The real issue: reliability, not "catching liars"

You often read that 78% of resumes contain at least one embellished piece of information (HireRight 2025 study). The temptation is to build an AI that "unmasks" every exaggeration: inflated title, career gap, frequent job changes, slightly optimistic "fluent English"…

That's a trap. Most of these "signals" are neither provable nor reliable — and, above all, they overlap with protected characteristics (a gap can be parental leave, illness, a career change; frequent job changes can reflect a precarious sector, not instability). A tool that penalizes them manufactures discrimination — and exposes you legally.

Candidalyze takes the opposite stance: verify only what is factually provable, and refuse to judge career choices. It's more honest, more useful in interviews, and compliant with GDPR and the EU AI Act.

What the AI checks: 4 factual inconsistencies

A dedicated pass re-reads every experience (title + dates) and every degree (title + year), and flags only these inconsistencies, each proven by the written dates:

Reversed dates — an end date earlier than the start date on the same position.
Real overlap — two full-time experiences overlapping by at least a full year. (A single transition year shared between two consecutive positions — "2015–2019" then "2019–2024" — doesn't count: that's normal and isn't flagged.)
Illogical degree order — a higher-level degree carrying an earlier year than a lower-level one (e.g. "Master's 2010" while "Bachelor's 2012").
Incompatible total duration — a self-declared seniority ("X years of experience") that the actual sum of dated periods doesn't support.

If the information isn't literally written on the resume, the AI ignores it. An empty result — no inconsistency — is common and perfectly normal.

What this changes in the Candidalyze report: these inconsistencies appear in the "Verify before deciding" section, each quoted word for word, with a neutral interview question to clear it up. When factual inconsistencies are proven, the reliability of self-reported claims drops and the score is pulled back to the "consider" zone until they're clarified — factually, without imputing intent.

What the AI refuses to flag — and why that's a deliberate choice

By design, the tool never flags:

periods without activity ("gaps");
the frequency or duration of positions ("job-hopping");
"inflated" titles or overqualification;
self-assessed language or skill levels.

These aren't inconsistencies: they're value judgments or proxies often correlated with protected characteristics (age, health, family situation, origin). Penalizing them would be both unreliable and contrary to anti-discrimination law, GDPR and the AI Act. An internal safeguard even blocks this kind of vocabulary in reports.

This isn't a shortcoming of the tool. It's its design.

For anything that needs context: a question, not a verdict

An unusual title, a skill critical to the role, a transition: these often deserve a closer look — but in the interview, not from an algorithm ruling in your place. So the AI generates a neutral question, focused on an observable behavior rather than a personal characteristic:

"You mention expertise in [technology]: can you walk me through a concrete project where you used it, and your exact role?"
"Between these two positions, how did you approach that period?" (without presuming anything)

The candidate provides the context. The AI hands you a well-framed question; you keep the decision.

Why this is an advantage, not a limitation

AI resume scoring is classified "high risk" by the AI Act (Annex III, point 4), with obligations applicable from August 2026: transparency, human oversight, non-discrimination, traceability. A tool that refuses discriminatory proxies by construction isn't less powerful — it's defensible. Facing a candidate, a works council or an audit, you can explain every signal: it's quoted, factual, proven by the dates.

Transparency: the candidate knows AI is involved, on objective, documented criteria.
Non-discrimination: no personal motive, no judgment of career trajectory.
Human oversight: the AI verifies and questions; you decide.

Frequently Asked Questions

Does the AI detect resume lies? No, and it doesn't claim to. It verifies 4 provable date inconsistencies and signals a drop in the reliability of self-reported claims. Everything else belongs in the interview, never in an accusation.

Does the AI penalize a career gap, a career change or frequent job changes? No, deliberately. These are neither flagged nor scored down. They may, if the role warrants it, lead to a neutral interview question — nothing more.

Is it legal to use AI for this? Yes, provided you respect GDPR and the AI Act: transparency toward candidates, objective criteria, human final decision. Candidalyze is designed around these principles from the ground up.

Do candidates know their resume is analyzed by AI? Yes. Transparency is a legal requirement; best practices recommend informing candidates in the job posting or upon receiving their application.

Does the AI analyze the candidate's social media? No. Candidalyze only analyzes the resume provided. No external data is collected or cross-referenced.

Conclusion

The perfect resume doesn't exist — and that's fine. The AI's job isn't to hunt down the imperfections of a career, but to objectively verify what can be verified — dates that contradict each other — and to hand you back control for everything else, with well-framed questions.

That's the difference between a tool that screens in your place (and exposes you) and one that helps you decide with full knowledge of the facts, even when analyzing 100 resumes in a day.

Ready to screen faster, and more fairly? Try Candidalyze free — 5 free analyses, no credit card required. See for yourself what the AI verifies… and what it leaves for you to decide.

Resumes: the Factual Inconsistencies AI Checks (and What It Refuses to Judge)