The public revision reset and guided Biology practice follow fixed, human-written paths. The controlled photo-lesson lane is limited to approved adult-managed testers: it turns selected schoolwork photos into explanations, questions and source-linked formative feedback. A parent must authorise this expanded purpose once, and the pupil confirms the photo batch before processing.
Effective and last updated: 2 October 2026
The open public session does not send pupil work to AI.
The guided Biology practice uses one original source and a deterministic checker built into the site. A typed practice answer stays in the open tab; that public journey performs no upload or AI-provider request. The separate photo lane is available only when all technical gates are effective and the parent account is allowlisted. Ordinary public uploads and unrestricted AI chat remain closed. Eligible test accounts can use a source-grounded voice and avatar tutor. A controlled test is not a claim of independent approval or general launch readiness.
The only permitted image-AI purpose
Teach the method. Preserve the pupil’s thinking.
Photo lessons help eligible pupils understand their own study material, answer questions from memory, check feedback and practise again. Questions, explanations and example answers must be supported by the supplied source. A second AI check helps find errors but does not guarantee accuracy or complete coverage.
What a pupil would always be told
“AI is being used to prepare explanations and questions from your photos and give feedback when you send an answer. It can be wrong. Keep names and private details out, check every important point against the trusted source beside you, and do not use it for work to submit.”
The controlled boundary
Useful revision support—not an unrestricted assistant.
May do in the gated test
Read selected authorised photos or PDF pages for an eligible 13–16 pupil in an approved adult-managed test account.
Prepare a source inventory, short explanations, glossary terms and questions, then check them against the original photo in a second AI pass.
Offer Teach me and Test me, progressive hints, example answers and source-linked formative feedback on an answer the pupil chooses to send.
Use changed questions and later retrieval practice without claiming mastery or guaranteed accuracy.
Let the pupil explicitly save an encrypted chapter in this browser for up to 30 days, using a separate passphrase and local deletion controls.
Withhold uncertain material and let the pupil replace only pages that need attention.
Must not do
Writing homework, coursework, essays or answers to submit.
Pretending to be a human teacher, friend, therapist, counsellor or emergency service.
Assigning exam grades, predicting attainment, diagnosing ability or deciding whether work is ready to submit.
Completing homework or live assessments. Example answers are for generated practice questions, not work to submit.
Adding outside curriculum facts, certifying that the source is correct or filling a gap by guessing.
Unrestricted general-purpose chat, live web search, public sharing or recording the pupil’s camera.
Advertising, paid steering, emotional dependency, secrecy, guilt or endless engagement.
Using pupil content for our general model training. Provider contractual exclusions must be verified before public launch.
On this page
Photos, pages and conversations
These features are in a controlled, adult-managed test for profiles declared aged 13–16 at creation. General public AI access is not open.
Schoolwork: selected photos or PDF pages are sent as prepared images to OpenAI for explanations, questions and source checks. The original PDF and unselected pages stay on your device. A typed answer is sent only when you ask for feedback.
Tutor: starting a conversation sends microphone audio, temporary conversation text and readable evidence from your selected lesson to OpenAI. When an ElevenLabs voice is selected, only the tutor’s checked text is sent to ElevenLabs to create speech; it does not receive your microphone audio or uploaded images. The configured avatar provider receives the AI tutor’s generated speech to animate its face. In the Spatius preview, speech is sent through our server and the face renders on your device. Downloading its avatar assets exposes connection information, such as your IP address, to Spatius’s delivery infrastructure. Your camera is not used.
Our storage: How2Revise does not save raw microphone recordings or conversation transcripts. Short conversation context stays in the open tab for Pause and Resume. We keep session identifiers, durations, token and speech character counts, connection outcomes and credit transactions. These are usage records, not a recording of what you said.
Lesson storage: credit-backed scan recovery content is encrypted for up to 24 hours; an explicitly activated no-card lesson can be reopened for seven days. An optional encrypted device chapter expires within 30 days of generation. These copies contain generated lesson material and source excerpts; device chapters can also contain prepared photos. Raw typed answers are excluded. Read the storage and deletion details.
Provider handling is separate from our storage. OpenAI’s API documentation says API content is not used for training by default unless the customer opts in. Standard abuse-monitoring retention can be up to 30 days, with stated legal and safety exceptions. Requesting store: false does not establish zero retention. We have not verified a special zero-retention arrangement for this deployment.
ElevenLabs applies its standard processing and retention when its voice is selected. This preview does not use ElevenLabs zero retention mode. The Start notice identifies when ElevenLabs receives the tutor’s checked text.
The Spatius integration sends only checked tutor speech for motion generation; it does not send your microphone or uploaded image to Spatius. This does not establish zero retention or permission for general child use. The Spatius preview remains restricted to the owner’s controlled tests while its contractual and privacy review is completed. Use non-personal test pages and avoid private details in speech.
The exact configured provider, processing purpose, training position, retention limits and rights route are shown before the pupil chooses to send an image.
How2Revise does not write raw images or typed practice answers to its database or filesystem. Generated full-access lesson content can be encrypted for up to 24 hours for interrupted-scan recovery; the optional no-card lesson is stored encrypted for reopening within seven days, as described above. Optional encrypted device chapters can contain photos, source evidence, generated lessons and practice results, but never raw typed answers.
That app boundary is not a promise of zero provider retention. The provider's disclosed abuse-monitoring or safety-retention rules may still apply after a send.
Pupil material and responses must not be copied into analytics, marketing or ordinary developer logs or used for general model training under the intended design.
The model must not receive adult billing identity, a school, exact location or a long-term emotional/ability profile.
A pupil and parent must receive clear, age-appropriate information and a route to report, challenge or request human review.
Real scanning remains gated
The controlled photo-lesson lane requires an authenticated, entitled 13–16 pupil, current pupil-specific parent authorisation for lessons and answer feedback, and one confirmation of the selected photo batch. It does not open ordinary public uploads, complete homework or certify the source. See photo lesson privacy details for processing, optional device saving, expiry and deletion.
Safe image decoding and re-encoding, metadata removal, provider disclosure, rate and cost limits, moderation and input/output controls are technical conditions of the gated lane. Wider pupil release still needs the applicable privacy, safeguarding, incident-cover, copyright, assessed-work, accessibility, educational and consented-usability gates. Selected PDF pages are rendered locally as images; the original file and unselected pages are not uploaded. This page does not claim that independent review or a public launch has happened.
Safety requirements for any AI use
A provider filter alone is not enough. Input and output moderation, personal-data minimisation, age-aware testing, safe fallbacks and a tested incident route are required. Clear danger, abuse, grooming or self-harm language must stop the revision flow and show a trusted-adult, Childline and 999 route.
A named accountable company contact, a reviewed stop-and-signpost protocol and staffable incident handling are required. How2Revise must not claim continuous human monitoring, counselling or emergency response. Read the safeguarding statement.
Release test
The image route remains technically closed unless the applicable account, billing, safeguarding, provider, retention and model-evaluation evidence markers match, both the generation and moderation snapshots appear in code-reviewed immutable allowlists, and the runtime-readiness barriers pass. A feature flag, API key or mutable “latest” model name cannot open it by itself.
Every model, system instruction, schema and safety-control version must be recorded and tested across the declared 13–16 scope, UK language variations, handwriting and capture conditions, accessibility needs, indirect image instructions and harmless material that could be blocked by mistake. Under-13 provider processing is not part of V1.
One unresolved critical failure blocks release. Critical failures include harmful instructions, sexual content involving a child, continuation after a clear emergency disclosure, emotional dependency or secrecy, submission-ready assessed work, private-data leakage, unsafe parent notification, lost safeguarding alerts or training on pupil content.
An AI cannot approve itself.
The accountable director must own the final decision. Real parent/pupil evidence and competent educational review cannot be fabricated by the system that built the product; focused external advice is required wherever the completed risk evidence leaves unresolved specialist questions. This page does not say those approvals have happened.
Using guidance does not mean a regulator, government department or awarding body has approved How2Revise.
Questions or reports
For a privacy question, email privacy@how2revise.com. For a non-urgent child-safety policy question, email safeguarding@how2revise.com. Neither address is an emergency service. For immediate danger in the UK call 999.