Sources
Attribution & Licences
Where the words, texts and questions on this site come from, and which parts were written by AI. The vocabulary, the reading texts and the questions are published under CC BY-SA 4.0, inherited from Wiktionary and the wikis the reading texts come from. One of those, Hurraki, is CC BY-SA 3.0 DE, so each reading text records its own licence alongside its source. If you reuse any of it, keep this attribution reachable.
What was written by AI
- The A1 reading texts. The 46 A1 texts were written by an AI assistant from the A1 word list, and so were their translations, glossaries, grammar notes and comprehension questions.
- Translations and glossaries from A2 to C2. The English translations, glossaries and grammar notes were written by AI. The German texts themselves are real articles from Klexikon, MiniKlexikon and Hurraki.
- Practice questions. The reading, true or false, Sprachbausteine and vocabulary questions, and their explanations, were written by AI. The article, conjugation and word order questions were not. They are built by rule from the Wiktionary tables and real example sentences.
- Grammar test explanations. The explanations on the A1 grammar gap-fill questions were written by AI and proofread by a second AI pass. Claude Opus 5.5 read every one and corrected the ones that were wrong.
- Word meanings. The short English meaning shown for each word was written by AI, shortened from the full Wiktionary definitions.
- Case and preposition. The case and preposition a word takes, shown for 1,943 words, was proposed by AI and checked against the German sentences on this site. For rare words those sentences often say nothing, and then the pattern rests on the AI alone.
- Grammar points. Every grammar point from B1 to C2 was written by Claude Opus 5.5, made by Anthropic. The A1 and A2 points were written by AI, and Claude Opus 5.5 rewrote the ones it got wrong. Most example sentences on these pages are real Tatoeba sentences; a few were written by AI.
- Example sentence audio. The recordings of example sentences are synthetic speech from Google Cloud Text-to-Speech (WaveNet). Word pronunciations are different: they are recordings from Wikimedia Commons.
- Listening tests. The 54 A1 to C2 tests and their 1,320 questions were written with OpenAI Codex and GLM 5.3 Flash, and the C1 and C2 tests with Claude Sonnet 5.5. Their original dialogues, announcements, interviews and talks follow Goethe task formats, using text choices in place of pictures. The recordings use Rime Coda's German Aura, Baldur, Kumara and Lina voices. The material has not been reviewed by a teacher.
- Authored reading tests. The 48 original A1 to B2 reading tests and their 1,140 questions were written with Claude Opus 5.5, OpenAI Codex and GLM 5.3 Flash. They follow Goethe task formats and appear in reading practice and the four-module mocks. No passage or question is taken from an exam paper. The new GLM reading and listening tasks passed blind AI answer checks through Hugging Face. The material has not been reviewed by a teacher.
- Writing tasks. The writing tasks from A1 to C2 were written with Claude Opus 5.5, Claude Sonnet 5.5, OpenAI Codex and GLM 5.3 Flash in the formats of the Goethe-Zertifikat. None is taken from an exam paper. Your answer is marked by GLM 5.3 Flash (Z.ai) through OpenRouter: the band for each criterion, the corrections and the rewrite. The A1 forms are checked against the task's own answers, with no AI. At C2 the rewriting of Teil 1 is marked by the model against a model solution, with the given word checked by rule.
- Speaking exams. The speaking exams from A1 to C2, their cards and the presentations Lena or Jonas give, were written with Claude Opus 5.5, Claude Sonnet 5.5, OpenAI Codex and GLM 5.3 Flash in the formats of the Goethe-Zertifikat. None is taken from an exam paper. In the exam, the examiner and the partner are played by GLM 5.3 Flash (Z.ai) through OpenRouter, apart from the examiner's opening and closing lines, which follow the Goethe-Institut's procedure. They are spoken by Rime's Coda voice model, and what you say is turned into text by Whisper. Your exam is marked by GLM 5.3 Flash. The C2 exam is taken alone with the examiner, as in the real exam.
- The Daily Voice Lesson and Ask the tutor. Lena and Jonas, their replies, corrections, quizzes and readings, and the tutor's answers about a test question, are generated live by GLM 5.3 Flash (Z.ai) through OpenRouter, with Hugging Face as the fallback. What you say into the microphone is turned into text by Whisper, and the lesson replies are spoken by Rime's Coda voice model.
AI-written material built on the sources below carries their licence, CC BY-SA 4.0.
What was not written by AI
The graded German reading texts from A2 to C2, the example sentences on word entries, the conjugation and declension tables, the learner mistakes, the word list and its levels, and the word pronunciations all come from the sources below.
Where the data comes from
| Source | Provides | Licence |
|---|---|---|
| minddory.com | The starting list of words | None stated |
| Goethe-Institut Wortlisten, Goethe-Zertifikat A1, A2 and B1 | The level of every word they list. Words they do not list are graded by how common they are | Used for levels only, not republished |
| Goethe-Institut Modellsätze, Goethe-Zertifikat A1 (Start Deutsch 1), A2, B1 and B2 | The formats, times and marking criteria for writing and speaking, and the parts, question types and repeat counts for listening | Used for formats only, not republished |
| Wiktionary via wiktextract, and German Wiktionary for compounds | Gender, article, plural, genitive, verb forms, auxiliary, comparative and superlative, the full English definitions, IPA, audio links, topic labels, synonyms, antonyms, and every conjugation and declension table | CC BY-SA 4.0 |
| Tatoeba | Example sentences for words and grammar points, and the corpus counts used to pick each word's primary reading | CC BY 2.0 FR |
| OpenSubtitles v2024 via OPUS | Most example sentences, mined from 65.7 million German–English pairs | See note below |
| FrequencyWords (OpenSubtitles 2018) | How common each word is in spoken German, which every list on this site is sorted by | CC BY-SA 4.0 |
| Leipzig Corpora Collection, German news 2015 | How common each word is in written German, and the spoken and written labels | CC BY |
| MERLIN learner corpus | The learner mistakes and their corrections on the grammar pages | CC BY-SA 4.0 |
| Klexikon | Most reading texts, and the passages the comprehension questions are built on | CC BY-SA 4.0 |
| MiniKlexikon | Reading texts in Leichte Sprache, mostly at A2 | CC BY-SA 4.0 |
| Hurraki | Reading texts in Leichte Sprache, mostly at A2 | CC BY-SA 3.0 DE |
| German Fluent | The 46 A1 reading texts and their study material, written by AI for this site | CC BY-SA 4.0 |
OpenSubtitles asks that work using the corpus link to opensubtitles.org and cite Lison & Tiedemann (2016), OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles (LREC). OPUS states that it does not own the underlying text and redistributes only what it believes it may.
Pronunciation audio is linked to Wikimedia Commons, not copied. Individual recordings carry their own licences, so check the Commons file page before redistributing a file itself.
The typeface is Plus Jakarta Sans by Tokotype, under the SIL Open Font License 1.1.
Known limitations
- The source word list was fully lowercased; capitalisation is restored only for words matched in Wiktionary.
- Homographs differing only by case (essen / Essen) are collapsed in the source; the alternative reading is kept alongside the entry.
- Goethe publishes word lists only up to B1. B2, C1 and C2 are cut by how common a word is, so those boundaries are a convention, not an official list.
- Reading texts are selected against a hard floor: no passage is published where a learner at that level would know less than 90% of the running words. A2 is the level where that bites: its texts are short, and they are drawn from Leichte-Sprache sources rather than from the children's encyclopaedia, because no ordinary German text clears the bar against a beginner's vocabulary. At A1 nothing cleared it at all, which is why those texts were written by AI instead.
Found a problem?
A wrong translation, a mislevelled word, a broken practice question, anything that looks off: hello@germanfluent.com. Include the page it is on and it gets fixed in the next build.
Immersion videos
The videos on Immersion are embedded from YouTube. German captions and any channel-provided English belong to the credited channel. Missing English lines and additional word meanings are machine-generated and labelled on each video page. Video levels are editorial estimates. The videos and their captions are not covered by this site's CC BY-SA licence.
Reusing this material
CC BY-SA 4.0 lets you copy, adapt and redistribute all of it, including commercially, on two conditions: credit the sources above, and license what you build under the same terms. A link back to this page satisfies the attribution requirement.