AI Disclosure

Last updated: August 31, 2026

Limud uses artificial intelligence throughout the product - tutoring, grading assistance, content generation, and personalization. This page focuses specifically on the automated inferences Limud makes about individual learners: what they are, how they are produced, what oversight exists, and how training data is handled.

1. The Evolving AI Note

After a student completes a graded assignment or an AI tutor session, Limud updates an evolving note about that student - their inferred help level (independent, moderate, high, or intensive support needed), a short narrative summary, strengths, growth areas, and teaching strategies that seem to work for them. This note is re-derived after each new learning event, not written once; it also keeps an append-only history so you can see how it has changed over time. It draws only on that student's own data on Limud: their interest/learning-style survey, mastery records, and recent graded work.

2. AI-Generated Versus a Deterministic Fallback

Limud asks its AI model (Google Gemini) to write this note. If the model is unavailable, misconfigured, or returns no usable result, Limud computes the note instead from a fixed, deterministic rule based on the student's stored mastery levels and average scores, so the learning loop still closes even offline. Every note record in the product distinguishes which of these wrote it, and both the teacher-facing and parent-facing views label it explicitly - "AI inference, not a finding" / "Written by Limud's AI" when the model wrote it, or "Rule-based fallback, no model wrote this" / "Written by a fallback rule, not the AI" when it did not.

3. Confidence and When a Claim Is Withheld

Each note carries a confidence level that increases as more learning events accumulate. The parent-facing view shows this confidence alongside the number of learning events behind the note, and below a confidence threshold it withholds the descriptive claim entirely rather than showing a low-confidence guess dressed up as a finding.

4. Human Review

The note is generated automatically and used in the AI tutor's prompts to personalize its responses without a human approving each update first. Teachers and parents can view the current note, its provenance (AI or fallback), its confidence, and its full history at any time; they cannot edit the note itself today. If you believe a note is inaccurate, you can raise it with your school or with us (see "Corrections" below), and grades themselves - separate from this note - remain entirely under the teacher's control.

5. Where the AI Note Is Used

The note is folded into the context Limud gives its AI tutor so tutoring responses can be personalized to a student's inferred needs. It is explicitly instructed not to mention its own existence to the student. It is not currently used for automated decisions that affect grades, placement, or discipline - grading is performed separately, and where AI assists grading, our Terms of Service state that AI grading is advisory and teachers retain final authority.

6. Training-Data Use

AI prompts and student data we send to Google's Gemini API are processed to generate the requested response; we do not permit Google to use this data to train its models. We do not use student data to train our own models either.

7. Managing This

A parent can withhold the "AI interaction" consent category described in our Children's Privacy page; we do not currently have a way to disable only the AI tutor and note for one child while leaving the rest of their account active, so withholding this category may limit which features are available. Contact privacy@limud.co if you need this.

8. Corrections and Questions

Email privacy@limud.co if you believe an AI-generated note or other automated inference about a student is inaccurate, or with any other question about how Limud's AI works.