AI is the engine of Blabble, so it deserves its own plain-English policy. This page explains how the AI actually works, the limits we’ve built into it, and your responsibilities when you use it. It forms part of our Terms of Service and sits beside our AI Ethics principles.

Scope

This policy covers every AI-powered feature in Blabble: document analysis, quiz / flashcard / mind-map generation, study planning, IB-style practice exams and grading, the Socratic tutor chat, voice transcription, and content moderation. Where this policy and the Terms of Service overlap, both apply.

How Blabble uses AI

Blabble’s AI is grounded: wherever possible it works from the materials you upload, rather than inventing from thin air. Each feature uses AI for a specific, bounded job:

Document analysis: Reads your uploaded notes and extracts key concepts, knowledge gaps and suggested focus areas.
Study materials: Generates quizzes, flashcards, mind maps, summaries, explanations and diagrams from your materials.
Study plans & exams: Builds exam-urgency-aware plans and IB paper-style practice exams, and grades your free-text answers against rubrics.
Socratic tutor: Guides your thinking with questions and explanations instead of simply handing over answers.
Transcription: Turns voice study recordings you make into text.
Moderation: Screens chat for unsafe content before the tutor responds (see “Safety & guardrails”).

The models & providers we use

All AI runs through a single provider, OpenRouter, using our own API key. We keep the provider list short on purpose:

Tutoring & generation: A general-purpose language model powers analysis, content generation, planning, practice grading and the tutor.
Safety moderation: A separate safety model screens chat messages for harmful content.
Transcription & OCR: Additional models via OpenRouter convert voice recordings to text and read text from scanned PDFs and images.

We may change the specific models we use to improve quality, safety or cost, while keeping the protections in this policy.

What we send the AI — and what we don’t

For each task, Blabble sends only the content that task needs: the relevant document text, your chat messages, your quiz or exam answers, or the highlight you asked about. It does not ship your whole account to the model.

Rule of thumb

Treat the AI like a capable but external assistant. Don’t paste in passwords, ID numbers, or sensitive personal information about yourself or others that isn’t needed for studying. What you send to a feature is what gets processed to produce your result.

We do not train on your work

We do not use your documents, chats, recordings or generated materials to train Blabble’s AI, and we do not sell your content or use it for advertising. Every inference request tells OpenRouter to use only zero-data-retention endpoints and to deny model providers that collect prompt data.

OpenRouter and the selected model provider still process the request to return an answer. OpenRouter keeps non-content request metadata such as token counts and latency and describes limited anonymous prompt categorisation. We minimise what we send and ask you not to submit information you would not want an external AI provider to process. See the Privacy Policy for the data and transfer detail.

Accuracy, hallucinations & verification

AI can be confidently wrong

Large language models can produce information that sounds right but isn’t — including made-up facts, flawed reasoning, or incorrect marks on a practice answer. Blabble’s grounding reduces this, but it cannot eliminate it.

Always check AI output against your textbook, syllabus and teacher before you rely on it — especially close to an exam.
Treat practice-exam grades and feedback as indicative coaching, not official marks.
If something looks wrong, it may be. Tell us, and don’t reproduce it in assessed work without checking.

Human oversight & your control

You stay in charge. AI output is clearly presented as AI-generated, you can regenerate, edit, ignore or delete it, and a human (you, and where relevant your teacher) always interprets it. For questions about any AI output, real people are reachable at blabbledevai@gmail.com.

How to read automated assessment

Blabble automatically scores quizzes and practice exams, estimates your mastery per topic, and builds an exam-readiness picture; teachers see related at-risk signals. These are designed to point you to what to study next.

Guidance, not a verdict

These signals are study aids. They do not set your grades, decide admissions or determine anything official about you, and Blabble does not make decisions with legal or similarly significant effects about you by automated means alone. See the Privacy Policy for your rights here.

Acceptable use of AI

The whole point of Blabble is to help you learn honestly. So:

Encouraged

Using AI to understand concepts, test yourself, plan revision, get unstuck, and practise under exam conditions.

Not allowed

Using AI to produce assessed work you’ll submit as your own, to cheat, or to disguise AI-written text as human.

Some doors are bolted shut by design

Blabble’s essay-writing and AI-“humanizer” (detection-evasion) features are switched off by default and intended to stay off. The tutor is built to guide your reasoning, not to hand you finished answers to copy. Trying to jailbreak these safeguards is a breach of these Terms.

Safety & guardrails

Blabble runs a two-layer safety system. First, the tutor is prompted to stay educational and age-appropriate. Second, an independent moderation model screens chat messages before the tutor replies.

Clearly harmful requests — for example self-harm, sexual content involving minors, weapons or serious violence — are blocked.
Sensitive academic topics (such as history, politics or biology) are allowed but handled with extra care, because students legitimately need to study them.
The safeguards fail safe: if a check is uncertain, Blabble errs toward protecting the student.

No filter is perfect. If you see the AI produce something unsafe or inappropriate, please report it so we can improve the guardrails.

AI credits & fair use

AI work is metered with a friendly AI-credits allowance so the service stays sustainable and available to everyone during the pilot. Heavier tasks cost a little more than light ones, but you only ever see one simple credit number. Teachers and administrators are unlimited. If you reach your allowance, your existing materials remain available and the allowance resets for the next period. Trying to circumvent these limits (for example with automated scripts) isn’t allowed.

EU AI Act transparency

We build in line with the EU AI Act’s transparency expectations. Blabble is an educational support tool, not a high-risk system that decides your education:

You are always told when you’re interacting with AI, and AI-generated content is labelled as such.
Blabble provides guidance and study aids; it does not make binding decisions about your grades, progression or eligibility.
Human oversight and contact are available for any AI output.
We document and review our AI use, and update this policy as the rules and our features evolve.

Report a problem or ask a question

Spotted an issue with the AI?

Tell us at blabbledevai@gmail.com — wrong answers, unsafe output, or anything that feels off. Your reports directly improve the safeguards for every student.

AI Use Policy