TL;DR

AI hallucination happens when a language model generates confident, plausible-sounding information that is factually wrong. In recruitment, this means AI tools can fabricate candidate qualifications, invent job history, or misrepresent credentials — and do so with complete conviction. Any recruiter using AI-assisted screening or sourcing needs to know this exists and account for it.

What AI Hallucination Actually Is

AI models do not look things up — they predict what text should come next. That distinction matters enormously. Large language models are trained on patterns in text, and they generate responses by predicting probable word sequences. When the model lacks reliable training data on a specific topic, it fills the gap by generating something that sounds correct rather than admitting uncertainty.

The term "hallucination" is borrowed from psychology, where it describes perception without an external stimulus. For AI, it describes output without grounding in fact. The model is not lying in any intentional sense. It simply has no mechanism to distinguish between what it knows and what it is making up. Both come out with the same confident, fluent tone.

Hallucinations range from subtle to spectacular. A model might slightly misstate a date, assign a publication to the wrong author, or invent an entire academic credential that sounds entirely plausible. The more specific the query, the higher the risk — because specific details are exactly where training data gets thin.

Why It Matters for Recruitment

Recruitment runs on information accuracy, and AI hallucination attacks that foundation directly. Several high-stakes scenarios arise in hiring where hallucinated content can cause real damage.

The clearest risk is AI-assisted resume screening. When a recruiter uses an AI tool to summarise or evaluate a candidate's background, the tool might generate a plausible-sounding summary that does not accurately reflect what the resume actually says. A candidate who worked in sales operations could be described as having held a revenue leadership role. An incomplete skills list might be padded with adjacent skills the model associated with the job title.

Reference checking via AI carries similar risk. If a tool is asked to assess publicly available information about a candidate — social profiles, publications, portfolio work — it may confidently attribute work or accolades that do not exist. Recruiters who use AI to research candidates without verifying output against primary sources are exposed.

Job description generation is another vector. AI-written JDs can include regulatory requirements, certification prerequisites, or compensation benchmarks that are invented or out of date. Publishing a JD with fabricated compliance requirements creates liability and wastes candidate time.

Candidate-facing AI chat tools present a third category of risk. If an AI assistant answers candidate questions about a role, benefits, or company policies and hallucinates details, the company may face breach-of-expectation complaints or legal exposure if an offer is made and the hallucinated terms are not honoured.

In Practice

A recruiter at a mid-size technology firm uses an AI sourcing tool to build a shortlist for a senior data engineer role. The tool returns a candidate profile summary stating the person holds a Stanford computer science degree and led a machine learning team at a well-known fintech company. Neither fact is accurate — the candidate attended a community college and held an individual contributor role. The recruiter, pressed for time, advances the candidate without checking the original LinkedIn profile. The discrepancy surfaces in background verification, the candidate is withdrawn, and the recruiter loses two weeks of pipeline time.

The failure was not the candidate's fault. The AI tool generated a confident, coherent summary that was wrong. The fix is not to stop using AI — it is to treat AI output as a first draft requiring human verification, not a source of record.

Key Facts