We gave the same prompts to Diffit and Gemini for Edu, then scored both outputs against a 20-criterion rubric grounded in WestEd's framework for high-quality instructional materials.
CCSS.MATH.CONTENT.5.NF.A.2 — "Solve word problems involving addition and subtraction of fractions referring to the same whole, including cases of unlike denominators, e.g., by using visual fraction models or equations."
Gemini's first draft printed $\frac{1}{2}$ where a student should see ½ — and called itself "clean, ready-to-print." Diffit's came ready, with the visual fraction models the standard names and an error-analysis task that pushes students to reason, not just compute.
CCSS.ELA-LITERACY.RI.5.3 — "Explain the relationships or interactions between two or more individuals, events, ideas, or concepts in a historical, scientific, or technical text based on specific information in the text."
Same assignment: a reading packet on Harriet Tubman from a National Park Service source. Diffit built the whole lesson — a passage, standard-aligned analysis activities, and an answer key, all referencing the same text. Gemini wrote good analysis questions, but about a passage it never produced — with no answer key, and only after the teacher followed up to make it student-facing at all.
CCSS.ELA-LITERACY.RH.9-10.2 — "Determine the central ideas or information of a primary or secondary source; provide an accurate summary of how key events or ideas develop over the course of the text."
This was Gemini's strongest result — both built solid primary-source lessons. Diffit built structured analysis activities, while Gemini leaned on open questions. And when the assignment named one source, Gemini's output cited ten — Wikipedia, a federal record, even a city-government page — while Diffit drew only from the source provided.
RI.5.3), releveled to Grade 2
Going from Grade 5 to Grade 2 means simpler sentences — but Diffit keeps the named figures, the dates, and the full military-leadership section. Gemini got simpler by deleting: Edward Brodess becomes "the man who owned Harriet," and the Combahee River Raid section disappears. A relevel should simplify the language, not delete the lesson.
It's important to note that Diffit didn't score perfectly on the rubric in a few areas, too. In the 2nd-grade rewrite, one prompt oversimplified a part of the reading; a Spanish translation used two different words for the same term ("pantanos" (swamps) in the reading passage but "marismas" (marshes) in one of the analysis activities); and the high-school lesson skipped images that may have helped with scaffolding. These are noted in the full analysis below.
The difference: a teacher can fix any of these inside Diffit in seconds — swap an image, edit a word — right in the packet. In a chatbot, that means re-prompting and hoping the rest of the lesson survives.
The side-by-sides above are three moments. Here's the full tally across all 20 criteria, for all three prompts.
| Brief | Diffit | Gemini for Education |
|---|---|---|
| 5th-grade ELA · Harriet Tubman | 17 / 19 | 4 / 17 |
| 5th-grade Math · Fractions | 17 / 17 | 7 / 17 |
| 9–10 ELA · MLK Letter | 18 / 19 | 10 / 18 |
Both tools' materials were graded against the four things WestEd found teachers care about most: accuracy, standards alignment, ease of use, and meeting student needs — 20 criteria in all. Every score has cited evidence. All 11 source files and the exact prompts are available upon request, so any district can rerun the comparison.
Grounded in the WestEd framework for instructional-materials quality (Bugler et al., 2017) and the Diffit Quality Constitution.
To see the full scoring or learn more about Diffit, contact schools@diffit.me.