Math Grounded in Real Curriculum: Illustrative Mathematics (IM v.360) + Claude

Math is the subject where AI slips quietest. Ask a general-purpose chatbot for a lesson on ratios and it will hand you something tidy: a definition, a couple of worked examples, a set of practice problems. It reads clean. And then you look closer and the examples reinforce a misconception, the "worked" solution takes a shortcut that doesn't generalize, or the sequence jumps a step students haven't been taught. Nothing is flagged, because the model has no source it's answering to — it generated fluent math, which is not at all the same as correct, well-sequenced math. In a subject built on precise, cumulative reasoning, that gap is where understanding quietly breaks.

Claude for Teachers closes it by planning from a real curriculum instead of a blank page. It brings in IM v.360 from Illustrative Mathematics — a trusted, widely used math curriculum — so Claude draws on vetted instructional material rather than improvising math from memory. This piece is about why that matters more in math than almost anywhere else: what grounded math lesson design buys you over generic generation, and how a curriculum teachers already trust keeps AI math honest.

Why math punishes ungrounded generation

Every subject rewards grounding, but math is unusually unforgiving of the lack of it, for three reasons that compound.

First, math is cumulative to a degree few subjects are. Today's lesson assumes yesterday's is solid; skip or scramble a step and the whole sequence wobbles. Ungrounded generation has no reliable model of that sequence, so it produces lessons that are locally plausible and globally out of order.

Second, math has right answers, and wrong ones look just as fluent. A misconception phrased confidently — that you can add fractions across the tops and bottoms, that multiplication always makes bigger — reads exactly as smoothly as the correct idea. Fluency is no signal of correctness here, which strips away the one heuristic a busy teacher might otherwise lean on.

Third, in math how you get the answer is the lesson. A generated solution that reaches the right number by a non-generalizing trick teaches the wrong thing even when it's "correct." Grounding matters because a trusted curriculum encodes the methods worth teaching, not just the answers worth reaching.

A grounded planning partner sidesteps all three by starting from material where the sequence, the correctness, and the methods have already been worked out by mathematicians and math educators — which is precisely what IM v.360 is.

What IM v.360 grounding changes

Bringing IM into the planning loop shifts Claude's job from inventing a math lesson to adapting a trusted one. Here's the difference laid out plainly:

That last row is the one teachers feel most. When a lesson is grounded, your review shifts from is this math even right? to is this right for my kids? — a completely different and far better use of prep time. You're a math teacher adapting trusted material, not a proofreader debugging an improvisation.

Fluent math and correct math look identical on the page. The only way to tell them apart at a glance is to know the source. Claude for Teachers plans from IM v.360, so the source is a curriculum math teachers already trust — not the model's best guess.

Standards on top, curriculum underneath

Grounding in IM v.360 is one half of a pair. The other is standards. The Learning Commons Knowledge Graph connector gives Claude the math standards across all fifty states — plus, crucially, the competencies and learning progressions beneath each one. The curriculum supplies the how; the standards graph supplies the what and the order. Together they make a math lesson that is both grounded in trusted material and aligned to your state's target, with scaffolds placed where the progression says they belong.

That pairing is exactly what keeps AI math from the two classic failure modes at once. Curriculum grounding guards against wrong methods and planted misconceptions. Progression awareness guards against out-of-order sequencing and skipped prerequisites. Neither alone is enough in math; both together are what make a Claude-planned math lesson trustworthy on first read.

The teaching skills reflect this at the file level. Anthropic's open-sourced k12-teacher-skills repo ships a dedicated math.md reference alongside its other per-subject packs — subject-specific guidance for how Claude should reason about a math lesson, distinct from how it approaches ELA or science. Math has its own habits of rigor, and shipping a math reference is how the skills respect them. It mirrors exactly what happens for science, where OpenSciEd + Claude plays the same role: ground each subject in the curriculum its teachers actually trust, and give Claude subject-specific guidance for the reasoning.

Where the math ecosystem picks up

Grounded planning is the foundation, but math has a rich connector ecosystem in Claude for Teachers that builds on top of it — and the pieces fit together precisely because they share the same curricular ground.

Coteach generates high-quality math diagrams grounded in K-12 curriculum, so the visual — often the hardest part of a math lesson to get right — is built on the same trusted footing as the plan; that pairing gets its own piece, Coteach + Claude: K-12 Math Diagrams Grounded in Curriculum. ASSISTments generates auto-scored, standards-aligned math problems for practice and assessment. The through-line is consistency: when the lesson, the diagrams, and the practice problems are all grounded in curriculum and standards rather than improvised separately, they cohere instead of contradicting each other. That coherence is something a teacher stitching together five different generic tools rarely gets.

What it feels like to plan

In practice, the grounding is invisible until you notice its absence elsewhere. You ask Claude for Teachers for a math lesson on a topic, at your grade and state, and what comes back is drawn from IM v.360 and mapped to your standards — a solid draft to adapt, not a wildcard to verify. Follow-ups behave, too: "reteach this for students who never solidified the prerequisite," "add a worked example that heads off the usual misconception," "extend it for early finishers." Each is a modification of trusted material, so it stays mathematically honest instead of drifting the way edits to an improvised lesson do.

You're still the math teacher. You still catch what doesn't fit your room. But you're spending that judgment on pedagogy and fit, not on debugging arithmetic and method — which is where a math teacher's judgment is actually worth the most.

The takeaway

Any AI tool will write you a math lesson. The one that matters is whether it wrote from a curriculum you'd trust or from the average of everything it once absorbed — and in math, that difference is the difference between building understanding and quietly planting a misconception. Claude for Teachers plans from IM v.360, with standards and progressions from Learning Commons underneath, so the lessons stand on real ground.

To feel the principle before you're verified, grounded marketplace tools reward the same discipline: give the model real material and real standards to work from, as lesson-plan-studio and the education-agent-skills bundle do their best work when you do. Ground the math in a curriculum, and the lesson stops being a guess. It becomes something you can teach from tomorrow.

---

Part of the Claude for Teachers series. Related: Teaching Science from Trusted Curricula: OpenSciEd + Claude · Coteach + Claude: K-12 Math Diagrams Grounded in Curriculum. Browse AI tutoring skills or more builder insights.