AI Policy

Version 1.1. Effective October 7, 2026. Version 1.0, effective August 24, 2026, applies until then. The change is the Butter Health sentence in Section 3.

We use AI tools in limited, disclosed ways, and a person stays responsible for everything we publish. This policy is part of our Terms of Service, and it describes commitments we hold ourselves to rather than general intentions.

1. Which tools we use

• Claude, made by Anthropic, for drafting and editing our own written content and for building the site and its tools.
• ChatGPT, made by OpenAI, for artwork and image generation, such as event banners.
• Google Workspace, which includes AI features in email and documents. Our Workspace is covered by a business associate agreement with Google, and Workspace content is not used to train Google's general models.

We use paid business or commercial plans on these tools, under settings that exclude our content from being used to train the providers' models. If we add another tool, we will name it here before we rely on it.

2. Where we use AI

• Daily news summaries. Our daily GLP-1 news posts are drafted with AI assistance from named source articles, and a person reviews them against the sources before they are published.
• Site and community content. Some written content is drafted with AI assistance and reviewed by a person before publishing.
• Coaching guides. Guides are built with our coach, checked against published research, and reviewed by a person. Every claim links to its source.
• Artwork. Some event banners and illustrations are AI-generated. We do not generate images of real people, and we do not use AI to alter a member's photograph.
• Building the site. The website, its tools, and its automation are written with AI assistance.

3. What we do not do

• We do not put member content into AI tools. Posts, comments, direct messages, live session chat, profiles, and support emails from members are not sent to Claude, ChatGPT, or any other AI tool, whether for summarizing, moderating, drafting a reply, or anything else. The one exception is the Butter Health list: answers a member gives through its signup form, polls and surveys may be analyzed with AI tools that do not train on them, under the agreement that member accepts when joining the list. Nothing else a member shares in the community goes into an AI tool.
• We do not use member posts to train AI models, and we do not license member content to anyone for that purpose.
• We do not use AI to make decisions about you. Nothing about your membership, your access, your pricing, or a moderation outcome is decided by an automated system. A person makes those decisions.
• We do not use AI to profile you or to infer health information you have not told us.
• We do not use AI to give medical advice. Nothing on Cozy Butter is medical advice, whoever or whatever drafted it.
• We do not run an AI chatbot that answers members in the community, and we do not have an AI account posting as a person. If we ever add one, it will be labeled clearly as AI.

4. Circle's own AI features

The community runs on Circle, which offers AI features in its platform. Those features are governed by Circle's own terms and privacy policy. Circle's AI agents are turned off in our community.

5. Labeling and accuracy

AI-assisted drafts can contain errors, which is why a person reviews them against sources before publishing. Where a piece of content is substantially AI-generated rather than AI-assisted, we say so on the content itself. If you spot something wrong, email [email protected] and we will correct it and note the correction.

Amendments

Changes made after a version takes effect are listed here, newest first.

October 13, 2026 (announced September 29, 2026). AI tools and member content. From October 13, we use AI tools to help us run Cozy Butter, with things like organizing messages, posts, comments and emails that members send us, creating content, and more. We use AI providers under business terms. From that date, the first point in Section 3, which begins “We do not put member content into AI tools,” no longer applies. The tools and uses named in Sections 1 and 2 are examples, and we may use other AI tools without listing each one here. The reviewing and labeling described in Sections 2 and 5 are our general practice rather than a promise about every item, and some content, such as our daily news posts, may be published automatically. The sentence “We do not generate images of real people” now reads “We do not generate images of a real person without that person’s permission.”

6. Contact

Questions about how we use AI can go to [email protected].