← Cases
United StatesTechnology and telecom2024 – ongoing

Notability

Notability Learn, an AI study layer for student notes

Notability cut user reports of problem quiz questions by about 45% after moving its AI study tool to newer Claude models

WorkingVendor case studyVendor ↗

Problem

Most notes students take in Notability were created once and never opened again. The company wanted to turn those notes into something students would study from.

Approach

Notability built Learn on Anthropic's Claude API. It turns a student's own notes into flashcards, multiple-choice quizzes with explanations, and summaries. The first version shipped in about two months, in time for the 2024 back-to-school release. Learn runs on Claude Haiku 4.5, and quiz questions for Pro subscribers go to Claude Sonnet 4.6. The team tracks quality through thumbs up and down reports and rolls out new models gradually.

Results

about 45%drop in user reports of problem content, from 33 to roughly 18 per 10,000 questionsVendor case study[1]
90% moretime spent in Notability by Learn usersVendor case study[1]
85%share of Learn users in an April 2026 cohort who returned the following week, against a 58% baselineVendor case study[1]

Takeaway

Notability tracked a simple user-report rate for each model change. That gave it a way to tell whether a newer model improved quiz quality.

Categories

Business function
IT and R&D
AI technology
Generative AI
Evidence
Vendor case study

Sources

Source summaries

[1]Anthropic27 Aug 2026

Anthropic's customer story describes how Notability built Learn, which creates flashcards, quizzes and summaries from student notes using Claude. It reports a drop in problem-content reports from 33 to about 18 per 10,000 questions, 90% more time in the app for Learn users, and higher week-one and week-eight return rates. The page carries no date; the Internet Archive first saved it on August 27, 2026.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026
Open case page ↗