OpenAI is buying scale through state-level data agreements
OpenAI's expansion of ChatGPT for Teachers added 55 school systems across 20 states and nearly 140,000 educators and staff, bringing total program coverage to more than 340,000 educators and staff across 30 states and more than two million students. The new cohort includes one in five of the 20 largest U.S. public school districts. OpenAI VP of Education Leah Belsky was explicit about the model's boundaries: 'We are not opening any student accounts or asking schools to figure this out on their own.' The program remains free for verified educators through June 2028 and excludes student accounts entirely.
The mechanism OpenAI is using to scale faster than one-district-at-a-time sales is a 16-state Data Privacy Agreement built through the Student Data Privacy Consortium, which gives participating states a standard legal route to adoption rather than forcing each district to negotiate terms from scratch. That is a procurement shortcut, and it is the kind of move that should catch the attention of any CIO who has watched a single vendor contract take a year to clear legal review. Belsky's second point, 'Access alone is not enough for educators,' is backed by real usage data: Township High School District 211 reported that its share of staff at the lowest 'AI-aware' tier fell from 43 percent to 11 percent over eight months, with 65 percent reporting active use of ChatGPT at work.
Google is embedding inside a tool teachers already trust
Google's bet runs through Khan Academy rather than around it. Six Google engineers spent six months embedded with Khan Academy's team through a Google.org Fellowship, rebuilding Khanmigo with Gemini-powered visual tutoring that generates interactive diagrams, charts, and geometric shapes when the system judges a visual will help a student understand a math or science problem. Students can drag elements of a diagram and the system recognizes the change and adjusts its response. Khan Academy founder and CEO Sal Khan called the partnership a case where 'Google is a tremendous partner,' crediting the embedded engineers with speeding up feature development significantly.
On the teacher side, Google rebuilt Khan Academy's Practice My Knowledge tool so teachers draft multiple-choice questions with Gemini, upload their own reference materials, and then edit or reject AI-drafted questions before anything reaches students. That approval gate is the detail worth noting: Google's design keeps a human decision point between generation and deployment, rather than pushing AI-authored content directly to students. Jen Carter, Head of Technology at Google.org, framed the goal as supporting 'Khan Academy's visionary approach to education' by making learning 'more effective, engaging, and accessible.' Google separately rolled out Gemini Notebook features including voice chat in nearly 100 languages and short video overviews in more than 80, available free to U.S. college students for a year through Google AI Pro.
Anthropic is still playing the training-first game
Anthropic's approach looks different from both competitors because it runs through partners rather than direct district contracts. Its partnership with Teach For All, a network spanning 63 countries, has already put AI fluency training in front of more than 100,000 educators. That is a training-and-coalition strategy rather than a procurement strategy, and it trades the speed of OpenAI's state-level data agreements for depth of relationship with teacher-training organizations that already have credibility inside schools.
An analysis from Outsell frames the three companies' education pushes as distinct bets on what will actually drive long-term loyalty among teachers: curriculum alignment, LMS embedding, or administrator control. That framing matters because each path implies a different long-term winner depending on how your district or university actually buys and renews software. A product that wins on curriculum alignment competes on content quality and instructional design. A product that wins on LMS embedding competes on how deeply it is wired into Canvas, Schoology, or Google Classroom. A product that wins on administrator control competes on compliance, auditability, and the kind of governance dashboards a district technology director actually signs off on.
The real differentiator is proof of usage, not access
Every one of these companies is now past the free-access-announcement phase and into the harder work of proving classroom usage translates into measurable outcomes. OpenAI's own summer 2026 AI Skills Jams, run with the Walton Family Foundation, drew more than 1,600 participants, with 93 percent reporting something reusable and 96 percent planning to apply it within 30 days. Those are self-reported training outcomes, not measured classroom results, and the gap between the two is exactly where a skeptical buyer should focus due diligence.
Lynwood Unified Superintendent Patrick Gittisriboongul's framing is the one worth repeating to any vendor pitching a district deal: 'AI should never replace the human work of teaching. It should really create more room for it.' That is the test every one of these three products should be measured against, not how many educators have accounts, but whether teachers report the tool actually freeing up instructional time rather than adding another system to manage.
What this means for procurement teams
If you are evaluating AI tools for a school system or university, the three vendors are no longer interchangeable free tiers competing on price. OpenAI is optimized for districts that want a fast, state-level legal pathway and are comfortable with a workspace model that excludes student accounts entirely. Google is optimized for districts and universities already standardized on Google Workspace, with the Khan Academy integration specifically valuable where Khanmigo is already in use. Anthropic is optimized for organizations that value training depth and coalition partnerships over rapid district-by-district rollout.
The practical move is to map your own procurement constraints, data privacy posture, existing LMS stack, and teacher training capacity against these three distinct strategies before any pilot begins, rather than running a bake-off on features alone. The vendor that wins your district will be the one whose underlying bet, curriculum, embedding, or governance, actually matches how your institution buys and sustains software, not the one with the flashiest demo.
