If you practised with ChatGPT, saw scores in the 80s and then got "nicht bestanden", you are not alone, and it does not mean your German is worse than you thought. Usually the practice score measured something different from the exam. This guide explains the five most common reasons and what to do for the retake.
First: which module did you actually fail?
Before you blame the AI, read your Zeugnis. At B1, B2 and C1 every module (Lesen, Hören, Schreiben, Sprechen) is scored separately out of 100 and passes at 60. If Schreiben passed and Hören failed, your AI writing practice was not the problem. The Zeugnis decoder explains every line.
If Schreiben is the module below 60, read on.
Reason 1: the AI was too generous with the Leitpunkte
The four Leitpunkte (content points) of the task are the most expensive part of Goethe writing. Examiners check how precisely each one is handled, and a missing point costs task points that grammar cannot make up.
On 6 October 2026 we pasted the same B2 forum post into five fresh ChatGPT chats. Leitpunkt 4 ("name alternatives") was missing on purpose. Every time, ChatGPT gave 82–89 points and wrote that all four Leitpunkte were covered, and a few lines later admitted that the alternatives were missing. The task score stayed at 22–23 out of 25 as if a quarter of the task were a detail. The full test with the text, the prompt and every result is in ChatGPT vs GoetheCoach.
This is exactly where real candidates lose points. In our practice data (312 graded texts, not official Goethe results) nearly 1 in 3 texts skipped at least one Leitpunkt, and at B2 the 4th point was missing in 46% of texts.
Reason 2: the AI guessed your word count
A language model reads text in tokens, not words, so unless it runs code it estimates the length. In our test ChatGPT guessed 190, 155 and 170 words for a text of 173. At B2 the minimum is 150. "About 190" tells you that you are safe; the real text may be right at the edge. Count your words yourself.
Reason 3: the same text got a different verdict each time
Ask the same question twice and an AI grader may find different errors. In our test one grammar error was flagged in three runs, called "style" in one and missed in one. One run gave 82 points and then added that a stricter examiner would give 75–80. If the number changes when you press regenerate, it is a feeling, not a measurement.
Reason 4: your practice texts were not written under exam conditions
This one has nothing to do with the AI's grading. In the exam you write against the clock, without a dictionary, without autocorrect and without anyone suggesting a better sentence. Many practice texts are written slowly, polished with AI help and then graded. That measures what you can produce with help, not alone. On the digital exam, typing speed adds another limit; what changed on test day in 2026 has the details.
Reason 5: format and register slipped
Examiners also check the text type and the register: a formal message needs the formal salutation, Sie and a fitting closing; a forum post needs a clear position. AI graders often accept a text that is "good German" in the wrong form. If your Zeugnis is close to 60, this kind of point is often what was missing.
What to do before the retake
- Retake only the module you failed. The Goethe-Zertifikat is modular from B1, so passed modules stay. The recovery roadmap after a failed exam walks through retake or appeal.
- Write the next practice texts like the exam: a real task from a Modellsatz, the official time, no dictionary, no AI while writing.
- Check every Leitpunkt with evidence. For each point, find the sentence that answers it. If you can't point to one, it's missing. If you use ChatGPT, ask exactly that: "Quote the sentence that answers each Leitpunkt, or write MISSING."
- Count your words with a word counter, not by asking the AI.
- Use a grader that scores by a fixed rubric. The free GoetheCoach Writing Check checks each Leitpunkt, counts words in code and rates B1 and B2 texts against the 60% pass mark. It is a practice estimate too; only Goethe examiners grade the real exam.
AI is still a very good study partner for grammar questions, vocabulary and speaking practice. It is the examiner role it is weak at. How to practise Goethe writing with AI shows how to use it for what it is good at.
Frequently asked questions
Can ChatGPT grade the Goethe writing exam accurately?
Not reliably. In our test of one B2 text graded five times, ChatGPT gave 82–89 points, said all Leitpunkte were covered although one was missing, and guessed the word count. Use it to explain errors, not to predict your score.
Why did I fail Schreiben when AI said 80+?
The most common reasons are a missing or thinly answered Leitpunkt, a text that was too short, the wrong format or register, and practice texts written without exam conditions. Check your Zeugnis to confirm Schreiben was the failed module.
Do I have to retake the whole Goethe exam?
No. At B1, B2 and C1 the exam is modular: you retake only the module you failed.
How should I use AI to prepare for the retake?
Write first, alone and under time. Then ask the AI to quote the sentence that answers each Leitpunkt and to explain your errors. Count words yourself, and get the score from a grader that uses a fixed rubric.
Is the GoetheCoach Writing Check an official score?
No. It is a practice estimate based on the four official criteria and published model exams. Only Goethe-Institut examiners grade the real exam.
Sources
- ChatGPT vs GoetheCoach: the same B2 text graded 5 times (test of 6 October 2026)
- Why B2 and C1 candidates lose points on Leitpunkte (312 practice texts)
- How Goethe writing is scored
Would your text really pass?
The Writing Check checks every Leitpunkt, counts your words and rates by the four Goethe criteria. Free, no sign-up.
Start the Writing Check