gōgy

AI Transparency: How Gogy Uses AI

Last updated: July 17, 2026

Gogy is a curriculum product. AI is one of the tools we use to build our lessons, not the product itself. This document explains how AI fits into Gogy. It is written for the teachers who use Gogy and for the school and district staff who evaluate it.

1. Students never interact with the Gogy platform

Gogy is a tool for teachers, and no student data ever enters it.

A teacher types a curriculum topic — like "the water cycle". Gogy generates a complete lesson: a student handout and a paired teacher script built on Gogy’s 12 instructional principles. The teacher reviews it, chooses what to use, and prints it. The student's only contact with Gogy is a printed packet their teacher hands them.

This design shapes how we use AI in our product. Gogy does not grade, assess, evaluate, monitor, or make any decision about any student. There is no student account, no student login, no student data, and no student surveillance.

2. How we use AI

What it does: Given a teacher's instructional topic, Gogy calls upon LLMs to generate some — but not all — text and images for a lesson. Much of the text in both the student handout and teacher script is static, written by Gogy’s curriculum team. When able, our lessons pull in real, appropriately-licensed images. We use AI to tailor content to a teacher’s topic and fill in any gaps.

What it does not do:

  • It does not interact with students.
  • It does not grade, score, assess, or make decisions about students.
  • It does not collect, store, or process student data of any kind.
  • It does not replace the teacher. The teacher reviews every lesson and leads it. The teacher script exists specifically to keep the human educator as the leader of instruction.

3. The AI we use

Gogy does not train or operate its own AI model. We use established third-party AI providers, typically accessed through a routing service:

RoleProviderWhat it does
Request routingOpenRouterRoutes most requests to the appropriate model provider
Primary text generationAnthropic (Claude)Lesson text and lesson evaluations
Supplementary generationOpenAI (GPT)Certain backend structures and lesson evaluations
Image generationGoogle (Gemini)Lesson images and layout; lesson evaluations fallback
Image matchingNomicConverts image descriptions into embeddings to choose images
AI monitoringLangfuseTrack the quality and reliability of our AI features

These are proprietary, closed-source large language models and an image-generation model, accessed by API — none are self-hosted by Gogy and none are open-source models we fine-tune. We chose external providers deliberately: it lets a small team deliver high-quality output without building or training models in-house, and it means model safety and security are maintained by organizations with dedicated safety teams.

If we change the AI models we use in a way that materially affects lesson quality or behavior, we will notify registered users and explain the change.

4. How your data flows through the system

Here is exactly what happens when you generate a lesson.

  1. Teacher enters grade, subject, duration, and types a curriculum topic (e.g., "photosynthesis")
  2. Passed to Gogy servers (DigitalOcean, US), along with the user’s state indicated in their profile
  3. Passed to OpenRouter (request routing) or Google Gemini (for images)
    • Passed to Anthropic for most lesson structure and text.
    • Passed to OpenAI for limited backend structures.
    • For image generation and image matching (Google, Nomic), what travels is a system-generated image description derived from the lesson — not the text you typed.
    • Passed to Langfuse to track the quality and reliability of outputs.
  4. Gogy servers assemble the lesson
    • Returned to the teacher (handout and script)
    • Original generated copy is stored, detached from the teacher
    • Any teacher edits to the lesson are stored in the teacher’s account only

A few specifics that evaluators ask about:

  • Only the topic inputs (grade, subject, duration, user-entered text) and the user’s state travel to the AI providers. No teacher PII (except state) and — because it never exists in the system — no student information. The input is a subject-matter prompt like "photosynthesis," not a person. The state is used to localize lesson content for relevance and engagement.
  • Provider retention and training: Our providers do not use Gogy's inputs to train their models. They either do not retain inputs or outputs (OpenRouter) or may retain inputs and outputs for a limited period for safety and abuse monitoring — currently up to 30 days (Anthropic, OpenAI) or 55 days (Google) — after which they delete them, consistent with each provider's API terms. Our embeddings provider (Nomic) retains nothing.
  • Gogy's own retention: We keep the original generated copy of each lesson, detached from the teacher's identity, and use these de-identified copies to operate and improve the service. We do not re-identify them, and we do not use teacher-edited versions for this purpose. Edited versions live only in the teacher's account.
  • Where it lives: All lesson and account data is stored in the United States.

5. Why Gogy is "directed to teachers," not "directed to children"

Because Gogy's lessons serve grades 3–8, an evaluator might initially assume the software is "directed to children" under COPPA and the FTC's 2025 amendments. It is not. Applying the FTC's multifactor "directed to children" test:

  • Intended and actual audience: Teachers. Accounts, purchasing, and all interaction with the platform are by adult educators.
  • Subject matter and interactive features: The platform is a professional lesson-preparation tool — topic entry, lesson selection, editing, downloading. There are no child-oriented interactive features, games, or characters.
  • Marketing and representations: We market Gogy to teachers, on professional and educator channels, and we represent it to third parties as a teacher tool.
  • Who actually uses it: Only registered adult educators. Students have no access path — no account, no login, no input.

The curriculum serves children; the platform is used exclusively by adults. No personal information is collected from any child because no child ever uses the Gogy platform. This is why COPPA's child-data obligations do not apply to Gogy — the conditions that trigger them never occur.

6. How we control quality

AI-generated content carries real risks — factual errors ("hallucinations"), uneven rigor, and bias. We manage these with layered controls.

6.1 Design constraints

Every lesson is designed and generated against Gogy's 12 instructional principles. These are not decorative — they constrain what the AI produces toward defined pedagogical patterns: lessons where students are talking, doing, and given a reason to care; where thinking is made visible; where instruction starts concrete and builds toward complexity. The structure is fixed by design. The AI fills it in for the requested topic.

6.2 Human and machine review

Our curriculum team reviews lesson quality on an ongoing basis. Any teacher who generates a lesson on Gogy reads it, adapts it to their needs, and decides whether to use it before anything reaches a student. We also run our lessons through proprietary machine evaluations that we use to refine our design constraints and review process.

6.3 Grade-level constraints

Teachers specify the grade band (3–8). Content is generated to that level, which constrains vocabulary, complexity, and framing to age-appropriate ranges.

6.4 Known limitations

  • Gogy is not a replacement for teacher-led instruction. It prepares materials; the teacher teaches.
  • Gogy is not designed for student self-study without a teacher.
  • Gogy is not a special-education compliance tool — it does not generate IEP-compliant or 504-compliant materials, and teachers must adapt content for students with disabilities.
  • AI-generated content can contain errors. Teachers should verify facts before classroom use. We recommend treating standards alignment as a starting point that you confirm, not a guarantee.
  • Gogy generates lesson content; it does not currently provide source citations for every factual claim. Teachers remain the verifying authority.

7. Bias and fairness

Bias in AI output is a known, ongoing risk. Our approach:

  • Representational range is built into the design. Two of our 12 principles — Value Every Student's Identity and Facilitate Open-Ended Expression — push lessons toward content that reflects a range of identities, experiences, and perspectives rather than a narrow default.
  • Explicit prompting and human review by our curriculum team checks generated content for representativeness and for stereotyping or exclusion.
  • Ongoing monitoring. Because model updates can change behavior, we treat bias review as continuous, not one-time. When we change models (Section 3), representativeness is part of what we re-check.
  • Non-discrimination. Gogy is committed to generating content that does not discriminate on the basis of race, color, national origin, sex, disability, or any other protected characteristic, consistent with Title VI of the Civil Rights Act, Title IX, Section 504 of the Rehabilitation Act, and the Americans with Disabilities Act (ADA).

We will continue to develop and publish our bias-monitoring practices as Gogy grows.

8. Reporting a problem

If a generated lesson contains an error, bias, inappropriate content, or anything else that concerns you, tell us — it is the fastest way we improve. Email [email protected] with the lesson and what you noticed. We review reports and use them to refine our design constraints and review process.

9. The framework behind our approach

Gogy's lessons are grounded pedagogically in our 12 instructional principles. Our approach to building responsibly with AI considers trustworthiness characteristics including validity and reliability, safety, accountability and transparency, and fairness — which aligns with widely recognized educator guidance on the responsible classroom use of AI.

10. Quick answers for evaluators

  • Does Gogy use AI? Yes. We generate lesson text and images from teacher-entered topics.
  • Third-party or in-house? Third-party: Anthropic, OpenAI, Google Gemini, and Nomic, typically via OpenRouter. No in-house model.
  • Open or closed source? API or self-hosted? Closed-source proprietary models, accessed by API. Not self-hosted, not fine-tuned by Gogy.
  • Does it collect student data? No. Students never interact with the Gogy platform.
  • Are teacher inputs used to train AI? No.
  • Is the model trained on student data? No. Gogy sends no student data to any provider, because we never collect it.
  • Where is data stored? United States. A monitoring provider processes limited operational logs in the EU — see our Privacy Policy.
  • Human in the loop? Yes. The teacher reviews and leads every lesson.
  • What are the limits? AI output can contain errors; teachers must review before use. Not for student self-study, not an IEP/504 compliance tool.

11. Contact us

For questions about our usage of AI, email [email protected].

© 2026 Gogy Lesson PBC

www.gogylesson.com

Sign Up for Our Newsletter

© 2026 Gogy PBC

PricingPrivacy PolicyTerms of ServiceAI Transparency
Our StoryInstructional Principles[email protected]