How Studio uses AI

Studio uses large language models — the same family of technology behind general-purpose AI assistants — to turn material you already have into teaching resources. That means:

• writing worksheets, quizzes, lesson plans, exit tickets, flashcards, task cards and vocabulary sets from a topic you type or a source you upload;
• summarising a PDF, a recording or a video into notes;
• turning audio and video into text;
• answering questions you ask in chat about material you have already added.

What it does not do is just as important. Studio does not mark student work, does not score or rank students, does not build a profile of any student, and makes no decision about anyone’s education. It writes material for a teacher to use, and stops there.

Generating a resource means sending your material to an AI provider and getting the result back. There is no way to do it without that, so here is exactly what goes where.

Where the AI runs, and where your work sits. The AI runs in the United States. We reach the models through OpenRouter, Inc., a United States company, and it may route our requests only to a named list of model hosts that we enforce on every single request: OpenAI, Microsoft Azure, Amazon Bedrock, Baseten, Cloudflare, CoreWeave, DeepInfra, Fireworks and Together, all of them headquartered in the United States. Turning a recording into text uses OpenAI’s Whisper model run at DeepInfra, in the United States.

Teachit Studio itself does not run there. The application, its database and its request logs run in Railway’s EU West region, in Amsterdam, and the files you upload and the documents we generate are stored in Backblaze B2’s EU region. The Privacy Policy names each provider, says what it does for us and where it operates, and sets out the safeguards we rely on when personal data leaves the UK.

• We do not train AI models on your material, and we do not permit our providers to. Your uploads are not used to make anyone’s model better.
• We do not sell your material, and we do not build datasets from it.
• Studio carries no advertising and shares nothing about you for advertising purposes.
• We do not use your material to answer anyone else’s request. What you upload is visible to you and to no other user.

This is the section we would most like you to read. AI writes fluently whether or not it is correct, which makes its mistakes harder to spot than a human’s. In Studio specifically, expect that it can:

• state a fact confidently and be wrong — dates, figures, attributions and scientific detail are the usual culprits;
• produce an answer key that does not match its own question, or mark a plausible distractor as correct;
• miss a reading level — text asked for at Year 4 can come back at Year 7;
• cite a book, article or study that does not exist, or attribute a real quotation to the wrong person;
• be out of date on recent events, because a model’s knowledge stops at a fixed point;
• reproduce bias present in the material it learned from, in whose examples it chooses and whose it does not.

We test for these and design against them. Quiz and worksheet answer keys go through a second automated pass that re-checks the key against the question and corrects it where it disagrees, and output that does not match the shape we asked for is regenerated rather than shown to you. None of that makes it reliable enough to hand to a class unread.

Studio is built on the assumption that a qualified adult reads everything before a student sees it. Every generated document is labelled as AI-generated in the product, and the label is a genuine instruction rather than a disclaimer:

• Check the facts, and check the answer key, before you print.
• Check that the reading level and the tone suit the students you actually teach.
• Edit freely. The output is yours — see the Terms of Use.
• Tell us when something is wrong. Every resource has a thumbs up or down with reason tags, and what you send reaches us.

Studio is a tool for educators, not a student product. Students have no account and no login. The one place a student takes part is a live quiz you run, and no AI is involved in it at any point:

• The questions were generated before the session started, and you saw them first.
• A student’s answers are compared against the saved answer key on our own servers. No model reads them, and they are never sent to any AI provider.
• Nothing about a student is stored beyond the session. Names, answers and timestamps are deleted within 24 hours.

We also ask you not to upload student personal information as source material — marked work, class lists, records. We cannot detect it, and source material does reach an AI provider. Section 8 of the Terms of Use covers this.

Different jobs in Studio run on different models, chosen by testing rather than by reputation. Before a model is put in front of teachers we run it against a fixed set of real lesson-prep tasks and check the output for factual accuracy, correct arithmetic in answer keys, appropriate reading level, and whether it invents sources.

We change models when a better one appears, and we may use different providers for different tasks. What does not change is the rule in section 3: whichever model runs, it is not permitted to train on your material.

If Studio produced something wrong, unsafe or unfair, we want to know. Use the feedback control on the resource, or email support@teachit.co.uk. A specific example — the resource, the question, what was wrong — is far more useful to us than a general report, and it is what we test the next model against.

Back to Teachit Studio