ABI LF - AI-generated Abitur answer auditieren
ABI LF - AI-generated Abitur answer auditieren
Einleitung
Mini-description: Audit an AI-generated Abitur answer for task fulfilment, evidence, analysis, argumentation, language, hallucinations and exam suitability. Then develop your own, stronger answer. The aim is not to let AI write an exam essay for you, but to sharpen your judgement, source discipline and independent writing.
Deutsch als letzte Hilfe: Du untersuchst eine absichtlich fehlerhafte KI-Antwort, belegst Dein Urteil am Material und schreibst anschließend selbst eine bessere Fassung.

Rule of the course: Context first → word family → visual clue → English explanation → monolingual dictionary → bilingual dictionary → German help last.
All practice data, names and institutions in the audit case are fictional. Recordings are optional and should remain private unless your teacher explicitly provides a protected school platform. Do not upload copyrighted exam texts, personal data or classmates' work to external AI systems. No external AI access is required for this course.
I-can goals
By the end of this aiMOOC, you can …
- Task fulfilment: identify the operator, focus and expected depth of an Abitur task.
- Evidence: distinguish accurate textual evidence from invented or distorted support.
- Analysis: explain how language, structure and rhetorical choices contribute to meaning and effect.
- Argumentation: judge whether claims, evidence and reasoning form a coherent line of argument.
- Language: improve precision, register, cohesion, grammatical accuracy and lexical range.
- Hallucinations: detect fabricated quotations, sources, facts and unjustified certainty.
- Exam suitability: decide whether an answer fits the operator, timing, permitted aids and examination format.
- Independent writing: produce your own improved response without turning an AI output into a replacement essay.
Abi 2027 check: what matters in Baden-Württemberg
English first: For the 2027 English Leistungfach Abitur, work from the official requirements, not from rumours or generic online advice. The official 2027 documents refer to the Bildungsplan 2016 for this cohort. In upper-secondary written assessment, all three Anforderungsbereiche matter, with AFB II and III carrying greater weight than AFB I.
| Area | Abitur 2027 orientation | What it means for your audit |
|---|---|---|
| Level | Leistungfach / erhöhtes Anforderungsniveau | Expect broader content, greater analytical depth, more complex material and more independent connections. |
| Written examination | About 255 minutes in total: Listening, a break, then Writing | Audit both quality and time feasibility. A beautiful answer that ignores the operator is not exam-suitable. |
| Writing | Choice between literary and non-literary material; tasks include focused summary, analysis and further writing | Separate summary from analysis. Do not confuse description with interpretation. |
| Analysis | Evidence-based treatment of content, intention, structure and relevant linguistic or formal features | Ask what a feature does, why it matters and how the evidence supports the interpretation. |
| Permitted aids | Monolingual and bilingual dictionaries plus an approved German spelling reference according to the 2027 rules | Build dictionary discipline instead of depending on an external AI tool. |
| Assessment of Writing | Content and language are assessed separately; the 2027 correction rules weight them 40 : 60 | A correct idea still needs precise, coherent and accurate English. |
| Overall modern-language examination | Listening 20 %, Writing 55 %, Kommunikationsprüfung 25 % | Do not treat the communication exam as a spoken essay; it assesses interaction as well as content and language. |
| Kommunikationsprüfung | Individual or tandem format with preparation time; monologic and dialogic parts | Train spontaneous explanation, response and interaction rather than memorised scripts. |
Current cohort note, 03.10.2026: The revised Bildungsplan version introduced from the school year 2026/27 starts with younger cohorts. For Abitur 2027, follow the specific 2027 Facherlass and your school's current instructions.
Official sources:
- Ministerium für Kultus Baden-Württemberg: Termine und Informationen zum Abitur
- Facherlass Abitur 2027, Stand 09.09.2025
- Beurteilungs- und Korrekturrichtlinien Abitur 2027, Stand 07.07.2025
- IQB: Englisch – Operatoren und Abitur-Dokumente
- Bildungspläne Baden-Württemberg
Media entry: Trust, then verify
Datei:Fact-Checking by Jananie Kulandaivelu for the Wikimedia Foundation.webm
Before reading the audit case, answer in English:
- What makes a fluent answer look trustworthy?
- Which parts of an answer can be checked directly against a source?
- Which claims require outside verification?
- Why can confident language hide weak reasoning?

Think–pair–share: Write one sentence beginning with “I would trust an exam answer more if …”. Then challenge your own rule with a counterexample.
Vorwissen: 90-second triage
Classify each item before you evaluate it.
| Item | Ask yourself | Quick label |
|---|---|---|
| A quotation from the source | Can I locate it exactly or paraphrase it accurately? | Source evidence |
| “This proves that all teenagers …” | Does the source justify this level of generalisation? | Inference |
| “A 2025 Oxford study found …” | Is that study actually supplied or verifiable? | External claim |
| “The repetition creates urgency” | Is repetition present, and is the claimed effect explained? | Analytical claim |
Micro-rule: Never reward a sentence simply because it sounds academic. Audit what it does.
Listening / Viewing: hallucinations and fact-checking
While viewing:
- Note two reasons why an AI system may produce plausible but unsupported information.
- Note one strategy for reducing the risk of accepting a hallucination.
- After viewing, distinguish “plausible”, “probable” and “verified” in one sentence each.
Viewing task: Create a three-column note: claim – check – verdict. Do not copy the video transcript. Record only the checking moves you can transfer to an exam-answer audit.
Lateral reading challenge: Explain why staying on one polished-looking page can be less reliable than opening independent sources. Then transfer the idea to an AI-generated answer: which statements require you to leave the answer and verify elsewhere?
Reading: a fictional mini audit case
Practice material only – not an official Abitur task. All names, figures and institutions below are invented for this course.
Source text
Northbridge Sixth Form tested optional AI feedback for one term. In an anonymous survey of 240 students, 64% said the tool helped them notice recurring language errors, while 31% said they sometimes accepted suggestions they did not fully understand. Teachers found that students improved most when they had to explain why they accepted or rejected each suggestion. Head of English Maya Cole argues, “A fluent sentence is not the same as a sound argument.” The school therefore keeps final drafting offline: students may use dictionaries and teacher feedback, but must write the assessed version independently. The pilot report does not claim that AI improves grades. It recommends teaching verification, source checking and revision as separate skills.
Task
Analyse how the writer builds the argument that AI feedback should support, but not replace, independent writing. Support your points with evidence from the text.
AI-generated answer to audit
The text proves that AI should be used in all final exams. The writer starts with a 2025 Oxford study showing that 90% of students get higher grades with AI, which makes the argument scientific. Cole says that “AI always makes students more accurate”, a direct quotation that creates authority. The statistic that 64% improved their grades confirms the claim. The text is neutral because it only gives positive evidence. Repetition of “AI” makes the tone emotional and persuasive. As a result, the reader must agree that banning AI would be unfair. The source also shows that Northbridge is a university in London, so the findings can be generalised to all British schools.
First audit: Do not rewrite yet. Mark each sentence with one or more labels: TF = task fulfilment, EV = evidence, AN = analysis, AR = argumentation, LA = language, HF = hallucination, EX = exam suitability.
Context vocabulary and word formation
Use the context before translating.
| Core word | Word family | English explanation in this course | German help |
|---|---|---|---|
| evidence | evident – evidently – evidential | information used to support or test a claim | Beleg / Nachweis |
| analyse | analysis – analytical – analytically | examine how parts work together and why they matter | analysieren |
| verify | verification – verifiable – unverifiable | check whether something is accurate and supported | überprüfen |
| argue | argument – argumentative – counterargument | develop a reasoned position rather than merely state an opinion | argumentieren |
| rely | reliable – reliability – unreliable | be dependable enough to use as support | sich verlassen auf |
| infer | inference – inferential | reach a conclusion from evidence rather than from direct statement | folgern |
| cohere | coherence – coherent – coherently | connect ideas logically and clearly | zusammenhängen |
| qualify | qualification – qualified | limit a claim so it matches the strength of the evidence | einschränken / präzisieren |
Dictionary routine: Read the full sentence → identify the word family → paraphrase in English → check a monolingual dictionary → use a bilingual dictionary only if needed → return to the sentence and test the meaning.
Visualising analytical depth

A useful analytical chain is:
Claim → Evidence → Analysis → Link
- Claim: What do you argue about the writer's method?
- Evidence: Which exact detail supports your claim?
- Analysis: How does that detail shape meaning, credibility, tone or reader response?
- Link: How does it answer the task?
Weak: “The writer uses a statistic.”
Stronger: “The survey figures introduce measurable evidence, but the contrast between 64% noticing errors and 31% accepting suggestions they did not understand complicates a simple pro-AI message. This supports the writer's argument for guided, critical use rather than replacement of independent judgement.”
This is a micro-example, not a replacement essay.
The seven-lens audit
| Lens | Key questions | Typical red flag |
|---|---|---|
| 1 Task fulfilment | What does the operator require? Is the answer focused on the exact question? | It discusses whether AI is good or bad instead of analysing how the writer builds an argument. |
| 2 Evidence | Is each quotation, statistic or paraphrase actually in the source and represented accurately? | A real-looking number is given a meaning the source never states. |
| 3 Analysis | Does the answer explain how and why a feature matters? | Device spotting: “There is repetition, which is persuasive.” |
| 4 Argumentation | Are claims connected logically? Are limits and counterevidence handled? | Absolute conclusion from limited evidence. |
| 5 Language | Is the English precise, varied, coherent and appropriate in register? | Vague verbs such as “shows” or “proves” where “suggests”, “contrasts” or “qualifies” would be more accurate. |
| 6 Hallucinations | Are facts, quotations, sources, locations or contexts invented? | “Oxford study”, invented quotation, invented institutional status. |
| 7 Exam suitability | Would the answer work under time pressure with permitted aids and independent authorship? | It relies on unverifiable outside claims or ignores the operator. |
Audit workshop: catch the exact failure
| AI sentence | Your audit question | Repair direction |
|---|---|---|
| “The text proves that AI should be used in all final exams.” | Does the source make that claim? | Return to the actual focus: support, verification and independent writing. |
| “A 2025 Oxford study …” | Where is the study supplied? | Delete the fabricated authority claim. |
| “64% improved their grades …” | What does 64% actually refer to? | Correct the evidence: noticing recurring language errors. |
| “The text is neutral because it only gives positive evidence.” | Is the evidence only positive? | Include the 31% risk and the offline final-drafting rule. |
| “The reader must agree …” | Does analysis justify certainty about every reader? | Use qualified language and explain likely effects. |
Worked audit note: The statement about “64% improved their grades” is not a minor wording error. It changes the meaning of the data. The source says 64% reported that the tool helped them notice recurring language errors; it explicitly does not claim improved grades. Therefore the AI answer turns evidence into a hallucinated conclusion.
Speaking: defend your revision
Communication-practice format:
- 60 seconds: explain the most serious flaw you found.
- 60 seconds: your partner challenges your judgement with a different reading.
- 60 seconds: respond, concede where appropriate and defend your final decision with evidence.
- Repeat with a new lens.
Useful language: “The evidence supports … only to the extent that …” / “I would qualify that claim because …” / “The quotation cannot support this interpretation because …” / “A more defensible reading would be …”
Privacy: Recording is optional. If you record yourself, keep the file private unless your teacher has explicitly set up a protected school workflow.
Writing: build your own better answer

Write your own answer to the fictional task. Do not “repair” every sentence of the AI version. Start again from the source.
Suggested process:
- Re-read the operator and underline the analytical focus.
- Select two or three strong pieces of evidence from the source.
- Build a claim–evidence–analysis–link chain for each paragraph.
- Include at least one qualification or contrast.
- Check language, cohesion and evidence accuracy only after the reasoning is complete.
Suggested practice length: 180–250 words, or the length your teacher sets. The aim is controlled analytical quality, not maximum length.
Non-negotiable self-check: zero invented quotations, zero invented studies, zero claims that exceed the evidence.
Optional mediation: explain a school rule
Fictitious German school notice for practice only:
„Bei dieser Übungsleistung schreibst Du die Endfassung selbst. Wörterbücher und die im Unterricht bereitgestellten Materialien sind erlaubt. Externe KI-Dienste dürfen für die Endfassung nicht verwendet werden. Wenn Du vorher mit KI-Beispielen gearbeitet hast, musst Du Angaben und Quellen selbst prüfen. Aufnahmen bleiben freiwillig.“
Mediation task: An English-speaking exchange student has missed the briefing. Explain in clear English what they need to know before completing the practice task. Select and adapt relevant information; do not translate word for word.
Hilfen und Differenzierung
| Level | Language support | Scope and pace | Analytical demand |
|---|---|---|---|
| Support | Sentence starters, vocabulary table, one lens at a time | Audit four sentences first; extend after feedback | Identify one accurate evidence problem and one analytical problem |
| Standard | English glossary plus dictionary routine | Audit the full answer and write a complete response | Connect evidence, function and effect in each paragraph |
| Challenge / Leistungsfach plus | Minimal sentence scaffolding; precise academic vocabulary | Add a timed second material or visual stimulus | Compare alternative interpretations, qualify claims and link material independently |
Support ladder: Context → word formation → visual clue → English explanation → monolingual dictionary → bilingual dictionary → brief German help.
Feedback: evidence before elegance
Use this four-step peer protocol:
- Strength: Identify one sentence where reasoning and evidence work together.
- Evidence risk: Mark one claim that needs checking, narrowing or deletion.
- Analytical depth: Ask one “how?”, “why?” or “so what?” question.
- Language edit: Improve one phrase for precision, cohesion or register.
Do not rewrite your partner's paragraph for them. Feedback should make the writer more independent.
Fehleranalyse: build a repair log
| Tag | Error type | Repair question |
|---|---|---|
| TF | Task focus | Am I answering the operator and the exact focus? |
| EV | Evidence | Is this detail really in the source? |
| AN | Analysis | Have I explained function and effect? |
| AR | Argumentation | Does the reasoning follow logically and acknowledge limits? |
| LA | Language | Is the wording precise, coherent and grammatically controlled? |
| HF | Hallucination | Did I invent or overstate a fact, quotation or source? |
| EX | Exam suitability | Could I produce and verify this independently under exam conditions? |
After receiving feedback, record only recurring error patterns, not a full model answer. Your goal is to make the same mistake less likely next time.
Transfer: from AI audit to source-based Abitur analysis
Use a teacher-provided excerpt, public-domain text or legally supplied classroom material related to an Abitur topic. First write a deliberately weak two-sentence interpretation yourself. Then audit it with the seven lenses and replace it with a stronger paragraph.
Copyright and AI rule: Do not paste complete copyrighted Pflichttexte, exam papers or film sequences into public platforms or external AI tools. Work from material legally provided by your school and follow the current school rules for digital services.
Abi-Check: written exam versus Kommunikationsprüfung
| Check | Schriftliche Prüfung | Kommunikationsprüfung |
|---|---|---|
| Core demand | Listening plus source-based writing | Monologic and dialogic spoken communication |
| Evidence | Precise source reference and accurate textual support | Relevant material use plus responsive spoken reasoning |
| Language | Lexical range, grammar, cohesion and register | Accuracy, fluency, interaction and discourse management |
| Strategy | Plan, analyse, write, revise under time pressure | Prepare key points, speak independently, listen and react |
| What not to do | Import unverifiable claims or memorised generic paragraphs | Deliver a memorised essay without genuine interaction |
| AI audit transfer | Verify every factual and textual claim | Challenge a partner's claim and respond with evidence and qualification |
Exam-day rule: Your school's current instructions and the official Baden-Württemberg documents take priority over this learning page.
Folgekurs
Continue with:
- ABI LF – Quellen lateral prüfen
- ABI LF – Analyse vom Beleg zur Wirkung
- ABI LF – Kommunikationsprüfung argumentieren
- ABI LF – Writing unter Zeitdruck
- ABI LF – Materialverknüpfung im Leistungsfach
Interaktive Aufgaben
Quiz: Teste Dein Wissen
What should you check first when auditing an exam answer? (The operator and exact task focus) (!The number of long words) (!Whether the answer sounds confident) (!Whether the conclusion is positive)
What counts as strong evidence in a source-based analysis? (Accurate support that can be located in the source) (!A plausible statistic remembered from the internet) (!A quotation that sounds appropriate) (!A general claim with no source reference)
Which example is a hallucination? (An invented study presented as if it were supplied by the source) (!A cautious interpretation based on a quotation) (!A paraphrase that preserves the source meaning) (!A clearly labelled personal evaluation)
What makes analysis stronger than device spotting? (It explains how and why a feature contributes to meaning or effect) (!It names as many techniques as possible) (!It replaces evidence with personal opinion) (!It repeats the task in every sentence)
What does the 64 percent figure in the fictional source actually describe? (Students saying AI helped them notice recurring language errors) (!Students receiving higher final grades) (!Teachers preferring AI to dictionaries) (!Students supporting AI in final exams)
Which feature best supports exam suitability? (Accurate evidence, operator fulfilment and work that is feasible under exam conditions) (!Unverifiable outside knowledge) (!A memorised introduction for every topic) (!The longest possible answer)
What is lateral reading? (Checking a claim by consulting independent information beyond the original page) (!Reading only the first paragraph repeatedly) (!Translating every unknown word immediately) (!Accepting a source because its design looks professional)
Which combination best describes strong language performance? (Precision, range, cohesion, appropriate register and grammatical control) (!Rare vocabulary regardless of meaning) (!Very long sentences in every paragraph) (!Informal phrasing to sound spontaneous)
What should an analysis paragraph connect? (A claim, accurate evidence, analysis and a link to the task) (!A quotation, a translation, a personal story and a slogan) (!A summary, a new topic, a guess and a conclusion) (!A statistic, an invented source, a rhetorical question and an apology)
How should you approach the Kommunikationsprüfung? (Prepare key ideas but respond genuinely in monologic and dialogic communication) (!Memorise a complete essay and recite it) (!Avoid reacting to your partner) (!Use external AI to generate live responses)
Memory
| Task fulfilment | Operator and focus |
| Evidence | Verifiable textual support |
| Analysis | How and why |
| Argumentation | Coherent reasoning |
| Language | Accuracy and register |
| Hallucination | Invented unsupported detail |
| Exam suitability | Fit to format time and permitted aids |
Drag and Drop
| Ordne die richtigen Begriffe zu. | Thema |
|---|---|
| Invented quotation | Hallucination |
| Wrong operator focus | Task fulfilment |
| Misrepresented statistic | Evidence |
| Device named without effect | Analysis |
| Unsupported leap in reasoning | Argumentation |
| Inappropriate informal register | Language |
| Depends on unavailable outside tools | Exam suitability |
Kreuzworträtsel
| Operator | Which task word tells you what intellectual action to perform? |
| Evidence | What do you call support that can be checked against the source? |
| Analysis | What explains how and why textual choices matter? |
| Cohesion | What connects sentences and ideas smoothly across a text? |
| Register | What term describes language choice appropriate to situation and audience? |
| Hallucination | What is an invented but confidently presented AI claim called? |
LearningApps
Lückentext
Offene Aufgaben
Leicht
- Audit-Markierung: Markiere im fiktiven KI-Text je ein Beispiel für evidence error, hallucination, weak analysis und overgeneralisation und begründe Deine Zuordnung auf Englisch.
- Evidence Card: Gestalte eine kleine Karte mit den Feldern “claim – source evidence – what it really shows – safe wording” und fülle sie für die 64-Prozent-Angabe aus.
- Word Family Map: Visualisiere die Wortfamilien zu analyse, verify, argue und rely und schreibe zu jeder Form einen eigenen Beispielsatz.
- Private One-Minute Explanation: Erkläre freiwillig in einer privaten 60-Sekunden-Aufnahme, warum ein flüssiger Stil keine verlässliche Analyse garantiert; lösche die Aufnahme anschließend oder behalte sie lokal.
Standard
- Seven-Lens Peer Audit: Prüfe einen selbst geschriebenen Absatz eines Partners mit den sieben Linsen, ohne den Absatz umzuschreiben.
- Lateral Check Protocol: Entwickle für drei externe Behauptungstypen ein Prüfverfahren und dokumentiere, welche unabhängigen Quellenarten Du jeweils suchen würdest.
- Timed Rewrite: Schreibe in begrenzter Zeit einen eigenen 180–250-Wörter-Text zur fiktiven Aufgabe und führe danach eine zweiminütige Evidenzkontrolle durch.
- Communication Drill: Simuliere zu zweit einen monologischen Einstieg und eine dialogische Nachfragephase; bewertet anschließend interaction, evidence und qualification.
Schwer
- Competing Interpretations: Entwickle zwei vertretbare Lesarten derselben Textstelle, vergleiche ihre Belegstärke und entscheide begründet, welche Grenzen jede Lesart hat.
- Cross-Material Audit: Verknüpfe einen Lehrertext mit einer Grafik oder Statistik und prüfe, ob Deine Schlussfolgerungen wirklich von beiden Materialien getragen werden.
- Rubric Stress Test: Entwirf eine kompakte Bewertungsmatrix für task fulfilment, evidence, analysis, argumentation, language, hallucinations und exam suitability und teste sie an zwei unterschiedlich schwachen Beispielabsätzen.
- Create a Safe Audit Case: Erstelle selbst einen kurzen fiktiven Ausgangstext und eine absichtlich fehlerhafte Antwort mit mindestens fünf unterschiedlichen Fehlertypen; tausche nur innerhalb des Kurses und ohne personenbezogene Daten.


Lernkontrolle
- Audit begründen: Erhalte einen neuen fiktiven KI-Absatz und priorisiere die drei schwerwiegendsten Probleme. Begründe, warum sie stärker wiegen als kleinere Sprachfehler.
- Evidenz transformieren: Formuliere aus einer korrekten Textbeobachtung zunächst eine zu starke Behauptung und anschließend eine angemessen qualifizierte analytische Aussage. Erkläre den Unterschied.
- Materialverknüpfung prüfen: Vergleiche Text und Grafik und identifiziere eine Schlussfolgerung, die nur von einem Material gestützt wird, sowie eine, die durch beide gestützt wird.
- Operatorentransfer: Zeige an demselben Material, wie sich eine Antwort auf “summarise”, “analyse” und “assess” in Ziel, Struktur und Denkleistung unterscheiden müsste.
- Prüfungsformat wechseln: Überführe eine schriftliche Analyseidee in einen 90-sekündigen mündlichen Beitrag und ergänze zwei mögliche dialogische Reaktionen auf Rückfragen.
- Fehlerprofil nutzen: Werte Dein persönliches Repair Log aus und entwickle drei konkrete Prüfhandlungen, die Deine häufigsten Fehler unter Zeitdruck verhindern sollen.
Lernnachweis
Für einen Lernnachweis zu diesem Thema ist wichtig:
- Du kannst den Operator und den Fokus einer anspruchsvollen Englisch-Aufgabe sicher bestimmen.
- Du kannst Belege exakt vom Ausgangsmaterial trennen und erfundene oder verzerrte Angaben erkennen.
- Du kannst Analyse als funktionale Erklärung von sprachlichen, strukturellen und argumentativen Entscheidungen durchführen.
- Du kannst eine kohärente, differenzierte Argumentation mit angemessener Qualifizierung entwickeln.
- Du kannst sprachliche Qualität nach Präzision, Bandbreite, Kohäsion, Register und grammatischer Kontrolle reflektieren.
- Du kannst Halluzinationen und nicht verifizierbare Autoritätsbehauptungen systematisch prüfen.
- Du kannst schriftliche Prüfung und Kommunikationsprüfung hinsichtlich Anforderungen und Strategien unterscheiden.
- Du kannst eine eigenständige verbesserte Antwort verfassen, ohne einen KI-Text als Ersatzaufsatz zu übernehmen.
- Du kannst Deine wichtigsten Fehler aus Feedback ableiten und in einem Repair Log in konkrete nächste Schritte übersetzen.
OERs zum Thema

Additional open and educational reference:
- Wikimedia Commons – freie Medien
- Stanford / Digital Inquiry Group – Civic Online Reasoning resources
- Stanford CRAFT – classroom resources for evaluating AI information
The embedded YouTube videos are linked for educational viewing and are not reproduced as course files. Wikimedia Commons media remain subject to the licence shown on their respective file pages.
Verknüpfte Lernbereiche
aiMOOC-Projekte
MOOCwiki · Deutsch
Nach dem Lernen ist vor dem Lernen
Entdecke direkt den nächsten Lernkurs. Weitere Inhalte erscheinen, wenn Du weiter nach unten scrollst.
Zur MOOCwiki-HauptseiteJetzt üben: AI competence →Mediathek
Mediathek wird aus dem Wiki geladen ...
Keine passenden Inhalte gefunden. Bitte ändere Suche oder Filter.
NEWSLernweltNOAH fragen