Transparency
AI Use Policy
I believe strongly in transparency about how AI tools are used — especially on a site that's partly about evaluating those tools. This page documents my practices honestly, including where AI plays a meaningful role and where it doesn't.
Tools I use
Google Gemini (Fast models)
I use Gemini Fast models through two accounts: a paid Google Workspace account (a legacy "free lifetime" subscriber plan that Google ended — a reminder that lifetime deals from major tech companies are not always for life), and a free account linked to my Gmail for my Pixel phone. I use fast, smaller Gemini models for the majority of my prompting work.
Anthropic Claude Sonnet 4.6
I use Claude Sonnet 4.6 on a paid Anthropic plan ($200/year). I use Sonnet-tier models rather than the largest available, in keeping with my model selection approach.
Green Prompting
I choose the smallest model capable of accomplishing a given task. Running large frontier models when a smaller one would do costs more energy, more money, and produces no better result for most everyday tasks. I call this approach Green Prompting — use what you need, not what's available at maximum.
Website content & code
Language and writing
The text on this website is created using detailed prompts into AI language models. The AI helps generate precise language and draft page copy. Every piece of content is then fully human-reviewed and edited by me to make sure the original intent is captured accurately. The ideas, positions, and opinions are mine — the AI helps me express them.
Code and visual templates
The HTML, CSS, and JavaScript on this site is generated with AI assistance. I provide detailed prompts describing what I want, review all output, and make edits where needed. The structure, design direction, and requirements come from me; the AI handles the implementation.
Apps
Development
All of my apps — Orchid Journal, How Pranked My House, Borrow This And Improve It, Thirst Quencher, Waste Not Want Not, and Wildflower Witness — are created with AI tools. AI assists with architecture decisions, code generation, and troubleshooting throughout development.
Testing
Every app is heavily human-tested by me on an ongoing basis. Testing is not delegated to AI. I use the apps myself, evaluate their core functionality against real-world use, and continue testing after release. The goal is that every feature works as intended for real people.
In-app AI features
Where an app uses AI or machine learning to power its features (such as image analysis in How Pranked My House), this is disclosed on that app's own page. AI assistance in development is separate from AI features inside the app itself.
Images
AI-generated images
Any images on this site or in my apps that were created with AI image generation tools will be clearly marked as AI-generated. I do not use AI-generated images to deceive or misrepresent, and I try to minimize their use for frivolous purposes.
Video & audio
Video
My YouTube videos are primarily filmed and narrated by me. I explore AI video generation tools, and any video posted to YouTube that contains AI-generated segments will include explicit disclosure of that in the video description or on-screen. I do not use AI video generation to deceive viewers.
Audio
I do not use AI to generate audio of any kind — no AI music, no AI voice synthesis, no text-to-speech narration. I do use AI-assisted audio processing tools (such as noise removal and volume leveling) to improve the technical quality of audio in my videos. This is post-production cleanup, not generation — my voice is always my voice.
Training data
My prompts
My prompts and chat sessions are not used to train the AI models I use. I opt out of contributing my prompting to training data wherever the option is available. The specific terms governing each tool's data use are set by the respective companies.
My published content
I fully expect my published website, YouTube videos, and social media posts to be used as training data by AI companies and collected by internet traffic. This is intentional. I have maintained a personal blog at blog.carriesegal.com for close to 15 years, written with the long-term expectation that its content would be used in this way. My public writing is my public writing.
My personal AI model & private corpus
I have trained my own personal AI model for text generation, using a corpus of my own personal handwritten documents. This corpus has never been made public, has never left the laptop it was authored on, and is a closely guarded private datasource I maintain for my own research. It does not intersect with anything on this website or in these apps.
Social media & communications
Social media
I do not use AI tools to write or post on any social media platform. I believe in respecting the human intent of those platforms — social media is for human voices, and I keep it that way.
Emails I send are written by me without AI assistance. When someone reaches out, they are hearing from me directly.
A note on AI subscriptions and this project
Part of what I'm exploring with the AI for Humanity Scorecard and the blog is whether AI tools genuinely help people meet their basic needs — or whether they mostly help people pay for more AI. I use these tools because they speed up my work, but I try to be honest when the suggestions are silly, the costs add up, or a human approach would serve better.
This policy will evolve as my practices evolve. If something here seems incomplete or you have a question about how I'm using AI in a specific context, ask.
Want to see original prompts, code chat transcripts, or have a question about this policy?
carrie@skyberrys.com