How to approve AI for students: a decision guide for schools

[ CONTEXT ]

Start from what students already do

Students are not waiting for a policy. 54% of US teens have used AI chatbots for schoolwork, and 10% do all or most of their schoolwork with chatbot help (Pew Research Center, Feb 2026). Among students from middle school through college, homework use of AI rose from 48% to 62% between May and December 2025 (RAND). 63% of teens who use AI for schoolwork get answers from it, and 35% do not understand their school's AI rules (Common Sense Media, Aug 2026).

So approval is less about letting AI in than about deciding which AI, under which rules, with which adults able to see the work.

Step 1: Decide what students must keep doing themselves

The best evidence on AI and learning points at one risk: AI doing the practice that builds the skill. In a field experiment with about 1,000 high-school students, answer-giving AI raised practice scores but cut unaided exam scores by 17%; a hint-first version of the same AI removed that harm, but it did not beat no AI. In a 30-month study of 26,811 Chinese students in grades 7-12, adopting AI raised homework scores 18% and cut monthly closed-book exam scores 20% within six months, while students who kept spending normal time on homework lost little (a working paper; not randomized).

Write down, for each department, what students must still do alone: the first attempt, the draft, the proof, the translation.

Step 2: Choose the kind of tool

School AI products fall into a few groups: assistants built into a school suite (Gemini for Education, Copilot Chat), activity platforms where teachers build AI tasks (SchoolAI, Brisk Boost, MagicSchool's student Rooms), tutors inside a curriculum (Khanmigo), and approved AI for everyday schoolwork under class rules (Nexus). Pick the group before the brand. Our comparisons date and source each product's facts.

Step 3: Check which rules students cannot switch off

Defaults are not locks. In Common Sense Media's testing, ChatGPT's Study mode offered "Show me the answer" in 43-90% of responses, and it found Gemini's guided mode easy to leave. Even teacher-written rules leak: chatbots set up by 27 middle-school teachers followed the teachers' intended rules 70.4% of the time (preprint). Ask every vendor which rules a student cannot turn off, and test them yourself with a request for the final answer.

Step 4: Settle visibility and privacy

Decide who can read what students type, and tell students before they start. Options range from full staff visibility to class-only review with a log of every read. Remember that monitoring software on a school device works outside any AI tool and may capture anything a student types. At schools that receive federal education funds, stored chats are likely education records that parents may ask to inspect. Read what teachers can see in student AI chats.

Step 5: Check provider terms for minors

The model behind a product has its own rules. Google's Gemini API terms bar use in services "likely to be accessed by individuals under the age of 18." OpenAI's API agreement bars letting minors use its services without a parent's or guardian's consent. Consumer Claude requires users to be 18 or older; Anthropic allows teen use through a business that puts safeguards in place. Ask each vendor which models students get, and who holds the provider account. Read choosing AI models for students under 18.

Step 6: Plan for a student who needs a person

Decide in advance who receives a message that suggests a student may be in danger, in which hours, with what backup, and whether a person reviews it before anyone acts. Abuse disclosures should follow your mandated-reporting process. Read AI chatbot safety for teens and use the safety escalation plan.

Step 7: Write it down and tell families

Turn the decisions into an acceptable use policy, a syllabus statement for each course, a student agreement and a letter to parents. Keep them short enough to read.

Step 8: Pilot, and measure what students can do alone

Start with one or two courses for about six weeks. Measure delayed unaided success, help requests that follow the student's own attempt, cost per active student and what teachers found useful. Usage counts are not learning. Read how to run a six-week AI pilot.

Where Nexus fits

If you block consumer AI, or want to, Nexus is the approved route. Your school chooses the models for each class. Each class has a learning mode (Hints, Guided, or Full answer with critique) that students cannot change. Students are told when a class's teachers can review a chat, personal chats are not shown to teachers, and Nexus records who opened each class conversation.