Writing Assistant Disclosure

AI Nutrition Facts  McGraw Hill Writing Assistant
Description Writing Assistant allows learners to ask for assistance with their short-response writing assignment and receive AI-generated personalized feedback aligned to the text and scoring rubric for the student's assignment.
Privacy Ladder Level	1
Feature is Optional	Yes
Model Type	Generative
Base Model	OpenAI – GPT-4.o (subject to change)

Trust Ingredients
Base Model Trained with Customer Data Writing Assistant is not trained on and does not have access to any user data or PII.	No
Customer Data is Shared with Model Vendor Writing Assistant runs on a private instance of base model that does not send data back to model vendor.	No
Training Data Anonymized	N/A
Data Deletion McGraw Hill's engineering team will keep a secure record of all user interactions to monitor performance. Data will be discarded after 1 year.	Yes
Human in the Loop While there are a number of safety and accuracy guardrails in place, learners receive instant output from the AI model in real-time conversation. For final feedback, teachers may choose to review the AI-generated feedback before sending it to the student.	Yes
Data Retention	1-year
Compliance Logging & Auditing McGraw Hill systematically reviews records of model input/output to audit performance. Model upgrades will be made as needed. Guardrails Writing Assistant employs input and output guardrails. It does not engage in dialogue with users when offered biased or harmful language or course content. The model has prompt-level instructions to ensure that output is grounded, relevant, and appropriate for the given learning context..	Yes
Input/Output Consistency	Yes
Other Resources Ask your Customer Support Representative for more information about Writing Assistant.

Writing Assistant Model Overview

Writing Assistant is built using a large language model (LLM). This LLM is a private instance of OpenAI GPT-4.o provided via Microsoft Azure AI. This model is given context for the specified article (and only that article) using a Retrieval Augmented Generation (RAG) pattern, which indexes McGraw Hill’s content.

Data Privacy and Security

McGraw Hill takes matters of security and bias very seriously, and we have performed extensive testing and monitoring to ensure Writing Assistant meets our high standards for educational use. We designed Writing Assistant to be secure in design and to minimize bias, inaccuracies, and inappropriateness in responses.

No Access to PII: Writing Assistant’s model does not have access to any personally identifiable information (PII) or specific user data.
No Data Sharing: Writing Assistant does not send data back to the model vendor for model training purposes.
Secure Data Handling: Our secure system records all model inputs and outputs for product improvement and model evaluations. Data will be discarded each year.
Bias, Accuracy, and Appropriateness Guardrails: Underlying each “action” (e.g., ask writing assistant) is a proprietary, lengthy prompt. These prompts have been designed to minimize potential biases, inaccuracies, or inappropriateness in responses. While McGraw Hill is dedicated to offering safest-in-class AI solutions for education, AI might occasionally produce biased or inaccurate information, and users must use critical thinking to evaluate model output as AI makes mistakes.

Continuous Improvement

McGraw Hill’s team is dedicated to the continuous improvement of our products, including Writing Assistant. To better serve learners, our team will systematically review deidentified model data and reserves the right to make changes to the underlying model as needed. This ongoing process ensures that Writing Assistant improves as a reliable and effective tool for education.

Teacher Choice

While AI technologies, like Writing Assistant, present many exciting opportunities within education, we want to ensure the choice to use these technologies remains firmly in the hands of teachers and administrators. The Writing Assistant can be deactivated by the district administrator at the district, school, or class level. A teacher can deactivate the Writing Assistant at the class level. The teacher can also choose to turn the auto-feedback feature off and on at the class level.