Greenhouse under the EU AI Act: how a structured ATS relates to Annex III point 4(a)

Greenhouse is in the Dutch tech market the favored ATS at scale-ups and mid-market employers that invest heavily in structured interviewing and bias mitigation. Companies like Mollie, Mendix and Bynder have used it for years. The Greenhouse AI roadmap (matching scores, AI-assisted writing, predictive sourcing) since 2024 changes the profile: from pure workflow ATS to AI-assisting recruitment platform.

This analysis walks through Greenhouse's public AI features, places them against Annex III point 4(a) and ends with vendor questions specific to tech employers.

What Greenhouse publicly offers

Based on Greenhouse product pages and release notes:

• Job Description AI — generative AI for job ads with inclusive language suggestions
• Candidate match scores — AI suggestions for matching candidates on a requisition
• AI-assisted candidate sourcing — external candidate discovery via integrated tools
• Predictive analytics — funnel analyses, time-to-hire predictions, source effectiveness
• Structured Interviewing — Greenhouse's core feature, supports scorecards and consistent evaluation (partly rule-based, not AI-driven)
• Anonymous Reviews — masking of demographic signals during evaluation

Greenhouse's philosophy is pronounced: structured interviewing, scorecards, bias mitigation. AI is added as supporting layer, not as decision-maker. For compliance that is profile-wise stronger than many competitors — but it does not absolve the deployer.

The seven checks applied to Greenhouse

1. Does the AI rank or score candidates?

Candidate match scores do this explicitly. For deployers actively using match scores for shortlisting: 4(a).

2. Does the AI optimize who sees a vacancy?

AI-assisted sourcing and integrated job boards determine targeting. Check which channels you use and which AI works within them.

3. Is CV parsing really only parsing?

Greenhouse parses CVs and uses information for matching. Skills inference is growing in the product roadmap. Check your release version.

4. Is the chatbot logistical or selective?

Greenhouse offers no autonomous screening chatbot. Candidate communication is largely template-based and human.

5. Does the assessment tool measure behavior or performance?

Greenhouse itself offers no psychometric assessments. Integrations with HackerRank, Codility, Karat (technical assessments) fall under that specific vendor.

6. Does the system continue post-hire?

No, Greenhouse is pre-hire. Greenhouse Onboarding (separate product) is workflow without heavy AI scoring.

7. Can you substantiate the vendor claim?

Greenhouse publishes AI positioning and privacy documentation. For model-specific documentation and bias evaluation: request in writing via Customer Success.

The classification call

Greenhouse deployments typically run:

• Greenhouse without candidate match scores active: structured ATS, limited AI Act relevance (Annex III point 4 not evident)
• Greenhouse with match scores and AI sourcing active: defensive 4(a) high-risk

For most Dutch Greenhouse users — mid-market and scale-up tech — the structured interview approach helps the bias mitigation argument, but match scores and sourcing are unavoidable in the modern configuration.

Vendor due diligence for Greenhouse

What to do now

For Dutch Greenhouse users — especially in tech and scale-up segment — this is the path: feature audit this week, written vendor confirmation within 30 days, documentation via HR AI hub. Your Structured Hiring discipline is your strong point for oversight; build your AI Act dossier on it.