EU Pay Transparency Job Ad Checker

Checks job ad URLs or raw job ad text against the EU Pay Transparency Directive ((EU) 2023/970) and its national implementations -- plus a bonus ruleset covering existing US state/city salary-disclosure laws (NYC, CA, WA, CO, IL, and more). Returns a per-ad compliance verdict (compliant / non-compliant / cannot-determine), the specific rule findings behind it (with a reference and source URL), the extracted salary, and a short plain-language fix suggestion.

This tool is informational only and is not legal advice. See "Legal" below.

Why this exists

Directive (EU) 2023/970 required all EU member states to transpose it into national law by 7 June 2026. That deadline has now passed. Some countries transposed on time, some are late, some still have a bill pending -- and several national implementations go further than the Directive's own minimum (e.g. requiring the pay range directly in the job ad, not just before an interview). For anyone posting or auditing job ads across several EU countries -- HR teams, recruiters, job boards, compliance/legal ops -- knowing which rule applies to which country's ad, and whether a given ad actually meets it, is genuinely hard to track by hand. This Actor encodes that per-country research into a versioned, sourced rules file and checks ads against it automatically.

Who it's for

• HR / talent acquisition teams publishing job ads across multiple EU countries who want a pre-publish compliance check.
• Recruiters and staffing agencies auditing client job ads before they go live.
• Job boards and ATS platforms wanting a bulk compliance pass across listings.
• Compliance / legal ops teams building a paper trail of what was checked, when, and against which rule version.

How it works

1. Fetch: for each job ad URL, a plain HTTP GET is sent (no headless browser).
2. Parse: the page's embedded JSON-LD JobPosting data (schema.org) is extracted when present -- Greenhouse, Lever, Workday, Personio, Teamtailor, SmartRecruiters and many other ATS platforms include this for Google for Jobs SEO, and it's a far more reliable source for salary/title/location than scraping visible text. When it's absent, readable page text is extracted instead (scripts/styles/nav stripped).
3. Resolve country: the job's country is taken from JSON-LD jobLocation, then the URL (ccTLD / locale path segment), then a country/city name mention in the ad text, then your input's country field as a final fallback -- see "Input".
4. Check: the resolved country's rule (from rules/eu_pay_transparency.json, or rules/us_state_laws.json for a US jurisdiction) is loaded, and the ad is checked for: a numeric pay range/starting figure, a pay-history question, a gender-neutral job title (heuristic), and a pay-transparency statement (informational).
5. Push: one dataset item per ad, plus a run-level summary written to the key-value store (SUMMARY) with the overall compliance rate and a breakdown by country/company.

Career pages can also be crawled: give a careersPageUrls entry and the Actor follows same-domain links that look like individual job postings (up to maxJobsPerCareersPage each), then checks each one the same way.

How compliance status is decided

For each dimension (salary range in ad, pay-history question, gender-neutral title, pay- transparency statement), the resolved country's rule says either required, not required, or unclear/only required before interview, not in the ad. Combining all dimensions:

• non-compliant: at least one dimension has a confirmed violation of a rule that is currently in force (e.g. the rule requires a range in the ad and none was found). One clear violation is enough, regardless of ambiguity elsewhere.
• cannot-determine: no ruleset exists for the resolved country; OR a ruleset exists but at least one dimension is genuinely ambiguous (e.g. the rule only requires disclosure before the interview, not in the ad itself, and the ad text has no pay info either way -- it may still be disclosed later, off-ad) with zero confirmed violations elsewhere; OR the resolved country's national transposition of the Directive is not yet in force (still a draft bill, enacted but not yet effective, etc.) -- a would-be violation there can't be graded non-compliant against a law that isn't binding yet, so it's downgraded to an informational finding (evaluated against the EU Directive's own baseline / the draft's likely shape instead) and the ad comes back cannot-determine. Check transpositionStatus on the result and rules/eu_pay_transparency.json for that country's actual status and expected effective date.
• compliant: a ruleset exists and is in force, zero violations, zero ambiguous dimensions.

A range that's found but implausibly wide (more than 2x between min and max) is a warning-level finding, not a violation -- it doesn't flip the status, but is worth a look. US bonus jurisdictions are always treated as current law (state/city statutes already in force), so this "not yet in force" downgrade only applies to EU countries.

Input

{
  "jobAdUrls": ["https://boards.greenhouse.io/example/jobs/1234567"],
  "jobAdTexts": ["Software Engineer (m/w/d), Berlin. We offer EUR 60,000-70,000/year..."],
  "careersPageUrls": [],
  "maxJobsPerCareersPage": 20,
  "country": "",
  "autoDetectCountry": true,
  "language": "",
  "includeUsBonusRules": true,
  "maxConcurrency": 3,
  "proxyConfiguration": { "useApifyProxy": true }
}

Field	Type	Default	Description
jobAdUrls	array of strings	--	Public job ad page URLs.
jobAdTexts	array of strings	--	Raw job ad text, one full ad per entry.
careersPageUrls	array of strings	--	Company careers pages to crawl for job links.
maxJobsPerCareersPage	integer	20	Cap on job links followed per careers page.
country	string	--	ISO 3166-1 alpha-2 (e.g. DE), or a US bonus code like US-NYC/US-CA. Used when auto-detect is off, or as the final fallback.
autoDetectCountry	boolean	true	Try JSON-LD -> URL -> ad text before falling back to country.
language	string	--	ISO 639-1 override for which phrase list is prioritized. All supported languages are scanned regardless -- see "Limitations".
includeUsBonusRules	boolean	true	Apply the bonus US ruleset when a job resolves to a US jurisdiction.
maxConcurrency	integer	3	Concurrent fetches.
proxyConfiguration	object	Apify Proxy	Usually not needed for ATS-hosted pages.

Output (one dataset item per ad)

{
  "sourceType": "url",
  "sourceUrl": "https://boards.greenhouse.io/example/jobs/1234567",
  "companyName": "Acme GmbH",
  "jobTitle": "Senior Backend Engineer (m/f/d)",
  "country": "DE",
  "countryDetectionMethod": "jsonld",
  "language": "en",
  "ruleSource": "eu",
  "status": "compliant",
  "findings": [
    { "ruleId": "pay_transparency_statement_found", "severity": "info",
      "message": "Ad includes pay-transparency/equal-pay statement language ('pay transparency').",
      "ruleReference": "Germany's implementing law", "sourceUrl": "https://..." }
  ],
  "extractedSalary": { "min": 70000, "max": 85000, "currency": "EUR", "period": "year", "raw": "70000-85000 EUR" },
  "salaryMin": 70000, "salaryMax": 85000, "salaryCurrency": "EUR", "salaryPeriod": "year",
  "payHistoryQuestionFound": false,
  "genderNeutralTitle": true,
  "transparencyStatementFound": true,
  "fixSuggestion": null,
  "lawName": "...", "lawUrl": "...", "transpositionStatus": "in_force",
  "jsonLdFound": true,
  "robotsNoindex": null,
  "checkedAt": "2026-09-28T12:00:00+00:00",
  "error": null
}

If a URL/text item can't be fetched or processed at all, one item is pushed with every field null except sourceType/sourceUrl and error (a description) -- the run does not stop. A ready-to-use Overview table view is available in the dataset UI/API. A run-level summary (compliance rate, breakdown by country and company) is written to the default key-value store under the key SUMMARY -- see the Actor's Output tab, or GET .../key-value-stores/{id}/records/SUMMARY.

See samples/sample_output.json for a real local run's output.

Rules coverage

Rules live in two versioned, sourced JSON files, each entry carrying a last_verified date and source URLs -- refresh them periodically, laws change:

• rules/eu_pay_transparency.json -- all 27 EU member states' transposition status of Directive (EU) 2023/970, as of last_verified: 2026-09-28: 5 in force (IT, LT, MT, SK, and partially EE), 1 partially in force (PL), 1 enacted but not yet effective (EL, effective 2026-11-01), 18 still a draft bill / not yet started, and 1 unclear (SE, which has publicly said it isn't pursuing transposition and wants the Directive renegotiated). See "How compliance status is decided" above for how a not-yet-in-force country is graded (cannot-determine, not non-compliant).
• rules/us_state_laws.json -- a bonus set of 17 US state/city salary-transparency laws, all currently in force (NYC, NY State, CA, WA, CO, IL, CT, MD, NJ, RI, HI, MN, MA, NV, Cincinnati/Toledo OH, DC).
• rules/eu_research_2026-09-28.json -- the raw per-country research notes (status, law, effective date, requirement flags, source, and caveats) that eu_pay_transparency.json was converted from; kept for provenance/audit trail.

See the coverage table in the build report for the full per-country breakdown (status, law name, effective date, source). A country/jurisdiction with no entry, or a US ad with no specific jurisdiction resolved, comes back cannot-determine rather than a guess.

Testing

• Unit + integration tests (pytest, offline, no network by default): salary/range extraction across currencies and languages, pay-history and vague-salary phrase detection, the gender-neutral-title heuristic, JSON-LD JobPosting parsing, country/ language auto-detection, the rules engine (including a structural check that all 27 EU countries and all US bonus jurisdictions resolve to a well-formed rule with a source), the compliance status algorithm (isolated from the rules file's actual content via synthetic fixtures), and a >=30-ad realistic fixture set across German, French, Spanish, Italian, Dutch, Polish, Portuguese, Swedish, Danish, Finnish, Czech, Latvian, Lithuanian, and English ads, plus JSON-LD HTML fixtures. Run with pytest.
• Live network smoke test (excluded by default): fetches a real Greenhouse-hosted careers page and one real job posting. Run with pytest -m network.
• Local end-to-end run: see the build report for actual pytest output and a real run against fixtures + live public job ad URLs.

Limitations

• This is a rules engine over hand-researched legal data, not a law firm. Every rule entry has a last_verified date and source URLs -- always check those before relying on a result. Transposition status changes; some countries' rules were still moving as of this build (see notes per country in the rules file).
• Salary extraction is regex-based pattern matching, not an NLP/financial model. It requires a currency symbol/code or a "k" suffix (e.g. "60k-80k") to accept a range -- a bare number range with neither (which is at least as likely to be an age, date, or hours-per-week range) is deliberately rejected. Numbers spelled out in words, and highly unusual formats, are not handled. See src/salary.py's docstring.
• Pay-history and vague-salary phrase lists are hand-curated, not a translation API. They cover 15 languages with a deliberately narrow, high-confidence phrase set to keep false positives low; absence of a match means "not detected", not "confirmed absent".
• The gender-neutral-title check is an explicitly-labelled heuristic covering English (legacy gendered nouns) and six grammatically-gendered languages (German, French, Spanish, Italian, Polish, Portuguese) with a curated term list. Other languages return "not evaluated", not a false "neutral". This is a signal worth a human look, not a linguistic authority -- see src/gender.py's docstring.
• Country/language auto-detection is heuristic (JSON-LD > URL ccTLD/locale segment > country/city name mention in text > your input's country). Ambiguous text (e.g. multiple countries mentioned) deliberately returns "not detected" rather than guessing.
• robotsNoindex only records the page's own <meta name="robots"> signal for reference; it is not itself a compliance finding and /robots.txt is not checked.
• Careers-page crawling is a same-domain, path-keyword heuristic (/jobs/, /careers/, etc.), not a sitemap parser -- it can miss JS-rendered link lists (no headless browser is used) or pick up a non-job page that happens to match the pattern.
• US coverage is a bonus set of 17 jurisdictions, not all 50 states -- an ad resolved to a US state/city not in rules/us_state_laws.json (or to plain "US" with no more specific jurisdiction determinable) returns cannot-determine.

Legal

This Actor is an informational compliance-checking tool, not legal advice. It reflects a snapshot of publicly available legal research (see each rule's source URLs and last_verified date in rules/eu_pay_transparency.json and rules/us_state_laws.json) and a set of heuristics for text patterns -- it can miss violations, flag false positives, and go stale as laws change. Before making any hiring, publishing, or compliance decision, verify against the official legal text and/or consult a qualified employment lawyer in the relevant jurisdiction. This Actor only fetches public job ad pages, the same way a browser would; it does not access any account or bypass any paywall or access control.

FAQ

Why did an ad come back cannot-determine? Either its country couldn't be resolved (no ruleset to check against), or the resolved country's rule for some dimension is itself unclear/ambiguous (see findings) with no other confirmed violation. Check countryDetectionMethod and the findings list.

Why didn't it find the salary I can clearly see in the ad? See "Limitations" -- extraction requires a currency symbol/code or a "k" suffix. A range with neither (or spelled-out numbers) won't be picked up. extractedSalary.raw shows exactly what was matched, if anything.

Is a flagged gendered title actually illegal? Not necessarily -- it's a heuristic signal (genderTitleHeuristicConfidence: "heuristic" in the underlying check), not a certified legal determination. Always verify manually.

How is this priced? See PRICING.md for the proposed pay-per-event plan.