You open the app store, type "3rd grade math," and forty results show up, half of them with "AI" somewhere in the name or description. They all promise the same thing: personalized learning, adapted to your kid's pace, backed by science. None of them explain how. Choosing, until now, has basically been an act of faith — trusting the reviews, the nice design, trusting that somewhere, someone bothered to check that what that AI generates is correct and age-appropriate for an 8-year-old, and not just a plausible-sounding answer spit out by a language model.
That "someone" is starting to exist. It's called Learning Commons, and while it doesn't solve the whole problem, it moves an important piece of the puzzle.
The Problem No One Could See (Because It Was Behind the Screen)
Building an AI-powered educational app that actually teaches something isn't just about wiring a chatbot to a nice interface. Someone has to decide what a 3rd grader should know in math, what mistakes kids typically make at that stage of learning, which skills are prerequisites for which, and how to check that what the AI model generates actually matches all of that — not just that it sounds convincing.
Until now, every company that wanted to do this right had to rebuild that foundation from scratch: gather curriculum standards state by state or country by country, map learning progressions, build their own verification systems. A process the initiative itself describes as "slow, expensive, and error-prone" — out of reach for most small teams, or for schools that wanted to build their own tools without depending on an outside vendor.
A Database, Not Another Chatbot
Learning Commons is a nonprofit backed by the Chan Zuckerberg Initiative, launched in 2025 to build, in its own words, "AI infrastructure that better connects the way students learn to the tools they learn with." It's not an app for kids, and it's not another AI assistant — it's the invisible layer underneath, and it's free.
It has three main pieces:
Knowledge Graph: an open, free dataset (published on GitHub) covering curriculum standards from all 50 U.S. states, learning progressions, common student mistakes at each stage, and definitions of cross-cutting skills like critical thinking.
Evaluators: tools that measure whether AI-generated content is accurate, age-appropriate, evidence-based, and aligned with real curriculum standards — instead of just trusting that the model "knows."
Agent Skills: a smaller set of tools designed to help teachers with lesson planning.
Since launch, its datasets have been downloaded more than 20,000 times, and more than 70 partners already use them or contribute data of their own — including OpenAI and Anthropic, alongside dozens of smaller edtech companies.
What Changes in a Real School
The theory is fine, but it's more interesting to see it in practice. In the Lisle 202 school district (Illinois), assistant superintendent Jason Markey used the beta version of the Knowledge Graph to build a formative math assessment app that mirrors the official state exam questions. Teacher Vincent Slowiak says the tool automatically flagged which prior standards each student needed to review before moving on — something he used to have to figure out by hand, digging through each kid's history.
Markey is honest about the limits: his team got "pretty close, pretty fast" to 97% of what they needed, but had to hire recent computer science graduates as interns to handle the less glamorous part — single sign-on, teacher dashboards — that no database, however good, solves on its own.
Other examples already underway: Really Great Reading, a literacy company, used the Knowledge Graph to build its reading outcomes system without having to pull data from half a dozen different sources — its own CEO estimates it saved them months of infrastructure work. And Eedi, an 11-year-old math research company, contributed its own data on why students get stuck on certain concepts — data that, according to its president George Mu, they'd been accumulating for years, but that "with the arrival of ChatGPT, suddenly became usable" at a much bigger scale.
Other tools already integrated include Canva Education, MagicSchool, TalkingPoints, and OKO Labs — names you might recognize if your kid's school already uses an AI tool for teachers.
What This Still Isn't
It's worth being honest about the limits here, so as not to swing into the opposite extreme of overhype: Learning Commons isn't a quality seal you'll see in the app store, or a certification a family can look for with a recognizable logo. It's infrastructure — something companies use (or don't) behind the scenes, voluntarily, and that doesn't yet translate into a visible label for someone just browsing for an app on their phone.
That means most of the apps you'll come across today won't tell you whether they're built on this foundation or not. But the fact that it exists changes one important thing: there's now a question worth asking, and a possible answer beyond "trust us."
What to Ask Before Trusting an AI Educational App
We already wrote a general checklist for choosing a good app (in Spanish) — age-appropriateness, ads, active participation, family values. All of that still applies. But when the app has AI built in, it's worth adding a few specific questions:
A. Does the company explain what curriculum standards it's based on? It doesn't have to name the Knowledge Graph specifically, but if there's no mention whatsoever of official curricula, educational frameworks, or being "aligned with state/national standards," that's a sign it might be improvising.
B. Does anyone check what the AI generates, or are they just trusting the model? Look for words like "verified," "reviewed by educators," or "evaluated" in the app description — and if you're unsure, ask customer support directly.
C. Do you recognize the company's name among known partners? MagicSchool, Canva Education, TalkingPoints, Really Great Reading, and OKO Labs already build on this foundation — it's not a closed list or an absolute guarantee, but it's a more concrete starting point than "seems trustworthy."
D. Ask the school directly. If your kid's school already uses some AI tool for teachers, ask if they know what it's based on. It's an increasingly reasonable question to ask, and increasingly likely that someone will know how to answer it.
E. Be wary of the empty promise. "Personalized learning with artificial intelligence," with no explanation of what that personalization is actually based on, remains, to this day, the most repeated and least verifiable phrase in the industry.
The Same Bridge, With the Blueprints Finally in View
Educational technology has always needed to be a bridge, not a wall — supporting a specific child's real learning, not replacing it with something that merely sounds smart. What changes with Learning Commons isn't that every AI app suddenly becomes good. It's that, for the first time, there's a shared, free foundation to build them well on — and a way to start telling apart which companies actually bothered to use it.
The next time an app promises "personalized learning with AI," the question no longer has to stop at "do I believe it?" It can simply be: personalized based on what?