, industry depth, seniority alignment, education, certifications, location, salary expectations, and availability. It also performs a de dicated CV quality analysis covering:

• Completeness — are key sections present?
• ATS compatibility — will parsing systems read it c
• Formatting quality — is it well-structured?
• Keyword coverage — does it align with the job desc
• Quantification rate — how many bullet points conta
• Clarity and language quality — is the writing clea
• Repetition detection — are phrases or structures recycled?
• Image-based detection — is it a scanned image rath

Advanced Tier

The advanced tier runs four parallel deep analyses:

Semantic NLP

Evaluates writing at a forensic level:

• Impact density score — what percentage of bullet pmetrics?
• Action verb score — how many use strong verbs ("les. weak ones ("assisted", "helped", "participated")?
• Template detection score — identifies AI-generatedhe score passes 60, recruiters get a "Template CV Detected" warning with the top three indicators listed: identicalhrasing, or lack of specific evidence.
• Writing breakdown — grammar, conciseness, clarity,
• Hidden requirement matching — finds evidence of skills the job description implies but doesn't explicitly list

Career Intelligence

Reconstructs the candidate's career trajectory:

• Progression trajectory — upward, lateral, downward
• Average and maximum tenure — flags unusually short
• Job-hopping flag — frequent moves without progression
• Loyalty signal — long-term commitment to employers
• Employment gaps — duration, type, and context for each gap
• Industry depth — years of experience per sector
• Career pivots — domain changes with summaries

It surfaces improbable progressions — the kind that lookon't withstand scrutiny.

Predictive Signals

Models future outcomes with explicit uncertainty:

• Retention probability — how likely the candidate is to stay
• Performance potential — estimated contribution lev
• Flight risk — broken down by tenure, overqualification, and salary gap contributors
• Growth potential — capacity for advancement
• Leadership readiness — and whether it's technical, people, or hybrid leadership
• Culture signals — startup vs. enterprise, remote v collaborative vs. autonomous

Every prediction carries a mandatory caveat: "This is a patterns — not a guarantee of future performance."

Compliance

Runs TGTBT analysis alongside:

• Inconsistency detection — conflicting dates, improcation timelines that don't align — each with a severity le vel and recommended action
• Work eligibility verification — confirmed, inferreg detail
• Bias guardrails — actively checks whether the AI'sf demographic bias, producing a guardrail summary the recru iter can review
• PII audit — what personal data was extracted, where it's stored, and for how long

Explainability, Not Black Boxes

Every score HCIA produces comes with an explanation. The explainability waterfall starts at a neutral baseline of 50 and walks the recruiter through each positive and negative contribution:

Direction	Factor	Impact
↑ Positive	10+ years of digital marketing experience
↑ Positive	MBA from top-tier programme	+10
↓ Negative	No direct GMV ownership stated	−10
↓ Negative	Missing experience with enterprise accounts	−10
Final		78

Recruiters can inspect the full justification, view a ra and see every strength and gap itemised. If they disagree, they can override the AI score with a reason — the n the audit trail.

Human Judgment Is the Final Step

HCIA doesn't auto-reject candidates. The pipeline flags, scores, and explains — but stage changes (Shortlisted, Rejected, Interview) are always manual. A CV flagged as "template-detected" might belongofessional résumé writer. A high TGTBT score might reflect an exceptional career, not a fabricated one. The explains the evidence they need to make that call.

What This Means for Hiring Teams

AI-generated CVs aren't going away. The response isn't to ban AI — it's to build screening systems that are:

• Transparent about uncertainty — every prediction c
• Specific about concerns — not "this looks off" but "identical bullet structure across all 12 entries"
• Audi``table — every decision, override, and stage change is logged
• Fair — bias guardrails run on the AI's own output,

HCIA's pipeline surfaces the signals that matter: not just whether a CV is "good," but whether it's real.