Can you stop AI companies training on your work?

Many makers ask the same question. "Can I stop AI companies learning from my work?" The honest answer has three parts: what the law says, what signals a website can send, and what nobody can promise.

This page is general information about the law in the United Kingdom. It is not legal advice. It covers the UK first. Whether a particular use of your work by an AI company is lawful is a question for a solicitor.

How Bernard helps

• Today, you can see who comes. Open Reports and press AI traffic. It shows which AI companies' crawlers visit your site and say who they are, including the ones that ask for your images, and Common Crawl's CCBot (is AI reading my site?).
• Today, you set your choice. Your media library has a Protect your work panel with two switches: one for your images and video, one for the words on your pages. While the images switch is on, the photos you have said are yours carry a "not for AI training" note inside the file (can I stop AI companies training on my photos and video?).
• Later: when a switch is on, Bernard will tell AI training crawlers to stay out in your site's robots.txt file, and then refuse the training crawlers that say who they are. Search engines and the AI assistants that send you customers stay welcome.

None of this can stop a crawler that lies about who it is, and Bernard does not say it does. The rest of this page explains why, starting with the law.

What UK law says in 2026

AI companies build their systems from large collections of text and images. Copying a work without permission can infringe copyright. The question is whether an exception covers the copying.

• The UK has one exception for this kind of copying. It lets a person with lawful access make copies for computational analysis for the sole purpose of non-commercial research. UK IPO guidance says the same: the exception "only permits the making of copies for the purpose of text and data mining for non-commercial research". It does not cover commercial use.
• The government looked at a wider exception and stepped back. On 18 March 2026 it published a report on copyright and artificial intelligence. It says that "a broad copyright exception with opt-out is no longer the government's preferred way forward". It also says: "We will not introduce reforms to copyright law until we are confident that they will meet our objectives for the economy and UK citizens."
• So the law has not changed. Today the UK has no legal opt-out for you to switch on. When you tell AI companies "no", you make a clear statement of your choice. It is not a legal right that UK law creates.

The courts are also still working through cases about AI training. The law is not settled.

The signals a website can send

AI companies and standards groups have created several ways for a website to say "do not train on this". Each one is a request. None of them is a lock.

• robots.txt. This is a file on every website that tells crawlers which pages they may visit. AI companies give their crawlers names, and some use different crawlers for different jobs. Anthropic, for example, names separate crawlers for training, search and a user's request and says each one respects robots.txt. So a site can refuse the training crawler and still welcome the search crawler.
• Content Signals. Cloudflare's Content Signals add a line to robots.txt that states separate choices for search, live AI answers and AI training. Cloudflare describes them as preferences, not technical enforcement.
• A rights field inside each image. The IPTC photo metadata standard has a Data Mining field. One of its values is "Prohibited for AI/ML training". The choice then travels with the file.
• A reservation on each response. A W3C community group report, TDMRep, describes a way for a website to reserve text and data mining rights in a file or in the reply to each request. It is a community report, not a W3C standard.

We did not find a published list of AI companies that read the image field, Content Signals or TDMRep. robots.txt is the signal that AI companies' own crawler pages, such as Anthropic's, say they respect.

What no website can honestly promise

Every one of these signals works only on a crawler that chooses to obey it.

A crawler can give a false name. In 2025 Cloudflare reported that Perplexity used undeclared crawlers to get around robots.txt blocks. A person can also save a picture and put it somewhere else. Once a copy is off your site, your website cannot control it.

So be careful with any service that says it "stops AI" or "locks" your images. Why no website can lock your images, and what works instead explains why that cannot be true on an open website.

A choice, not a default

Saying "no" to AI training has a cost for some businesses. A teacher who wants AI assistants to recommend her courses may want them to read her pages. A potter who sells her own work may want to reserve her photos and keep her shop open to search. These are different choices, and both are reasonable.

Search engines and the AI assistants that send customers are a separate question from training. A good setup keeps them welcome.

That is why Bernard's Protect your work panel has two separate switches, one for your images and video and one for your page text, and why neither switch touches search engines. The AI traffic report already shows the three kinds of AI visit apart: reading to answer a person's question, reading to keep AI search up to date, and reading to learn from. So you can see what each choice would affect before you make it.

When all of it is live, the honest promise is this. We say no in every way the law and the big AI companies recognise, we turn away the robots that admit who they are, and we keep the evidence. Hey AI, you can show me in search results, but don't steal my work describes the full plan.

Your next step in Bernard

Open Reports and press AI traffic, or ask your assistant: "Which AI companies read my site this month, and which of them asked for my images?" The answer tells you whether training crawlers are already coming, and how much of your AI traffic comes from real people asking questions.

Bernard is an AI-assisted website and marketing platform for small businesses. He builds a new site or moves an existing one, keeps it fast and readable to search and AI tools, and looks after it. Once measuring is on, he reads your numbers each week and suggests what to do next, with a button that does it. On the Nautilus Medium and Nautilus Large plans he also plans your marketing every six weeks and drafts your posts, emails and pages. Everything lands on a draft you approve. One flat price per site.

---

See what Bernard would do with your business. Already have a website? Paste its address and he measures it up, free, and shows you the new place before you decide anything. Starting from nothing? Tell him what you are building.