ChatGPT is a powerful general-purpose assistant.
Here’s why schools and districts also choose Diffit.
For supplemental materials and differentiation, districts that value accuracy, alignment, and quality turn to Diffit. To illustrate, we gave the same exact prompts to Diffit and ChatGPT, then scored both outputs against a 20-criterion rubric grounded in WestEd research.
3 prompts · 8 artifacts · 60+ scored criteria·June 2026
Evaluation Criteria:
- Classroom-ready & standards-aligned
- Complete & ready to teach
- Content integrity
- Differentiation that holds up
1. Classroom-ready & standards-aligned
Prompt: Adding and subtracting fractions with unlike denominators · Grade 5
Standard: 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 to represent the problem. Use benchmark fractions and number sense of fractions to estimate mentally and assess the reasonableness of answers."
5.NF.A.2 asks for more than correct answers — it calls for visual fraction models and for students to use benchmark fractions to estimate and check the reasonableness of their answers. Diffit renders the models and builds in the estimate-and-check step. ChatGPT's set is accurate, but it hands students plain word problems and leaves both to them.
Diffit
1Visual fraction models
2Estimate, then check
ChatGPT
1No visual models
2No estimate step
2. Complete & ready to teach
Prompt: MLK's "Letter from Birmingham Jail" · Grades 9–10 · source: Stanford King Institute
Standard: 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."
ChatGPT's question design is reasonable, but it shipped a reading-analysis worksheet with no reading. Every task says to use "the text," yet the letter is nowhere in the packet, and the questions stay generic enough to answer without it. Diffit weaves King's actual words and structure into the activities and answer key, so a teacher can hand it out as-is.
Diffit
1Source material included
2Key quotes the letter
ChatGPT
1No source material
2Generic key, no quotes
3. Content integrity
Prompt: Harriet Tubman reading packet · Grade 5 · source: National Park Service
Standard: 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."
Both packets are RI.5.3-aligned — the difference is fidelity to the source you gave it. Diffit worked the NPS biography into a specific, vivid passage — Edward Brodess, the Bucktown weight, the Combahee River Raid. ChatGPT returned a flat, generic summary and cited Wikipedia in the student passage, despite being handed the National Park Service source.
Diffit
1Specific details from provided source
2Only uses your source
ChatGPT
1Generic summary
2Cites Wikipedia
4. Differentiation that holds up
Prompt: The Grade-5 Harriet Tubman packet (RI.5.3), releveled to Grade 2
Both hit a Grade-2 reading level — but the difference is what survives the relevel. Diffit keeps the same story and analysis — Brodess, the marshland skills, the Combahee Raid, the relationship and cause-effect work — just in simpler sentences. ChatGPT cut the lesson to a generic ~130-word summary, dropped every named detail, and replaced the analysis with recall. A relevel should lower the reading level, not the content.
Diffit
1Full story kept at Grade 2
ChatGPT
1Cut to a 130-word summary
Where Diffit fell short
We scored Diffit's own work critically, too.
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.
At a glance
Scored against 20 quality criteria, on every prompt.
The side-by-sides above are just examples. Here's the full tally — every score is backed by cited evidence in the full evaluation.
| Prompt | Diffit | ChatGPT |
|---|
| 5th-grade ELA · Harriet Tubman |
17 / 19 |
16 / 20 |
| 5th-grade Math · Fractions |
17 / 17 |
14 / 18 |
| 9–10 ELA · MLK Letter |
18 / 19 |
9 / 19 |
This isn't just our opinion
The rubric comes from research, not from us.
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 source files and the exact prompts are available upon request, so any district can rerun the comparison.
Grounded in WestEd's study, "How Teachers Judge the Quality of Instructional Materials: Selecting Instructional Materials, Brief 1 – Quality (Bugler et al., 2017)", and the Diffit Quality Constitution.
To see the full scoring, read the Diffit Quality Constitution, or learn more about Diffit, contact schools@diffit.me.