When you submit an essay on LinguaProva, our AI grader evaluates it against the same four criteria a human telc grader would use. This article walks through each criterion, what graders count as an A vs a B vs a C, and the patterns that most often pull scores down.
We mirror the official telc rubric. Each criterion is rated A, B, C or D against the C1 target, and the four scores add up to 48 points — the same maximum the official exam uses.
The four criteria, in plain terms
| Criterion | In one line |
|---|---|
| Aufgabengerechtheit | Did you actually answer what the prompt asked? |
| Korrektheit | Are your grammar and spelling accurate at C1 level? |
| Repertoire | Do you use a rich, varied range of words and structures? |
| Kommunikative Gestaltung | Is your text well organised and easy to follow? |
1. Aufgabengerechtheit (task fulfilment)
This is about covering the prompt, not about how beautifully you write. If the task asks you to weigh pros and cons and conclude with your position, the grader checks that all three parts are present and that there is a clear "roter Faden" — one continuous thread of argument from intro to conclusion.
Scores at a glance:
- A — every point in the prompt is addressed in depth, position is defended with concrete arguments.
- B — most points covered, audience/situation respected, but one aspect is underdeveloped.
- C — only partially on-topic, or the position is vague, or you addressed the wrong text type.
- D — the text barely engages with the actual prompt.
2. Korrektheit (accuracy)
At C1 the standard is "very few or no mistakes in simple structures, occasional mistakes in complex ones". Conjugation, case, word order, agreement, spelling — all of it counts. Mistakes in complex subjunctive forms or relative clauses cost less than mistakes in basic verb conjugation.
Scores at a glance:
- A — errors are rare and only in advanced structures; they never block understanding.
- B — errors mostly in complex structures, never disrupting the reader.
- C — multiple errors even in simple structures, or comprehension is occasionally affected.
- D — frequent errors, parts of the text are hard to understand on first read.
3. Repertoire (range)
How wide is your vocabulary and how varied is your sentence structure? Using obwohl, sofern, demzufolge, infolgedessen where they fit signals C1 range; sticking to aber, weil, dann, also throughout signals B1/B2 even if everything is correct.
4. Kommunikative Gestaltung (cohesion & form)
Paragraphs, connectors between ideas, register appropriate to the audience, and a clear progression: introduction → argument → conclusion. Even a grammatically perfect essay can score low here if it is one giant block of text or if ideas jump around without transitions.
Patterns that cost the most points
AI grader vs human grader: how similar are they?
Our AI grader is calibrated against publicly available telc sample essays and the official descriptors from the telc Hochschule handbook. In practice:
- The AI is consistent — the same essay always gets the same score. Human graders vary by ±1 grade per criterion.
- The AI is strict on Repertoire — it notices when you reuse the same construction across paragraphs more easily than a human reading at speed.
- The AI is more lenient on idiomatic feel — a human native speaker may flag a phrase as "off" without being able to name a rule violated. Our grader rarely penalises these.
Net effect: if the AI gives you a 36/48, expect a human grader on exam day to give you somewhere between 32 and 40. Use the AI feedback to identify weak criteria, not to predict an exact score.
How to use the feedback
- Read the criterion-by-criterion breakdown first. The total is less useful than knowing which of the four is dragging you down.
- Re-write the same essay focusing on the weakest criterion. If Repertoire was a C, force yourself to replace 10 ordinary words with C1-level alternatives. If Kommunikative Gestaltung was a C, restructure into 4 paragraphs with explicit transitions.
- Move to a different prompt after improvement. Iterating on the same topic without changing the prompt makes you over-specialised on one argumentative angle.
Practise your essay
We have over 80 argumentative prompts modelled on real telc Übungstest topics, each with a 70-minute timer and the same word counter the AI uses. Start a session on the pricing page — the free trial includes essay grading.