Ask and AI use

This statement explains where rTraining uses AI, what it is and is not used for, what is sent to the companies that provide it, and what we keep. The short version: AI helps people read and helps our authors draft, and a person stands behind everything a learner is taught or assessed on.

Where AI is used, and where it is not

Used:

• Ask and the section tutor answer learners' questions from the text of the courses they hold.
• Drafting courses. Our authors use AI to draft course sections and assessment questions from the sources they give it.
• Narration. The spoken version of each section, and the read-aloud of assessment questions, is a synthetic voice generated from the course text.
• Course cover images are generated illustrations.

Not used:

• Marking. Assessments are marked automatically against a fixed answer and a published pass mark. No AI model is involved in marking, and none decides whether anybody passes.
• Decisions about you. Appeals, reasonable adjustments, malpractice findings, withdrawing a certificate and lifting a lockout are decided by people.
• Publishing. No accredited course version, or one applying for accreditation, can be published without a named author and a separate, independent reviewer, whoever or whatever drafted it. Any other course is published by our staff, credited to the named people or to the rTriibe team.

Ask and the section tutor

Inside a lesson, Ask about this section answers questions about the course you are reading. The Ask page in your account answers questions across every course you hold, and about how rTraining and your own account work. Ask is part of Premium.

What it is for, and what it will not do

• It answers only from the text it is given: the courses you hold, rTraining's own help pages and, on the Ask page, a summary of your account. It cites the course and section each answer comes from. General knowledge is not a source for it, even when it is right.
• If your courses do not cover a question, it says so, and suggests your manager, your setting's policy or its lead for the topic.
• It is not for emergencies. A message that reads as an emergency happening now gets a fixed reply telling you to call 999 and follow your setting's emergency procedure, without being sent to the AI model at all.
• It will not help with assessments. It is switched off while you have an assessment open, and a message that looks like one of our assessment questions is refused before it reaches the model. The check is made inside our own database, so the AI provider is never sent our questions.
• For emergency procedures in a course, it quotes the course's own steps, in the course's order, rather than paraphrasing them.
• It is automated and can be wrong. The course text is always the version of record. If an answer and the course disagree, the course is right, and you can mark the answer as unhelpful and say why.

What is sent to answer you

Ask is answered by OpenAI, as our processor. Each question sends:

• your question, up to 800 characters;
• your last three questions and answers in the same conversation, read from our database;
• the published text of the courses you hold: the whole course inside a lesson, or the sections that best match your question on the Ask page;
• the list of guidance and statutory sources those courses cite;
• the rTraining help pages that match your question;
• on the Ask page only, a summary of what your own account shows: your plan, the courses you can open, how far through each you are, the dates of your latest certificate for each and when it needs renewing, any due dates your employer has set, and which catalogue courses your plan does not include.

Never sent: your name, email address, employer, or any identifier for you; anything about another learner; and anything from your assessments, meaning no questions, answers, scores or attempts.

Separately, your question alone is sent to OpenAI's moderation check to see whether it is abusive, and when a new chat on the Ask page is named, the first question and the start of its answer are sent once, with nothing about you. We use OpenAI's API with storage switched off (store: false), and under OpenAI's API terms it does not use data sent through the API to train its models. OpenAI may keep API data for a limited period to detect abuse, as its API terms allow, and then deletes it. OpenAI handles this data in the United States.

What we keep

• Your questions and the answers, the section, and any rating you give, for 90 days. Then the words are deleted. You can delete a conversation yourself before then: it leaves your history at once, and we keep the record until the 90 days are up for safeguarding and misuse review.
• Who can read them: the people who write the course see questions and answers, never who asked. Our staff can read them with your name, to check the tutor is answering safely, to deal with misuse and to help when you contact support. Your employer never can.
• A usage record with no words in it: how much text was sent and received and what each question cost, because your monthly allowance is counted from it. It has no end date set, and it is deleted when your account is deleted.
• Nothing you type into Ask, and nothing it answers, is sent to our analytics service.

Limits

• A monthly allowance of use, counted in the cost of the model's work. You can buy a top-up if you run out; see cancellation and refunds.
• No hourly or daily limit. If many questions are sent at once, Ask asks you to wait a moment before the next.
• Misuse pauses it. A message asking for other people's data, for the tutor's instructions or keys, or to get round its rules, or one the moderation check finds abusive, is refused and recorded with what you typed, for the same 90 days. Three in 24 hours pause Ask, and the section tutor, for your account for 24 hours. The pause is automatic. You are told on screen that Ask is paused, why, and when it will start again, and it lifts by itself at that time. Your courses stay open to read throughout. Our staff are told, and can read the questions that caused it. If you think a question was refused wrongly, tell us and a person will look at it.

Please do not type personal details

Do not type a child's name, or anything about a real person's health or circumstances, into Ask. It does not need them to answer, and they would be kept for 90 days. If you are worried about a child or an adult at risk, follow your setting's safeguarding procedure; see our safeguarding statement.

How courses are drafted

• Drafting. Course sections and assessment questions may be drafted by Anthropic's Claude models from the sources our authors choose. The model is given the sources and the course plan. It is never given anything about a learner. Anthropic handles this data in the United States.
• Our staff use Claude through a connector to work on draft courses. The connector signs in as a named member of staff, every change it makes is recorded in our audit log as made through the connector on that person's behalf, and questions it adds go to review rather than into the live bank. It can read the names, qualifications, registration numbers and email addresses of the authors and reviewers on our register, and nothing about learners.
• People are responsible. An accredited course version, or one applying for accreditation, cannot be published unless a named author and a different, independent reviewer are recorded against it, and the database checks that they are two different people. A course with no accreditation may be credited to the rTriibe team instead. Where a reviewer is named, a member of staff has to confirm that the reviewer has reviewed the version. See conflicts of interest.
• Where a course states a legal duty, a clinical step or a number, it comes from the primary guidance the course cites, and each section records the sources it relied on.

Narration and images

• Narration is generated by ElevenLabs from the course's own text, when the course is prepared rather than while you listen. Only course text and question text are sent. No learner data is ever sent. The voice is synthetic. Every course can be completed in silence, and the written text is the version of record.
• Checking the narration. A voice model can say something it was not given, so each recording is transcribed by OpenAI and the transcript compared with the course text, and a person looks at anything that does not match. Only our own recording is sent.
• Cover images are generated by OpenAI from our own descriptions of the course. They are abstract illustrations, and none of them carries information the page does not also give in words. No learner data is sent.

Questions or concerns

If you think Ask gave a wrong or unsafe answer, mark it and tell us at training@rtriibe.com. How we handle your data is in the privacy notice, and the full list of the companies involved is on the sub-processors page.