AI and your work
Keeping your work out of AI training
Generative AI models were trained on artists’ work, scraped without permission. Plinth sites refuse that use in every way a website can. This page covers what we do, where it stops working, and how we use AI ourselves.
What every Plinth site does
- Refuses AI crawlers in robots.txt. Every site lists the known AI-training crawlers (OpenAI’s GPTBot, Anthropic’s ClaudeBot, Common Crawl’s CCBot, ByteDance’s Bytespider, Meta’s crawler, Perplexity, Amazon, Cohere, and more) and the opt-out switches for Google’s and Apple’s AI training. Search engines can still index your site, so people can find you.
- Says “noai” on every page and every image. Pages carry
noai, noimageairobots tags; images carry the same in their headers. Dataset tools such as img2dataset skip images marked this way. - Reserves your rights under EU law. Every page and image declares a text-and-data-mining reservation using the W3C TDM Reservation Protocol, the machine-readable opt-out that EU copyright law gives rights holders.
- Turns away crawlers that say who they are. When a known AI crawler asks for a page or an image, the request is refused.
- Never publishes your originals. Visitors get resized copies, at most 2400 pixels wide, with camera, date and location data stripped. Your full-resolution file stays private.
- Says it in the footer. Every page footer states that your work may not be used to train AI or build datasets.
What no website can do
These measures work on companies and tools that follow the rules, which most of the big ones say they do. A scraper that pretends to be a normal browser can still copy any image that’s publicly visible. Right-click blockers and image overlays don’t change that. They annoy real visitors and scrapers get around them easily.
If you want protection built into the image files, the University of Chicago makes two free tools you can run before uploading:
- Glaze makes small, hard-to-see changes that make it harder for AI models to copy your style.
- Nightshade changes images so that a model trained on them without permission learns the wrong things.
Upload the protected files as they are. Plinth makes smaller copies for display and doesn’t apply filters or “enhancements”.
How we use AI ourselves
We use a language model for two things. Both are optional, and you have the final say on each.
- Suggesting series. It gets the capture date and five colors from each image (never the image itself) and suggests groups like “Spring 2026”. You can rename or rearrange all of them.
- Drafting a statement. It turns your answers to three questions into a first draft, using your words and without making up shows, influences or materials. You can edit it or write your own instead.
We don’t send your images to any AI service, we don’t use your work or your words to train models, and we don’t generate images.
Your rights
You own your work. We store and display it only to run your site. You can remove any piece, or download all of it, whenever you like.
Make your site