All guides
AI writing feedbackTELC B1German writing

AI vs Human Grading on a German B1 Essay: An Honest Comparison

We took the same TELC B1 writing sample and evaluated it with AI feedback and with TELC rubric criteria. Here's what each approach catches — and misses.

1 April 20266 min read

AI writing feedback for German exam preparation is increasingly common. But how useful is it compared to what a trained human examiner would mark? The honest answer is more nuanced than most people expect — AI and human grading are genuinely good at different things.

This post breaks down what each approach catches well, where each one falls short, and how to use them together effectively.


How TELC B1 Writing Is Actually Marked

First, the context: TELC B1 writing is assessed by a trained human examiner using three criteria:

CriterionWhat it assessesMax points
KommunikationDid you address all 4 required points? Are ideas clear?15
Formale RichtigkeitGrammar, vocabulary, spelling accuracy15
KohärenzLogical structure, flow, use of connectors15

The pass mark is 27/45 (60%). A candidate can pass with imperfect grammar if their communication and coherence are solid. They can fail with perfect grammar if they miss required points.

Any useful feedback tool — human or AI — needs to assess all three criteria separately.


What AI Writing Feedback Does Well

Identifying systematic grammar errors

AI evaluators are highly effective at finding patterns of grammatical error across a piece of writing. Common B1-level errors that AI catches reliably:

  • Case errors: wrong article form (einen vs. einem, der vs. dem), especially in dative constructions
  • Verb position: subordinating conjunctions (weil, dass, obwohl, wenn) that should send the verb to the end
  • Preposition collocations: interessiert an vs. interessiert für, warten auf vs. warten für
  • Konjunktiv II: inconsistent or incorrect use of würde + infinitive vs. indicative forms
  • Plural forms: common irregular plurals that learners misremember

These errors appear in predictable patterns. An AI evaluator doesn't tire, doesn't skim, and finds every instance.

Flagging vocabulary repetition

AI reliably identifies when the same word or phrase appears too many times — for example, using wichtig five times in a 160-word letter, or repeating ich finde as the opener for every sentence. It typically suggests synonyms with context for when they're appropriate.

This is useful because vocabulary variety is part of the Formale Richtigkeit criterion, and most learners don't notice their own repetition patterns.

Speed and availability

This is not a trivial point. A trained tutor charges €15–€30 per written piece and responds within 24–72 hours. AI feedback is instant and available at 11pm before an exam the next day. For candidates writing 10–20 practice letters over a preparation period, the difference in availability and cost is significant.


What Human Examiners Do Better

Register assessment

This is where human examiners have the clearest edge. Register in German — the level of formality — is subtle and context-dependent in ways that are hard to codify.

A learner might write formally correct German that slides slightly toward casual mid-letter, or use phrasing that is technically grammatical but sounds overly stiff for the prompt's context (sehr geehrte Damen und Herren to address what the prompt frames as a community group). Human examiners with native-level German instinct catch these shifts. AI tools often rate register as "appropriate" when it has drifted.

At B1 level, register problems are typically a Kohärenz rather than a Formale Richtigkeit issue — they affect the overall coherence and appropriateness of the text.

Task completion: the detail level

The TELC B1 Schreiben task gives four specific bullet points to address. "Did you address all four?" sounds like something AI can check — and it can, for clear omissions. But the devil is in the detail.

A learner might write: "I enjoy sports." The prompt's third bullet point was: "mention your hobbies and ask a question about the organisation's activities." The learner addressed the hobby part but completely skipped the question. AI evaluation often misses partial point coverage — it sees that hobbies were mentioned and marks that bullet as addressed. A human examiner reads the rubric strictly and sees the missing question.

Partial point coverage is one of the most common ways candidates lose 5 marks on Kommunikation. Human examiners catch it more reliably.

Coherence at the paragraph level

AI evaluation of Kohärenz is good at the sentence level — it can check for connecting words and logical sequencing. But multi-paragraph coherence — whether two paragraphs make the same argument from different angles without connecting them, or whether the conclusion feels disconnected from the body — is harder for AI to assess reliably.

Human examiners read for overall argument flow in a way that current AI evaluation doesn't fully replicate.


A Practical Comparison Table

What's being assessedAI reliabilityHuman reliability
Grammar errors (case, verb position)HighHigh
Vocabulary repetitionHighMedium (depends on attention)
Preposition collocationsHighHigh
Register consistencyMediumHigh
Task completion (all 4 points)MediumHigh
Partial point coverageLow–MediumHigh
Multi-paragraph coherenceMediumHigh
SpeedNear-instant24–72 hours
Cost per essayLow€15–€30
ConsistencyVery highVariable

The Right Way to Use Both

AI feedback and human feedback are not competing for the same job. The most effective approach:

Use AI feedback for most of your practice. Write a letter, get AI feedback, fix the specific errors identified, write the next one. Doing this across 10–15 practice letters will systematically reduce your grammar error rate and vocabulary repetition. AI is better than no feedback for all of this.

Use human feedback strategically. With 2–4 weeks to go before the exam, get 2–3 letters reviewed by a trained examiner or language tutor who knows the TELC criteria. The goal here is register calibration and strict task completion checking — the two things where human examiners have the clearest advantage.

The combination works: AI gets you to 80–90% of your potential improvement cheaply and quickly; human feedback handles the marginal 10–20% that requires genuine reading of the text.


What "AI feedback aligned with TELC criteria" actually means

There's a meaningful difference between asking a general AI chatbot "is this good German?" and using a tool built specifically for TELC B1 evaluation. A generic AI response often:

  • Provides encouragement rather than a score
  • Doesn't break down Kommunikation, Formale Richtigkeit, and Kohärenz separately
  • Isn't calibrated to what a 15/15 vs 10/15 vs 5/15 looks like on each criterion

A purpose-built feedback tool should score each criterion separately, explain what specifically is wrong, and indicate how far above or below the pass threshold the writing sits. That's the information you actually need to improve.

Try AI writing feedback on the TELC B1 mock exam →

Ready to practise?

Take a free full-length TELC mock exam.