The new recruitment: AI as copilot

Traditional recruitment has a scale problem. A hiring process for a mid-level position generates between 150 and 300 applications. A dedicated recruiter can review 20 to 30 CVs in depth per day. That means most applications are rejected with a 6-second glance. Literally: eye-tracking studies show that the first screening of a CV lasts an average of 6 to 7 seconds.

AI does not replace the recruiter. What it does is eliminate low-value tasks (writing generic job posts, copying data between systems, filtering CVs by keywords) so the recruiter can dedicate their time to what really matters: evaluating cultural fit, negotiating terms, building relationships with candidates.

In this module you will learn to apply AI at every stage of the recruitment process, from writing the job post to the final decision. Not as abstract theory, but with prompts, tools and concrete workflows you can implement tomorrow.

Writing job descriptions with AI

The job post is the first touchpoint with the candidate. And most job posts are terrible. Endless requirement lists, empty corporate language, zero information about salary or culture. The result: the best candidates do not bother applying.

What AI can do for your job posts

Generative AI is especially good at three tasks related to job descriptions:

• Generate complete drafts from a minimal briefing: job title, department, key requirements and what makes your company special. In 30 seconds you have a first draft that used to take 45 minutes to write.
• Detect exclusionary language. Words like "ninja", "rockstar", "aggressive" or "young and dynamic" drive away diverse candidates. AI can identify these patterns and suggest inclusive alternatives.
• Optimize for job board SEO. Each job board (LinkedIn, Indeed, Glassdoor) has its own algorithm. AI can adapt titles and descriptions to maximize visibility without sacrificing clarity.

Prompt to generate a job description

This prompt works with any LLM (ChatGPT, Claude, Gemini):

"Write a job posting for [job title] at [company/sector]. The person will report to [position]. Must-have requirements: [list]. Nice-to-have requirements: [list]. Salary range: [X-Y EUR]. Work format: [remote/hybrid/on-site]. Include: job description in 3 paragraphs, main responsibilities (5-7 bullets), requirements (separating must-haves from nice-to-haves), what we offer (real benefits, not empty phrases), and a CTA inviting to apply. Tone: professional but approachable. Avoid exclusionary language by gender, age or origin. Maximum length: 600 words."

Automated review of existing job posts

If you already have published job posts, you can use AI to audit them. A useful prompt:

"Analyze this job posting and evaluate: 1) Clarity of responsibilities. 2) Balance between must-have and nice-to-have requirements (rule: no more than 5 must-haves). 3) Inclusive language. 4) Value proposition for the candidate. 5) SEO for job boards. Score each criterion from 1 to 5 and suggest concrete improvements."

The typical result: most job posts have too many must-have requirements (reducing the candidate pool) and too little information about what the company offers (reducing interest from good candidates).

Automated candidate sourcing

Sourcing is the most time-intensive phase of recruitment. Searching for passive candidates (not actively looking for work) on LinkedIn, GitHub, portfolios, professional networks. An experienced sourcer can identify 15 to 20 relevant profiles in a full day of searching.

Where AI applies in sourcing

• Advanced boolean searches. Instead of building manual queries in LinkedIn Recruiter (title AND skill AND location NOT company), you can ask AI to generate the optimal query from a natural language description of the ideal candidate.
• Profile analysis. You upload a batch of LinkedIn profiles and AI classifies them by relevance, identifying those that best match your target profile. Not just by keywords, but by career trajectory and career coherence.
• Personalized outreach messages. Generic InMails have a 15% response rate. A personalized message mentioning a recent project or a common interest goes up to 40%. AI can generate these messages at scale.
• Hidden talent identification. Candidates who do not have the exact title but do have the skills. A data engineer who has published articles about machine learning could be a viable candidate for an ML Engineer role, even though their current title does not reflect it.

Prompt for personalized outreach

"Generate an InMail message for [candidate name], currently [position] at [company]. I am contacting them for the [title] position. Personalize based on: [specific profile detail: project, publication, skill]. Tone: professional but human, without sounding like a template. Maximum 150 words. Include: why we thought of this person, what makes the position interesting, and a clear CTA (15-minute video call)."

CV screening and triage

CV screening is where AI provides the greatest time savings. But it is also where it generates the greatest risks if misused. Let us look at both sides.

What AI does well in screening

• Structured data extraction. AI can read a CV in PDF, Word or plain text and extract: name, experience, skills, education, languages, in a structured format. This eliminates manual data entry into the ATS.
• Matching against requirements. Automatically compare each CV against the job requirements, generating a fit score. Not as a binary filter (meets/does not meet), but as a ranking with explanation.
• Red flag identification. Unexplained time gaps, date inconsistencies, skills that do not match the career trajectory. Not as grounds for automatic rejection, but as points to explore in the interview.
• Executive summary. For each candidate, a 3-line paragraph the hiring manager can read in 10 seconds and decide whether to see the full CV.

What AI should NOT do in screening

• Automatically reject candidates. AI classifies and suggests. The rejection decision is human. Always.
• Filter by university or former employer. This amplifies socioeconomic bias. Someone having studied at a public university says nothing about their ability.
• Infer protected data. Gender, age, ethnic origin, disability. AI should not attempt to detect or use these variables.

Prompt for CV screening

"Analyze this CV for the [title] position. Must-have requirements: [list]. Nice-to-have requirements: [list]. Generate: 1) Fit score (1-10) with justification. 2) Candidate's strengths for this position (3 bullets). 3) Areas to explore in interview (2 bullets). 4) Executive summary (3 lines). Do NOT infer gender, age, origin or any protected data. Do NOT filter by university or company name."

Optimizing your ATS with AI

Most medium and large companies use an ATS (Applicant Tracking System): Workday, Greenhouse, Lever, Personio, Factorial. These systems manage the candidate pipeline from application to hire. And most are underutilized.

How AI improves your ATS

• Intelligent CV parsing. Traditional ATSs use rule-based parsing that fails with non-standard formats. New AI-powered parsers understand the content regardless of format.
• Semantic search in your talent base. Instead of searching by exact keywords, you can search by concepts: "candidates with experience in legacy system migrations" will find profiles even if they do not use that exact phrase.
• Communication automation. Confirmation emails, status updates, interview scheduling. AI generates personalized messages (not the generic "Dear Candidate") and schedules them at the optimal time.
• Predictive analytics. Which sources generate the best candidates? What is the time-to-hire by department? Where are candidates lost in the pipeline? AI analyzes your historical data and generates actionable insights.

ATS with native AI vs integrations

Two paths. Modern ATSs (Ashby, Lever) already include AI functionality. Enterprise ATSs (Workday, SuccessFactors) require integrations with external tools. If you are evaluating a change of ATS, native AI is a relevant selection criterion. If you cannot change ATS, integrations via API or connectors (Zapier, Make) are the alternative.

Preparing interviews with AI

The interview is the most critical phase of the process. And paradoxically, the least standardized in many companies. Each interviewer asks whatever they want, evaluates according to personal criteria, and takes notes (if they take them) in a different format. The result: inconsistent decisions and systematic bias.

AI for structuring interviews

• Generating interview guides. Based on the job profile and key competencies, AI generates a structured guide with behavioral (STAR) questions, technical questions and evaluation criteria. This ensures all candidates for the same position are evaluated with the same parameters.
• Questions adapted to the candidate's CV. AI reads the CV and generates specific questions: "I see you led the cloud migration at [company]. Tell me how you managed the team's resistance to change."
• Evaluation scorecards. Evaluation templates with predefined criteria and a numerical scale. The interviewer fills out the scorecard immediately after the interview, reducing the primacy/recency effect.

AI for analyzing interviews

• Automatic transcription. Tools like Fathom or Otter transcribe the interview and generate a summary with key points. This frees the interviewer from taking notes and lets them focus on listening.
• Consistency analysis. Compare the candidate's responses with what their CV says. Not as a lie detector, but as a tool to identify discrepancies that merit a follow-up question.
• Cross-interviewer calibration. If three people interview the same candidate, AI can compare scorecards and flag significant divergences for discussion in the debrief.

Prompt to generate interview guide

"Generate an interview guide for the [title] position. Key competencies: [list]. Structure: 1) Opening (2 min): break the ice, explain the process. 2) Behavioral STAR questions (20 min): 4 questions, one per competency. 3) Technical questions (15 min): 3 practical questions. 4) Candidate questions (10 min). 5) Close (3 min): next steps. For each question, include: the question, what to look for in the answer, red flags, and scoring criteria (1-5)."

AI-powered candidate experience

Candidate experience is marketing. Every person who goes through your selection process forms an opinion about your company. If the experience is bad, you not only lose that candidate. You lose their network, their Glassdoor reviews, and potentially a future customer.

Where AI improves the experience

• Pre-screening chatbots. A chatbot on your careers page can answer the 20 most frequent questions (schedules, benefits, selection process), ask basic pre-screening questions, and schedule interviews. Available 24/7, instant response.
• Proactive communication. The worst sin of recruitment is silence. AI can automatically generate and send status updates: "Your application has moved to the interview stage" or "We have decided to move forward with other candidates for this position".
• Personalized feedback. Instead of the generic rejection email, AI can generate a message that mentions something positive about the candidate and suggests development areas. It takes 30 seconds to generate, but the impact on employer brand is enormous.
• Intelligent scheduling. AI coordinates schedules between candidate and interviewers, suggests optimal times, sends reminders, and automatically reschedules if there are cancellations.

Candidate experience metrics

What is not measured is not improved. Three essential metrics:

1. CNPS (Candidate Net Promoter Score): post-process survey. "On a scale of 1 to 10, would you recommend a friend to apply to our company?" Target: > 50.
2. Time to feedback: days from the interview to the result communication. Target: < 5 business days.
3. Drop-off rate: percentage of candidates who abandon the process at each stage. If you lose more than 30% between first and second interview, something is wrong.

Recommended tool stack

You do not need to buy everything. Start with the basics and scale based on hiring volume:

Basic level (< 20 hires/year)

• General LLM: Claude or ChatGPT for writing job posts, CV screening, interview guides.
• ATS: Factorial, Personio or similar (many include basic AI features).
• Cost: 20-50 EUR/month.

Intermediate level (20-100 hires/year)

• LLM + ATS with AI: Lever, Greenhouse or Ashby with integrated AI.
• Sourcing: LinkedIn Recruiter + AI for outreach.
• Interviews: Fathom for transcription.
• Cost: 200-500 EUR/month.

Advanced level (> 100 hires/year)

• Complete suite: enterprise ATS + AI modules for every stage.
• Custom chatbot: for careers page and pre-screening.
• Analytics: recruitment dashboards with prediction.
• Cost: 1,000-5,000 EUR/month depending on volume.

Errores comunes in AI recruitment

1. Automating rejection, not evaluation. Many companies use AI to reject quickly, but still evaluate manually with subjective criteria. It should be the other way around: use AI to evaluate with consistent criteria, and let humans decide the rejection.
2. Not informing the candidate. GDPR (Art. 22) and the AI Act require that you inform the candidate when you use automated decisions. Not doing so is a real legal risk.
3. Blindly trusting scores. A score of 8/10 does not mean the candidate is good. It means their CV resembles what AI has learned as "good". Those are different things.
4. Optimizing only for speed. Reducing time-to-hire from 45 to 30 days is positive. Reducing it to 10 days by sacrificing evaluation quality is not.
5. Not auditing the system. AI learns from historical data. If your past hires had bias (and they probably did), AI will replicate it. Audit results quarterly: who passes the filter and who does not, by demographics.

Ejercicio practico

Puntos clave