How we classify listings
Every listing gets three labels: language, type and pay. Here is exactly how they are made — and where they can go wrong.
1. Where listings come from
We read the public career pages of companies through their applicant-tracking systems (Greenhouse, Lever, Personio and others) twice a day. We respect each site's robots.txt, identify ourselves as InternsValleyBot and never scrape job boards such as LinkedIn, Indeed, StepStone, XING or Glassdoor. Employers can also post directly.
We store only a short excerpt and our own summary, and always link to the original posting — you always apply on the company's site.
2. The labels
Language — what the posting requires, not the language it's written in:
Type:
Pay — and whether it meets the minimum wage (€13.90/hour in 2026, €14.60/hour in 2027):
3. How a listing is labelled
- An AI model (Anthropic's Claude) reads the posting and suggests each label. It knows German phrases such as “fließende Deutschkenntnisse” (German required) or “Deutschkenntnisse von Vorteil” (German is a plus).
- Every label needs a quote. The model must copy the exact sentence from the posting that supports it. We check that the quote really appears in the posting. No quote → the label becomes “unclear”.
- Each label gets a confidence score. Below 70%, a stronger model takes a second look, and the listing goes to our review queue.
- Unpaid voluntary internships longer than 3 months are always reviewed by a person.
- Monthly pay is converted to €/hour (monthly ÷ (hours per week × 4.33)) only when the posting states both the amount and the hours.
4. What the badges mean
- Auto-classified — labelled automatically as described above.
- Checked by our team — a person reviewed and corrected the labels.
- Verified by employer — the company submitted the listing and the labels itself.
Every listing also shows its source and when we last checked that it is still online.
5. When we get it wrong
Automatic labels can be wrong. Always read the original posting before applying. If you spot a mistake, use Report incorrect info on the listing — we review every report. Nothing on this site is legal advice.
Background on the rules: our guides.