Student data privacy
Student records are protected under FERPA, and services collecting personal information from children under 13 are covered by COPPA.[1,2] Many districts use the Student Data Privacy Consortium’s national agreement and registry.[3]
| Ask | A good answer |
|---|---|
| Will you sign our data privacy agreement (or the SDPC national agreement) as written? | Yes, or a named list of changes. Check the registry for agreements they've already signed. |
| What student data do you collect, including chat logs and photos of student work? | A specific list, and nothing beyond what tutoring and reporting need. |
| Is student data used to train AI models, yours or a third party's? | A clear no, in the contract, covering every model provider they use. |
| Which AI model providers process student data, and under what terms? | Named providers with no-retention or no-training terms you can read. |
| Where is data stored, how long is it kept, and how do we delete it? | A retention period and a deletion process the district controls. |
| Have you had a data breach? How did you notify schools? | A straight answer, with dates and what changed. |
Curriculum alignment
A tutor that pulls students into its own sequence competes with your curriculum instead of supporting it.
| Ask | A good answer |
|---|---|
| Can students work on problems from our adopted curriculum, or only from your library? | It serves your materials, and you can see the mapping before you buy. |
| Does it use our curriculum's vocabulary, models and strategies? | Examples from your curriculum, shown in the demo. |
| How is it aligned to our state standards, and who checked the alignment? | Standard-by-standard mapping, not just “standards-aligned.” |
| What grades and courses does it cover today, not on the roadmap? | A list that matches your grades now. |
Teacher workload
The data only helps if teachers have time to act on it.[4] Ask what the tool adds to a teacher’s week, not just what it saves.
| Ask | A good answer |
|---|---|
| How long does setup take per class, and who does it? | Minutes, with rostering through your existing systems. |
| What does a teacher do with the reports, and how long does it take? | A concrete routine: groups, reteach points or check-ins, readable in a few minutes. |
| Does it give grouping and next-step suggestions, or only raw data? | Suggestions tied to misconceptions, which teachers can accept or change. |
| Does it work with our learning management system and rostering? | Named integrations you already use. |
Rollout and support
Most tools fail on implementation, not features. Plan who uses it, when and how often before you sign.
| Ask | A good answer |
|---|---|
| What does a pilot look like, and what will you measure with us? | A defined pilot with a pre- and post-measure you agree on before it starts. |
| How much use per week does it need to work? | A specific dose, with the reasoning or study behind it. |
| What training do teachers and coaches get, and is it included in the price? | Named sessions, included or priced separately in writing. |
| Who do teachers contact when something goes wrong, and how fast do they hear back? | A named support channel and a response time. |
| Does it work on our student devices and network, including with filters? | A technical checklist, tested in your environment during the pilot. |
Evidence of learning impact
ESSA defines four tiers of evidence.[5] Design matters as much as the model: the same AI helped or hurt depending on whether it gave answers.[6]
| Ask | A good answer |
|---|---|
| Is there a study of this product's effect on learning? Can we read it? | A full report, not a one-page summary, with its ESSA tier if it claims one. |
| Who ran it, with how many students, and how were they compared? | Independent or clearly described researchers, a comparison group and a sample like yours. |
| Did it measure learning without the tool, after use? | Yes. Performance with the tool switched on can rise while learning falls. |
| If there's no study yet, what pilot data can you share? | Results from named schools with the measure used, and a plan to study it. |
How the AI behaves
These are the checks from the 20-minute test, asked out loud.
| Ask | A good answer |
|---|---|
| When, if ever, does the tutor show an answer or a worked solution? | Never during practice, or a clear rule your teachers can control. |
| Does it read handwritten work? In which formats? | Photo, stylus or typed steps, demonstrated with messy work. |
| How does it decide which misconception a student has, and how often is it wrong? | A described method and an error rate they've measured. |
| What stops it from going off-topic or giving inappropriate responses? | Specific guardrails and a way to review conversations. |
Cost
Many school-facing tools don’t publish prices. Get the full number in writing.
| Ask | A good answer |
|---|---|
| What is the price per student per year, and what's included? | One number covering licenses, training and support. |
| What happens to the price after the first year? | Renewal terms stated in the contract. |
| Can we pilot before committing to a multi-year contract? | Yes, with a clear exit. |
See how the tools answer some of these already in the comparison, and check which criteria matter most for your use in best by need.
Sources
Every claim above links to one of these. Vendor pages were read on November 10, 2026; check them again before you buy, since products change.
- FERPA (34 CFR Part 99) and guidance for K-12 school officials and vendors. U.S. Department of Education, Student Privacy Policy Office.
- Children's Online Privacy Protection Rule (16 CFR Part 312). Electronic Code of Federal Regulations.
- Student Data Privacy Consortium: National Data Privacy Agreement and the agreement registry. Access 4 Learning Community.
- Assessment and Classroom Learning (Black & Wiliam, 1998). Assessment in Education: Principles, Policy & Practice.
- Evidence for ESSA: programs that meet ESSA evidence standards. Center for Research and Reform in Education, Johns Hopkins University.
- Generative AI without guardrails can harm learning: Evidence from high school mathematics (Bastani et al., 2025). Proceedings of the National Academy of Sciences.