TL;DR

Generative AI refers to AI systems that produce new content - text, images, code, audio - based on patterns learned from large training datasets. In recruitment, generative AI is used to write job descriptions, personalize outreach messages, generate interview questions, summarize candidate profiles, draft offer letters, and produce content at a scale that was previously impossible without proportional headcount increases.

How Generative AI Works

Generative AI is not a search engine and not a database - it generates outputs that have never existed before. The distinction matters. When a recruiter asks a generative AI system to write a job description for a senior data engineer, the system isn't retrieving a job description from a database. It's generating a novel piece of text by predicting the most likely sequence of tokens given the prompt and the patterns it learned during training.

The foundation of modern generative AI in recruitment is the transformer architecture, specifically large language models (LLMs) trained on vast text corpora. These models learn the statistical relationships between words, phrases, and concepts at enormous scale. The result is a system that can produce coherent, contextually appropriate text on almost any topic, in almost any style, given the right instruction.

In practice, recruitment applications of generative AI involve two layers beyond the base model. Fine-tuning trains the model on domain-specific content - job descriptions, recruiter emails, interview frameworks - to improve output quality in recruitment contexts. Prompt engineering shapes the model's output through precisely structured instructions: specifying tone, length, required inclusions, and what to avoid.

Retrieval-augmented generation (RAG) extends generative AI further. Rather than relying solely on what the model learned during training, RAG systems pull relevant data from external sources at inference time - a company's existing job library, a candidate's Bullhorn profile, a client's job order - and inject it into the prompt. The model generates content grounded in real, current data rather than general patterns.

Why It Matters in Recruitment

Content creation is a significant portion of recruiter time, and most of it doesn't require human judgment. SHRM estimates that recruiters spend 20-30% of their working hours on content and communication tasks: writing job descriptions, personalizing outreach, drafting rejection messages, summarizing interview feedback, and preparing offer documentation.

Generative AI compresses these tasks dramatically. A well-prompted model produces a first-draft job description in 30 seconds. A recruiter who writes 15 job descriptions per week spends 6-8 hours on that task. With generative AI handling first drafts, that drops to 90 minutes of editing and refinement. Over a year, that's roughly 300 hours recovered per recruiter.

The quality argument is more nuanced. Generative AI produces better average quality than the average human first draft - consistent structure, clearer language, fewer omissions. But it produces worse quality than a skilled recruiter's best work applied to a role they understand deeply. The practical model is human-AI collaboration: AI handles the scaffolding, human expertise handles the signal.

Generative AI in Practice

An in-house [talent acquisition](/glossary/talent-acquisition) team at a 3,000-person SaaS company is hiring aggressively - 200 positions across engineering, sales, and operations over a 6-month period. Writing unique, compelling job descriptions for each role historically took 45-60 minutes per posting, consuming 150+ hours of TA coordinator time before sourcing even begins.

The team implements a generative AI workflow inside their Greenhouse integration. Hiring managers submit a structured intake form. The system generates a complete job description using the intake data, the company's established tone and inclusivity guidelines, and compensation benchmarks from their internal data. The TA coordinator reviews and edits the draft - a 10-minute task versus a 45-minute writing task.

Candidate outreach gets the same treatment. The recruiting team uses Candidately to generate personalized InMail messages for passive candidates, pulling each candidate's current role, skills, and career trajectory into a prompt that produces a message specific to them rather than a template. Reply rates on personalized AI-generated outreach average 28% versus 12% for templated messages in the same campaign.

Key Considerations