How Products Vary
We found three categories of products.
Three of the twenty products we studied are general-purpose chatbots powered by large language models (LLMs).
ChatGPT, Claude, and Gemini
Most other AI-powered products we saw on the tour use these LLMs as a base for their functionality. The parent companies of these models (i.e., OpenAI, Anthropic, and Google) each offer a combination of free and paid products aimed at educators or school systems.
ChatGPT limits access below the age of 13 and requires parental consent for 13–17 year olds; Anthropic restricts access to Claude to account holders who are 18+, providing no account access for minors; Google offers some products and functionality for use with K–12 students.
Five of the twenty products are education-focused, multipurpose platforms.
Note: The product counts in this section reflect the 16 products that had complete profiles at the time of publication (i.e., September 2026).
Brisk Teaching, Google Classroom (Gemini is a general-purpose chatbot but Google Classroom offers a range of functionality), Kira Learning, MagicSchool, Playlab
These products give teachers tools to choose from or the means to build their own tools. They can be used by teachers to assist with a range of tasks such as lesson planning, grading, differentiation, IEP language drafting, and family communication. These platforms have student-facing elements, but what those look like depends on what the teacher, school leader, or school system leader assembles, builds, or assigns (e.g., working through a practice activity or quiz, getting feedback on a draft, chatting with an AI tutor for help on an assignment, or using a custom app the teacher built for an activity in a specific lesson).
Thirteen of the twenty products are targeted instructional tools.
Amira Learning, Atlas by Kiddom, Coursemojo, Enlighten, Goblins, Khanmigo by Khan Academy, LitLab, Magma Math, Magpie Literacy, OKO, Paloma, Quill, and Snorkl
Of these 13 products, five are designed for multiple subjects, four are focused exclusively on literacy, and four are focused exclusively on math.
Though it is too soon to make definitive claims about the value each category can offer, three things are clear:
1. The general-purpose chatbots pose the clearest threat to shortchanging student thinking.
2. The multipurpose platforms are likely to be used in a wide variety of ways, yielding a broad range of outcomes.
3. The targeted instructional tools—particularly the products that have a clear opinion about what teachers can do during product use to support learning and nudge teachers to support those actions—are more likely to deliver consistent learning experiences across classrooms.
We found five dimensions of product design that we think will matter to students’ learning experiences and learning outcomes.
1. AI's role: Students reported liking the ability to ask questions and liking products that helped them understand what they needed to do differently to get the right answer immediately.
• In seven products, the AI asks questions of students, without an option for students to ask questions.
• In seven products, questioning is bidirectional: the AI asks students questions, and students can ask the AI questions.
• In two products, the AI does not interact with students directly; instead it analyzes work and surfaces insights for the teacher.
Of the 14 products in which the AI interacts directly with students, the feedback it offers differs:
• In four products, the AI checks the answer and tells students only whether their answers are correct or incorrect, without explaining what went wrong.
• In 10 products, the AI flags a response as right or wrong and, when applicable, provides an explanation of what the student did incorrectly and how to improve.
2. Teacher's role: Products with a clearer perspective on the teacher's role were used more in line with their intended approach and more consistently across classrooms.
• Nine products have a clear point of view about what teachers should be doing while students are engaging with the tech and are designed to support teacher actions (e.g., dashboards that nudge teachers based on patterns of student work).
• Seven do not specify what teachers should be doing during product use, instead encouraging a range of ways teachers could engage.
• All 13 of the products with a clear instructional focus have a teacher dashboard that gives teachers insight into student work, and most, but not all, have a leader dashboard for instructional coaches, school leaders, and system leaders.
3. Individual/social setup: With some exceptions, we found the products designed exclusively for solo use to be more boring to students.
• Four products are designed exclusively for independent student use.
• Two products are designed to be used solo during parts of the lesson and in small groups during other parts of the lesson.
• One product is designed only to be used in small groups.
• One product is designed to be used with parents.
• Eight products can be used in a range of ways.
4. Dosage recommendations: Products that were clear about dosage were more likely to hit the target dosage.
• Thirteen products have a clear recommendation on how frequently and how long students should engage with them. Dosage recommendations vary from daily to weekly, 10 to 30 minutes per session. Some products have started to move away from "minutes" to focus more on activity completion.
• Three do not offer a clear recommendation.
5. Curriculum alignment: In contexts with a core curriculum in place, teachers and leaders valued products that more closely reflected their materials—preferring products that totally synced with the questions and design of the core curriculum. In contexts without a fully implemented curriculum, teachers valued products that gave them a big bank of tasks or questions to choose from.
• Curriculum-embedded: Six products are built to reflect and pair with the specific texts, questions, and scopes and sequences in particular curricula.
• Curriculum-connected: Six products provide their own content matched to a particular curriculum's scope and sequence but do not contain, and are not designed to reflect, the specific texts or tasks in a given curriculum's materials.
• Standards-only: Four products work only from state standards. They do not read or engage a curriculum's content; these products may produce texts and learning experiences for students based only on the standards.
Evidence of impact on student learning
Note: The product counts in this section reflect the 16 products that had complete profiles at the time of publication (i.e., September 2026).
Independently evaluated, causal studies that show evidence of positive impact on student learning overall and for students in priority groups (i.e., students of color, students experiencing poverty, multilingual learners, and students with disabilities) should be the expectation for all student-facing products.
Currently, seven products have independent reviews that examine student achievement across student groups in the US; two products have conducted studies outside the US. Four products have external studies underway; three products are currently looking only at internal data.
This fall, we’re partnering with Stanford University’s AI in Education Lab. They will conduct secondary reviews of the evidence of impact that product developers have provided. By winter 2027, they will provide summary information about the strength of each study and the magnitude of impact overall and for students in priority groups.