Estimated reading time: 7 minutes · Last updated:
Parents earning $150,000 or more are taking up AI faster than any other income group, and they are the most likely to steer their teenagers toward it. That is the finding of a RAND Corporation study published on 12 November 2026. Between early 2025 and mid-2026, the AI use income gap separating the highest- and lowest-income families grew from 15 to 21 percentage points, driven mostly by top earners. Researchers Heather L. Schwartz and Melissa Kay Diliberti also found that a teen's belief that a parent approves of AI tracked that teen's own use more closely than any other factor they tested except age. Their warning runs in an unexpected direction. The heavier, encouraged use may put the academic performance of higher-income teens at risk.
Key takeaways
- Gap widened: RAND found the difference in AI use between the highest- and lowest-income U.S. families rose from 15 to 21 percentage points over 15 months, driven by parents earning $150,000 or more.
- Top earners encourage more: 17 percent of the highest-income parents told RAND they urge their children to use AI wherever school rules allow, even though most parents counsel restraint.
- Perceived approval matters: In RAND's data, a teen's sense of parental support for AI was more closely linked to that teen's AI use than sex, family income or parents' views on AI harms. Only age was a stronger factor.
- Guardrails, not bans: Schwartz and Diliberti urge states, districts, parents and AI developers to limit unrestricted AI use on students' first attempts at schoolwork, favouring guardrails over outright blocks.
Table of contents
- Key takeaways
- How Far the Income Gap in AI Use Has Grown
- What Higher-Income Parents Tell Their Teens About AI
- Why Perceived Parental Approval Predicts Teen AI Use
- What RAND Wants Schools, Parents and AI Companies to Do
- Will the AI use gap keep widening?
- What to be careful about
- Frequently asked questions
How Far the Income Gap in AI Use Has Grown
The study draws on the RAND American Life Panel, a survey panel built to be nationally representative. Two subsets supplied the data. The RAND American Parents Panel covered about 1,000 parents. The RAND American Youth Panel covered about 1,800 young people aged 12 to 18, enrolled in grades 5 through 12. Parents answered in February 2025 and again in May 2026. Teens answered in May 2025 and May 2026.
Across that 15-month stretch, the difference in AI use between families at the top and bottom of the income scale moved from 15 percentage points to 21. That is a widening of about 6 points. The growth was uneven. Use among parents earning $150,000 or more climbed faster than in any other group, and that surge did most of the work in pulling the two ends apart.
A percentage-point gap is the plain difference between two shares. A 21-point gap means the share of top-income families using AI sits 21 points above the share at the bottom. RAND's one-page overview does not list the underlying share for each income group. Those shares would show whether lower-income use is flat or simply rising more slowly.
What Higher-Income Parents Tell Their Teens About AI
Across income levels, most parents told their teenagers to use AI for schoolwork sparingly. The clearest exception came from the top of the income scale. There, 17 percent of parents said they encourage their children to use AI as much as school rules permit.
That is still a minority, even among the wealthiest families. But it marks a sharp break from the cautious advice most households give. Schwartz and Diliberti argue that parents' own AI habits, combined with this kind of encouragement, are feeding an income-based gap in how much teens use AI for school or homework.
The wording of that guidance matters. Telling a teen to use AI wherever the rules allow sets a ceiling based on permission, not on learning. RAND's own recommendations draw a different line. They separate AI that completes the work from AI that helps a student sharpen ideas they already have. Advice to maximise use within the rules makes no such distinction.
Why Perceived Parental Approval Predicts Teen AI Use
The researchers matched 533 teens with 427 parents living in the same households. That made a dyadic analysis possible. A dyadic analysis pairs each teen's answers with those of a parent from the same home, so attitudes and behaviour can be compared within a family rather than only across the population.
The strongest signal came from the teens' side. How much a teen believed a parent supported AI use was more tightly linked to that teen's own AI use than any other factor the team examined, apart from age. Factors that came out weaker included the teen's sex, the family's income and the parent's views on whether AI causes harm.
Two cautions apply. The link is a correlation, so it does not show that encouragement causes the use. Teens who already use AI heavily may also read more approval into what their parents say. The finding also concerns perception. What a teen thinks a parent believes appears to matter, whether or not the parent ever spelled it out.
What RAND Wants Schools, Parents and AI Companies to Do
For states and districts, RAND proposes two steps. The first is to equip schools with concrete, research-backed messages for students and families. These would explain how relying on AI from a student's first try at an assignment can undercut learning. The second is to put policy-layer controls on devices that schools hand out. These are centrally managed settings, and they would keep general-purpose AI chatbots off-limits while students make their initial attempt at the work.
Parents are asked to talk directly with teens about where the line falls between AI finishing an assignment and AI refining the student's own reasoning. They are also asked to look for technical limits on phones, tablets and other personal devices that curb open-ended AI use.
For AI developers, the ask is narrower. RAND wants guardrails rather than blanket bans, built to separate harmful cognitive offloading from use that supplements a student's learning. Cognitive offloading means handing mental work to a tool so the student no longer does that thinking. The recommendation accepts that some AI help is beneficial and puts the burden on product design to tell the two apart.
| Audience | Recommended action | Where it applies |
|---|---|---|
| States and districts | Evidence-based messages on why unrestricted AI on first attempts can harm learning | Schools, students, families |
| States and districts | Policy-layer controls blocking general-purpose chatbots during initial attempts | School-issued devices |
| Parents | Explicit talks on AI doing work versus AI refining thinking | At home |
| Parents | Technological guardrails limiting unrestricted AI use | Phones, tablets, personal devices |
| AI developers | Guardrail features, not outright blocks, separating harmful offloading from supplemental use | AI products used by students |
Will the AI use gap keep widening?
The case for
- RAND sets out concrete, low-cost steps such as talking points and device controls that districts can adopt without new legislation.
- Developers are asked for guardrails rather than bans, which leaves room for supplemental AI help that RAND treats as beneficial.
- Because perceived parental approval tracks teen use so closely, conversations at home are a lever families already control.
The case against
- The gap grew by about 6 percentage points in 15 months, driven by the fastest-growing group, which suggests momentum at the top.
- School-device controls do not reach phones and tablets at home, where RAND relies on parents to act voluntarily.
- 17 percent of top-income parents already back maximum use within school rules, a stance the recommendations would need to change.
What to be careful about
- The findings rest on self-reported survey answers, which can differ from how much parents and teens actually use AI.
- The link between perceived parental approval and teen AI use is a correlation and cannot show that encouragement causes the use.
- The academic risk to higher-income teens is a warning from the researchers, not a measured decline in grades or test scores.
- RAND's overview does not give per-group usage shares or margins of error, so the size of the 15- and 21-point gaps cannot be checked against sampling noise.
The bottom line
RAND's findings complicate the usual account of technology and inequality, in which poorer families fall behind. Schwartz and Diliberti flag a different exposure. Teens in the best-off households are encouraged at home and live with parents who use AI heavily, and they may lean on it in ways that cost them learning. That is a warning about risk, not a measured drop in grades, and the survey shows association rather than cause. What the data does establish is that the gap between top and bottom earners grew by about 6 points in 15 months. It also shows that teens read their parents' signals closely. Schools, families and AI companies now have a specific list of guardrails to weigh.
What to watch
- Watch for whether any state or school district adopts the policy-layer chatbot controls RAND recommends for school-issued devices; no timetable has been set.
- Watch for AI developers releasing student guardrails that separate harmful cognitive offloading from supplemental help, the approach RAND favours over outright blocks; no date has been set.
- Watch for a further round of the RAND American Parents Panel and American Youth Panel surveys, which would show whether the 21-point gap keeps widening; no date for a new survey has been given.
Frequently asked questions
Who funded the RAND study on parents, teens and AI?
The research was sponsored by the Gates Foundation and carried out by RAND Education, Employment, and Infrastructure. RAND states that its research reports go through peer review and do not necessarily reflect the views of sponsors.
Does RAND think students should not use AI for homework?
No. RAND's concern is unrestricted AI use on a student's first attempt at schoolwork, which it says can harm learning. Its recommendations favour guardrails over outright blocks and accept that supplemental AI help can be beneficial.
Can the RAND AI survey findings be reused?
Yes. The report, numbered RR-A4742-5, is published under a Creative Commons Attribution 4.0 International License. That licence allows copying, adaptation and commercial use with credit, and the report's DOI is 10.7249/RRA4742-5.
Related reading
