1. What is generated by AI

Content generated on the platform is labelled in the interface so that teachers and students can tell it apart from human-authored material, and synthetic audio is marked as machine-generated.

2. Where it goes wrong

These are general-purpose language models, not a graded-materials engine. In practice expect:

• Factual errors. Models state confident, plausible things that are untrue. Anything factual in a lesson needs checking.
• CEFR drift. Requested level and delivered level do not always match; vocabulary above the target band is common.
• Flawed answer keys. A question can have more than one defensible answer, or a key can simply be wrong.
• Pronunciation and accent artefacts. Synthetic speech mishandles names, acronyms and loanwords, which matters more in a listening exercise than elsewhere.
• Transcription errors. Whisper drops or invents words in noisy audio and with strong accents.
• Cultural bias. Training data skews towards some varieties of English and some cultural framings.

3. Human oversight

A teacher stands between every generation and every student. Lessons are created, reviewed, edited and assigned by a person; nothing reaches a class automatically. You can edit or delete any generated text, regenerate audio, and discard a lesson entirely. That review step is the oversight mechanism this product relies on, and it is why the disclaimer in section 2 is not merely a formality.

4. What we do not do

The platform is not a high-risk AI system under Annex III of the AI Act, because it does not do any of the following, and we will update this page before that changes:

• score or grade students in a way that affects their marks or progression;
• decide admission to, or assignment within, an educational institution;
• monitor or detect prohibited behaviour during tests;
• evaluate learning outcomes in a way that determines a student’s education;
• infer emotion from a student’s voice, face or behaviour;
• profile students or make automated decisions with legal or similar effects.

Live-session answers are shown to the teacher as classroom feedback while the session runs. The platform does not convert them into a grade, and the teacher decides what, if anything, they mean.

5. Your content and model training

We do not train AI models on your lesson content, and we do not let our providers do so: Gemini and Whisper are used through paid API tiers whose terms exclude submitted content from training. What each provider receives is set out on the sub-processors page.

Do not put personal data about identifiable students into a prompt — see section 6 of the Privacy Policy.

6. Teaching with AI-generated material

Practical advice

• Listen to the audio once before class. It takes a minute and catches most problems.
• Check the answer key against the transcript, especially for inference questions.
• Tell students the audio is synthetic. It is honest, and it is usually an interesting discussion in itself.
• Treat factual content as a prompt for discussion rather than as a source, unless you have verified it.

7. Questions and reporting problems

If a generation produces something harmful, biased or plainly wrong, send it to support@listeningnokidding.com. Reports genuinely shape the prompts and the guard rails, and we would rather hear about it than not.

We will revise this page as the AI Act’s obligations phase in and as the models behind the product change.