Bloom learns how to teach better by looking at how its lessons actually go — not at who your family is. Nothing here identifies a student, parent, or counselor, nothing here is sold or shared, and none of it is used to train AI models.
1. What is collected
Structured facts about studying, and nothing else:
- practice attempts — the topic, the kind of question, how many were right out of how many, and the day;
- evidence of a skill or a misconception — the topic, which misconception came up, and the chosen option or numeric answer for a graded question;
- study sessions — the sequence of topics and step types, how it ended, and how long it ran;
- teaching runs — the topic, which version of the lesson was used, how many hints and retries it took;
- tutor decisions — which level of help was offered, why, and whether it was delivered;
- numbers from a weekly progress report, such as streaks and changes in mastery;
- grade level, graduation year, and US state — the context that makes the rest mean anything;
- that a safety screen fired, which category it matched, and the day.
Dates are kept to the day. Times within a day are dropped.
2. What is never collected
No names, emails, addresses, or logins. No essays or essay drafts, no chat conversations with Bloom, no journal entries, no student stories, no counselor notes, no uploaded documents, no transcripts, no test scores, and no financial information of any kind. No free text a student wrote. No school name. Nothing from the student-private side of the parent–student wall — those records are excluded by construction, not by a filter someone could forget to apply.
And to say the important one twice: this data is not used to train AI models. It is counting and comparison, not training.
3. The pseudonym
Every row carries a pseudonym instead of a student. The pseudonym is derived with a secret key that Bloom's app never holds, so the app cannot work out which pseudonym belongs to which student, and the pseudonym cannot be turned back into a name, an email, or a family. Without that key, the research records are rows about learning with nobody attached to them.
Anything Bloom ever publishes from this data has to describe at least twenty different pseudonyms, so no single family can be picked out of a number.
4. Why it exists: help the next family
Bloom's teaching gets better when it can see which explanations actually work — which misconception a real ninth grader hits on slope, which sequence of steps got them past it, which version of a lesson produced understanding and which one produced a retry. That is the whole purpose.
There is no incentive for taking part: no discount, no credit, no unlocked feature, and no different price. Paying families to hand over data changes what the data means. Taking part simply helps the next family who sits down with the same lesson.
5. Leaving deletes the contribution
Turn the program off and every row carrying that family's pseudonym is deleted. Deleting a student, or erasing the whole account, does the same thing. The only exception is a published aggregate that already counted at least twenty families: a number already printed is not clawed back, and it contains nothing about your family in the first place.
6. After a plan ends
Contributions stay after a subscription ends, because a family joined the program, not the plan. If you would rather they did not, opt out before you leave — or ask us afterwards — and everything contributed is deleted, the same as any other opt-out.
8. Training program (opt-in)
Everything above is counting and comparison. The training program is a separate, second decision, and it is the only part of Bloom that uses a family's data to train an AI model. It is off by default. Nothing described in this section happens unless a parent turns it on.
What joining adds, and it is one thing: the short answer a student typed for a graded question — a number or a math expression, such as -3/4 or 2x+1 — kept beside the topic and the misconception it went with, so Bloom can learn which wrong answers a real student actually writes. The answer we were expecting is kept beside it.
Never prose. A typed answer that is not a number or an expression — a sentence, a name, an email address, a school — is not shortened or cleaned up, it is dropped. Answers longer than forty characters are dropped too. Nothing else changes: no chat, no essays, no journal, no messages, no free text of any kind, and every rule in sections 2 and 3 above still holds, including the pseudonym.
A parent opts the household in, on the Settings page, under “Help train Bloom's tutor.” A student aged 13 or older can opt themselves out of the training program on their own Settings page, and keep everything else exactly as it was — leaving the training program does not leave the program in section 1.
Leaving deletes the contribution, the same as section 5: switch the training program off and every short answer contributed under that pseudonym is deleted. Deleting a student or erasing the account does the same thing.
7. How to opt out
Open Settings in Bloom and switch off “Help improve Bloom's teaching.” It takes one click, it applies to the whole household, and a student can switch it off for themselves. There is no form to fill in and nobody to email for permission.
See also our Privacy Policy and our data practices in detail.