diff --git a/README.md b/README.md
index adee035..0bd3b15 100644
--- a/README.md
+++ b/README.md
@@ -156,7 +156,10 @@ Being upfront about what this package does **not** do is part of its job:
    `'public'` (which also exempts them from the catch-all). Names or other non-pattern PII typed
    into unclassified fields can still slip through.
 3. **Heuristics are best-effort.** The non-schema mode catches common field names and
-   email/phone/IBAN patterns. Novel field names or unusual formats can slip through — use
+   email/phone/IBAN patterns. Phones match in international (`+49 …`, `0049 …`) and German/Austrian
+   national format with a separator after the area code (`030 1234567`, `0221/123456`,
+   `(0221) 123 456`); a national number written as one digit run (`0301234567`) is left alone so
+   ids and postal codes are not mangled. Novel field names or unusual formats can slip through — use
    `strict: true` where that is unacceptable.
 4. **Quasi-identifier re-identification.** Kept generalized values (postal code + city +
    birth year + consumption data) can, in small populations, narrow down individuals. If you
diff --git a/src/anonymization.test.ts b/src/anonymization.test.ts
index d56a9f4..640b698 100644
--- a/src/anonymization.test.ts
+++ b/src/anonymization.test.ts
@@ -338,6 +338,31 @@ describe('non-schema-aware mode', () => {
     expect(anonymizer.scrub(once)).toEqual(once);
   });
 
+  it('scrub replaces national phone numbers, idempotently', () => {
+    const phones = ['030 1234567', '0221/123456', '0170-1234567', '(0221) 123 456', '089 123 456 78'];
+    for (const phone of phones) {
+      const once = anonymizer.scrub({ text: `Hotline: ${phone}.` }) as any;
+      expect(once.text).toMatch(/^Hotline: \+00 000 \d{8}\.$/);
+      expect(anonymizer.scrub(once)).toEqual(once);
+    }
+  });
+
+  it('scrub leaves ids, dates, times, versions and amounts alone', () => {
+    const value = {
+      id: '01M5NNM83TV1',
+      ts: '17297001234567890',
+      order: 'A-2007891234567',
+      date: '2026-10-24',
+      time: '05:30 Uhr',
+      version: 'v0.123 4567',
+      postal: '01067 Dresden',
+      amount: '0,50 EUR',
+      url: 'https://example.com/030-1234567',
+      iso: '2026-10-24T05:30:00.000Z',
+    };
+    expect(anonymizer.scrub(value)).toEqual(value);
+  });
+
   it('anonymizeUnknown applies field-name heuristics', () => {
     const event = {
       event_type: 'contact.created',
diff --git a/src/pseudonymizer.ts b/src/pseudonymizer.ts
index 8ac3bdd..e744ba0 100644
--- a/src/pseudonymizer.ts
+++ b/src/pseudonymizer.ts
@@ -33,7 +33,14 @@ export type Pseudonymizer = {
 
 const EMAIL_PATTERN = /[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}/g;
 const IBAN_PATTERN = /\b[A-Z]{2}\d{2}[ ]?(?:[A-Z0-9][ ]?){11,30}\b/g;
-const PHONE_PATTERN = /(?:\+|00)[1-9][\d\s\-/()]{6,}\d/g;
+// International: "+49 170 1234567", "0049 170 1234567". The "00" form only at a token start,
+// so digit runs inside ids, timestamps or amounts ("1729700123456") are left alone.
+const PHONE_PATTERN = /(?:\+|(?<![\w.])00)[1-9][\d\s\-/()]{6,}\d/g;
+// National (German/Austrian) format: trunk prefix 0 + area code, then a separator:
+// "030 1234567", "0221/123456", "0170-1234567", "(0221) 123 456". The separator keeps bare
+// digit runs (ids, postal codes) out; "0[1-9]" keeps the "+00 000 …" pseudonyms from re-matching.
+const NATIONAL_PHONE_PATTERN =
+  /(?<![\w.:/-])(?:0[1-9]\d{1,4}[\s/-]|\(0[1-9]\d{1,4}\)\s?)\d{3,}(?:[\s-]\d+)*(?![\w:/-]|\.\d)/g;
 
 export const createPseudonymizer = (params: {
   orgId: string;
@@ -106,6 +113,7 @@ const makeScrubText =
     value
       .replace(EMAIL_PATTERN, (match) => (match.endsWith(`@${PSEUDONYM_EMAIL_DOMAIN}`) ? match : email(match)))
       .replace(IBAN_PATTERN, iban)
-      .replace(PHONE_PATTERN, phone);
+      .replace(PHONE_PATTERN, phone)
+      .replace(NATIONAL_PHONE_PATTERN, phone);
 
 const normalize = (value: string) => value.trim().toLowerCase();