Ethical AI for All Learners
At Quill.org, we've built AI-powered literacy tools that help millions of students become stronger readers and writers, meeting learners where they are with tailored support.
10M+
Students Impacted
300+
AI Models Developed
300+
Teachers in Advisory Council
100K+
Responses Evaluatedby Humans Annually
Our Four Commitments to Ethical AI
These four principles guide our development process and ensure our AI tools effectively and ethically support student learning.
1. We Research First, Then Code
We ground our tools in learning science and research before any development begins. This ensures our AI is built from the ground up to best support all learners.
2. We Build Custom Datasets
Educators build the training datasets that are embedded into Quill’s AI, making them the creators, rather than the recipients, of our AI learning tools.
3. We Evaluate Early and Often
We test the efficacy and reliability of our AI tools rigorously at every stage. Continuous feedback ensures our AI remains effective and responsive to student needs.
4. We Partner with Teachers
Our AI is built with, not for, educators and the students they serve. Our AI supports—not replaces—the work of educators guiding student growth.
1. We Research First, Then Code
We shape the future of AI learning by first studying what's effective in today's classrooms.
Before writing a line of code, we collaborate with researchers and educators to precisely define what successful learning looks like by collecting and analyzing a large set of examples.
Start with a crystal-clear definition of successful learning
We establish a crystal-clear definition of success to train AI that can help students meet that mark. Our AI provides coaching that mirrors a veteran educator's feedback.
Collect an extensive set of examples that illustrate successful learning
We start by building hard-copy exercises and testing them with real students. Through this process, we've identified key principles for designing effective AI-powered learning experiences.
Define effective student prompts, responses, and high-quality feedback
We design effective writing prompts, define exemplary student responses, and deliver effective teacher feedback to guide our AI development process.
2. We Build Custom Datasets
After defining effective learning, we create specialized datasets that demonstrate it.
We ensure educators, not generic LLM training, make judgment calls about student work. We provide our AI with examples of educators' real judgments to modify its "thinking" process.
Develop custom datasets to show the AI how to evaluate student responses
For every writing prompt on Quill, we build a dataset of 40-100 examples of student responses paired with high-quality feedback written by educators.
Thick wrapper AI tools are more effective than thin wrapper tools
We create a 'thick wrapper' around the AI by providing 5,000-8,000 words of directions, sample responses, and teacher feedback for every evaluation.
Prompt engineering and fine-tuning allow you to customize and control AI
We employ both multi-shot prompting (50-100 examples per prompt) and fine-tuning to customize our AI.
3. We Evaluate Our AI Early & Often
We can't improve what we don't measure.
We know raw metrics aren’t enough to the whole story. We rigorously assess student experiences and outcomes to ensure our AI is reliable and meeting learner needs.
For every student task, develop a custom benchmark dataset
We create benchmark evaluation datasets with at least 300 manually graded responses, distinct from our training examples.
A/B testing infrastructure enables developers to know when AI has the right training data
We run multiple tests to compare different approaches and identify exactly where improvements are needed.
Continuous human evaluation powers the cycles of rapid improvement
Our ten full-time curriculum developers at Quill—primarily former classroom teachers—manually evaluate over 100,000 student responses each year.
4. We Partner With Our Teacher Advisory Council
We build effective AI by testing and iterating with our community of educators.
We make educators central partners in our development process. We collaborate with over 300 teachers in our Teacher Advisory Council to evaluate and review every AI-powered writing prompt we create.
Every student writing task goes through three rounds of distinct testing
Before releasing an activity, we share it with our Teacher Advisory Council for three distinct rounds of testing and feedback.
The council includes educators from across the United States
Our council includes teachers from 42 different states, with over 68% working in Title-1 schools.
Our team learns and improves from every testing round
Through rigorous analysis and Advisory Council feedback, we are able to improve the training data so that it is more responsive to classroom needs.
On the Future Fluent Podcast, Dive Into Quill’s Ethical AI
About the Podcast Episode:We know how to teach people to improve their writing—but it takes a lot of work. In this episode of Future Fluent, Betsy Corcoran and Jeremy Roschelle talk with Peter Gault, the founder of nonprofit Quill, which gives students feedback on 500 million sentences a year. Quill's been using AI for years and is now sharing its "playbook" on how to build ethically — and effectively — with AI.
Read Our Generative AI Playbook
After six years of building AI-powered literacy tools and developing more than 300 unique AI models, we've distilled our most valuable insights into a comprehensive playbook that shares our entire process.
For Education Leaders
As you evaluate AI solutions for your schools, this playbook gives you the critical questions to ask vendors and the frameworks to assess if their tools are truly designed for student success. You'll understand what makes AI effective in educational settings and be able to distinguish between superficial AI implementations and those with deep pedagogical foundations.
For EdTech AI Developers
Skip years of trial and error by learning from our experience. We share our step-by-step development process, technical approaches, and proven methodologies for creating AI that truly enhances learning. From dataset creation to evaluation frameworks, you'll find actionable insights that can be applied immediately to your own AI development work.
For Philanthropic Partners
When funding educational AI initiatives, it's crucial to identify organizations with the expertise and ethical approach to create meaningful impact. Our playbook provides the evaluation criteria and key indicators of effective AI implementation, helping you direct resources to organizations genuinely capable of advancing educational equity through technology.
Our biggest lesson? AI is malleable.
While out-of-the-box AI can be unpredictable, carefully annotated datasets and robust evaluation infrastructure make it possible to mold AI for effective, equitable learning.
Download Our Free Resource
Get our complete 12-page playbook with detailed insights and practical guidance.
Download PlaybookWhy this matters:
Pioneering AI Innovation for Public Good
AI technology holds immense potential to address societal challenges at scale, but it won't do it on its own—it has to be actively shaped by actors whose mandate is to put the public interest first. Mission-driven organizations are essential for creating technology that truly serves humanity's needs. We hope this playbook will inspire others to unlock visions of a positive technological future in which AI advances benefit everyone.
The educational landscape is facing unprecedented challenges. With literacy and math scores at their lowest in the past 30 years, our commitment to ethical AI development isn't just about technology. It's about creating a better educational future for all learners.
Our methodical approach means that when a student interacts with our AI, they receive immediate, meaningful feedback that mirrors what they would get from a veteran educator. This real-time coaching creates opportunities for growth, particularly for students in schools that may lack sufficient resources. Low-income, Title I schools stand to benefit most from carefully designed AI solutions. By adhering to our four key principles, we ensure our tools genuinely accelerate learning outcomes rather than creating frustrating or ineffective experiences. With educators at the center of the process, we're not just building tools—we're building bridges for better learning outcomes. When AI is developed ethically and intentionally, it becomes a powerful force for addressing educational disparities and helping every student develop the critical thinking and literacy skills they need to succeed.
For more information about our approach to ethical AI, contact us at:
EthicalAI@quill.orgAbout the Authors
Peter Gault
Executive Director
Peter has been developing AI models at Quill since 2018 working with Quill’s team of educators and technologists. His favorite part of his role is diving into new datasets to deepen the student learning experience.
Maheen Sahoo
Managing Director of Strategic Partnerships, AI for Education
Maheen Sahoo has spent the last decade working in EdTech to build education partnerships and educator communities. At Quill, she leads Quill's responsible AI strategy and builds partnerships with mission-aligned organizations.
Afua Bruce
Public Interest Technologist & Author
Afua Bruce is a technology executive and author who leverages technology to address complex societal challenges. Her work bridges technology innovation with public service, using data science and AI to create sustainable social impact solutions across sectors.