How to Protect Your Writing from AI Scraping
Updated October 2026. This post first ran in October 2025; the facts, lawsuits, bot names and platform settings below have been checked and revised.
You’ve spent months on your manuscript, maybe years. You rewrote sentences until they sounded right and built your characters from nothing. Then you find out the whole book was fed into an AI training dataset, and nobody asked you or paid you a cent.
Sounds like a nightmare, doesn't it? Maybe it sounds like something that could not possibly happen.
Well, it has already happened to a lot of authors, and at least one court has weighed in. In 2025, a US federal judge found that Anthropic had downloaded millions of pirated books to build a permanent digital library, and that this was not fair use. The case, Bartz v. Anthropic, settled for US$1.5 billion and covered close to half a million books.
The numbers get hard to picture fast. GPT-3, the predecessor to the model behind ChatGPT, was trained on about 300 billion tokens. Newer models use far more. Meta trained its Llama 3 models on more than 15 trillion tokens, or around 11 trillion words by a common estimate. Write 1,000 words a day, every day, and you’d finish 11 trillion words in about 30 million years...
So where did all those words come from?
Where the Training Data Came From
A lot of the books inside AI training datasets were pirated. Writers have spent years fighting shadow libraries like Library Genesis (LibGen), and those same libraries, along with datasets built from them like Books3, ended up feeding the systems now being sold as replacements for human authors.
The authors’ case against Anthropic is the furthest along. In June 2025, Judge William Alsup ruled that training on books Anthropic had bought legally could count as fair use. Keeping pirated copies in a permanent library could not. Anthropic settled, and the court gave final approval on July 20, 2026. Authors get roughly US$3,000 per book before fees, paid in instalments. For trade and university press titles, the money is split 50-50 between author and publisher by default.
Authors lost the first round against Meta. In Kadrey v. Meta, the company admitted it downloaded books from LibGen and other pirate sites over BitTorrent and used them to train its Llama models. In June 2025, Judge Vince Chhabria ruled for Meta on the named authors’ training claim. He went out of his way to say the plaintiffs lost because they made the wrong arguments, and that his decision said nothing about whether Meta’s use of the books was legal. A separate claim, that Meta shared pirated files with others while torrenting them, is still before the court, with a hearing set for early 2027.
The OpenAI case is still waiting on a ruling. The Authors Guild and 17 authors, including John Grisham, George R.R. Martin and Jonathan Franzen, sued OpenAI in September 2023 and later added Microsoft as a defendant. In April 2025, their case was folded into one New York proceeding before Judge Sidney Stein, along with the New York Times lawsuit and other suits from authors and publishers. On September 4, 2026, the authors asked the court to find that OpenAI and Microsoft infringed their copyrights and to throw out the fair use defence. OpenAI filed its own motion the same day arguing its use was fair. A decision is expected in early 2027.
And this isn't just about US authors. Canada has a fight of its own. In November 2024, the Toronto Star, Postmedia, the Globe and Mail, The Canadian Press and CBC/Radio-Canada sued OpenAI in Ontario. OpenAI argued the case belonged in a US court. In November 2025, the Ontario Superior Court of Justice disagreed and ruled that it has jurisdiction. OpenAI has appealed, but for now the case is going ahead here.
Want to know if your own book got swept up? In 2025, The Atlantic published a searchable database of LibGen titles. A match tells you your book was in the pirate library, though it can’t tell you which companies used it. The Anthropic settlement site lists every book covered by that deal. Only books registered with the US Copyright Office qualified, and the deadline to claim has passed.
The Authors Guild's Response
The Authors Guild has taken a firm stance against unlicensed training. Its statement on the issue reads: "Generative technologies built illegally on vast amounts of copyrighted works without licenses—that is, without giving authors any compensation or control over the use of their works—are used to cheaply and easily produce works that compete with and displace human-authored books, journalism, and other works."
In other words, your own writing is being used to build the products that compete with you.
In January 2025, the Guild launched its Human Authored certification for members, a seal that tells readers a book was written by a person. In March 2026, it opened the program to any author with books published in the United States. Guild members certify for free, everyone else pays US$10 per title, and each certified book gets a number in a public database.
More than 3,000 authors had certified 5,000 titles by the time it opened up. The whole thing runs on trust, with authors verifying their identity and signing a licence agreement stating the book is human-written. The Guild doesn’t screen manuscripts for AI before certifying them. If you’re Canadian, there’s a catch: right now you only qualify if your book was published in the US.
Why AI Companies Want Good Writing
In 2016, Microsoft released a chatbot called Tay that learned from conversations on Twitter. Within a day, users had taught it to post offensive messages, and Microsoft pulled it offline. It was an early and very public lesson that what goes into a model shapes what comes out (duh).
AI developers today take that lesson seriously. When they train their models, they filter and weight their data toward edited, published writing like novels, journalism, scientific papers and carefully written blog posts. Writing like yours, basically. The years you spent learning your craft and finding your voice are exactly what these companies want.
Books were only part of it. Common Crawl, a nonprofit, has spent more than a decade archiving huge numbers of web pages for researchers. OpenAI and other AI developers trained on those archives, which saved them from crawling the web themselves. For GPT-3, filtered Common Crawl data was the biggest single part of the training mix. If you’ve ever posted writing online, there’s a good chance some of it is in there.
Meanwhile, the money keeps climbing. OpenAI brought in about US$3.7 billion in 2024. Its 2025 revenue was reported at around US$13 billion, and its chief financial officer said annualized revenue had passed US$20 billion by the end of that year.
Next to numbers like that, what writers have been paid looks tiny. Authors in the Anthropic settlement should see their first payments in November. A small number of others have earned something through licensing deals, like the one HarperCollins offered its authors in 2024.
Most of the authors whose books built these systems have been paid nothing.
Practical Steps to Protect Your Writing
1. Add a "No AI training" notice
If you self-publish or control your copyright page, add a notice that says plainly you don’t allow AI training. The Authors Guild offers this sample wording:
"NO AI TRAINING: Without in any way limiting the author’s [and publisher’s] exclusive rights under copyright, any use of this publication to “train” generative artificial intelligence (AI) technologies to generate text is expressly prohibited. The author reserves all rights to license uses of this work for generative AI training and development of machine learning language models."
A scraper can still ignore it, but the notice puts your intent in writing, and that could help if you ever take legal action. It works a bit like a “No Trespassing” sign on your property. If you publish online, you can put the same notice at the top of your posts.
If you’re traditionally published, ask your publisher whether they include something similar. Penguin Random House, for one, now prints a no-AI-training statement on its copyright pages. The Guild also has model contract clauses that stop publishers from using or licensing your book for AI training without your written permission. They’re worth raising with your agent or publisher.
2. Block AI crawlers with robots.txt
A robots.txt file is a short text file at the root of your website that tells crawlers which pages they can visit. Crawlers don’t have to obey it, and some don’t, but the major AI companies say theirs do.
AI crawlers now come in three kinds, and you can decide on each one separately:
Training crawlers collect pages to train future models. For most writers, these are the ones to block.
Search crawlers index pages so AI search tools can show and link to them. If you block these, you won’t show up in AI answers or citations, and you may lose some readers.
User-requested fetchers load a page when someone asks a chatbot to read it. OpenAI and Perplexity both say robots.txt rules may not apply here, since a person made the request.
Anthropic has retired its old “Claude-Web” and “anthropic-ai” names and now uses ClaudeBot, and both OpenAI and Anthropic have added separate search bots. This list covers the major AI companies:
# AI training crawlers
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: Applebot-Extended
Disallow: /
User-agent: meta-externalagent
Disallow: /
User-agent: Bytespider
Disallow: /
# AI search and user-requested fetchers (optional)
# Remove this section if you want to appear in AI search answers
User-agent: OAI-SearchBot
Disallow: /
User-agent: ChatGPT-User
Disallow: /
User-agent: Claude-SearchBot
Disallow: /
User-agent: Claude-User
Disallow: /
User-agent: PerplexityBot
Disallow: /
User-agent: Perplexity-User
Disallow: /Blocking Google-Extended keeps your pages out of Gemini training, and you’ll still appear in regular Google Search. You’ll also still appear in Google’s AI Overviews, which use the regular search crawler. Annoyingly, the only way to stay out of those is to leave Google Search altogether.
CCBot is Common Crawl’s crawler. Blocking it keeps your new work out of future archives, but Common Crawl has said it can’t remove pages it has already captured.
On WordPress with Yoast SEO, you can edit robots.txt directly under Yoast SEO > Tools > File editor. Yoast SEO Premium also has a Bot Blocker under Settings > Advanced > Crawl optimization, with on/off switches for GPTBot, CCBot and Google-Extended. The WP Robots Txt plugin is another option. If you’re on WordPress.com, turn on “Prevent third-party sharing” in your site’s privacy settings.
Squarespace has a toggle for AI crawlers under Settings > Crawlers. You can’t edit the file itself there, so the list above won’t help, and Squarespace’s own list may miss some bots.
Wix has a robots.txt editor built into its SEO settings.
GoDaddy is less clear-cut. Whether you can edit robots.txt depends on your plan, so look in your SEO settings first and ask support if you can’t find it.
Sites that run through Cloudflare have their own AI crawler controls, which are worth a look. Cloudflare started blocking AI crawlers by default for new domains in July 2025. As of September 15, 2026, it also blocks AI training and agent crawlers by default on ad-supported pages for new sites and free-plan sites. If your site has no ads, that default doesn’t apply, so you’ll need to turn blocking on yourself. Cloudflare is also testing ways for site owners to charge AI companies for access.
Got FTP access to your site files? Create or edit robots.txt in your root directory and paste in the lines above.
3. Opt out on the platforms where you publish
Many sites where writers share their work now have AI settings, and they tend to be buried. Check every account you have.
Substack has a setting under Settings > Privacy called “Tell AI tools not to train their models on your content.” Substack warns that it only works with tools that respect it, and that your posts may show up less often in AI search.
Tumblr users can turn on “Prevent third-party sharing” in their blog’s visibility settings.
DeviantArt marks all work as off-limits to third-party AI datasets unless you opt in. Your AI labels and defaults are in Studio if you want to double-check.
Medium asks AI crawlers to stay away through its robots.txt file, and its terms of service have banned scraping without consent since 2023. You don’t have anything to change, though the block only works on crawlers that choose to honour it.
LinkedIn started training on Canadian members’ profiles and public posts by default in November 2025, and it may share that data with Microsoft. You can switch it off under Settings & Privacy > Data privacy > Data for Generative AI Improvement, but opting out only covers data from that point on.
X (formerly Twitter) lets you turn off data sharing for Grok training under Settings and privacy > Privacy and safety > Grok & Third-party Collaborators. Like LinkedIn’s setting, it only affects future training.
Can’t find a setting? Write to the platform’s support team and ask them directly how they use your content for AI training and how to opt out. They owe you a straight answer.
4. Be strategic about what you share online
Be picky about which platforms get your complete work. Where a site still allows AI scraping, post a short excerpt or a summary there and link to the full piece on your own site, where you control the robots.txt file.
Your best work can sit behind a paywall, or in a format that’s a pain to scrape. Someone determined will still get it. At least you’ve made them work for it.
And if you're posting artwork alongside your writing, be sure to watermark it and upload lower-res versions.
Artists should also know about Glaze and Nightshade, two free tools from the University of Chicago. Glaze makes small changes to an image that help stop AI from copying your style. Nightshade goes further and alters images in ways that confuse any model trained on them. They won’t protect your work completely, though they make it a lot less useful to anyone who takes it.
5. Monitor your content
Set up Google Alerts for your book titles, character names and any distinctive phrases. It’ll miss things, but it can tell you when your writing pops up somewhere it shouldn’t.
If you find unauthorized copies, move fast. The process depends on the country where the site is hosted. For sites hosted in the United States, send a DMCA takedown notice to the host, which has to take the material down promptly to keep its legal protection.
Canadian authors can use this process too. For sites hosted in Canada, our notice-and-notice system applies. You send your notice to the internet provider or host, which passes it on to the person responsible, and nothing gets removed automatically. Canadian notices also can’t include demands for payment. Many large platforms have their own takedown forms as well. Takedowns can’t undo training on data that’s already been collected, though they can help limit future scraping.
Not sure how takedown notices work? Stygian Media can help. I’ve filed plenty of them myself, and I’m happy to walk you through it. Email me at minion@stygiansociety.com
A Word About Copyright Law
Copyright already gives you the exclusive right to control how your work is copied, distributed and adapted. There are exceptions, like fair use in the US and fair dealing in Canada. But as the Authors Guild points out, plenty of tech companies have acted as if anything online is free to take until someone objects. They’ll honour an opt-out if you push, but they rarely ask permission first.
US courts are still working this out, and as the Anthropic and Meta rulings above show, the first decisions disagree.
Canada’s debate is happening in Parliament. The federal government consulted the public on AI and copyright in 2023 and 2024, and in April 2026 a House of Commons committee recommended requiring consent before copyrighted works can be used for AI training. The Copyright Act still hasn’t been updated to deal with AI training, though, and there’s no binding federal AI law yet.
(This is general information. If you’re dealing with a specific situation, talk to a lawyer.)
The Bigger Picture
Right now, the only way to fully protect your writing from AI scraping is to keep it in a notebook and never publish it, and that’s no way to build a career. What you can do is make your work harder to take. Every step above adds another barrier, and taken together they create a record that you objected. That record could matter as the law catches up, and it’s still being written.
The Authors Guild and other groups are pushing for legislation that would require consent before books can be used for training. Meanwhile, the lawsuits against OpenAI, Meta and others are working their way through the courts, and those rulings, plus the appeals that follow, will help decide what AI companies are allowed to do with your work.
What the Industry Predicts
When BookBub asked 18 publishing leaders for their 2025 predictions, bestselling author Craig Martelle gave one that’s stuck with us: “Old-school authors will dip and then rise anew in popularity as people realize that the human touch is critical for a storyline that reaches their soul.”
There are signs he’s onto something. All those authors rushing to get the Human Authored seal suggest writers believe readers care who wrote what they’re reading.
That demand only helps writers if the work stays theirs. So far, the worst outcome for an AI company caught training on pirated books has been a settlement. Until that changes, the book you’re writing now could end up in the next model too.
Find Your People
When we founded The Stygian Society (now Stygian Media), it was a “by authors, for authors” project that put human creativity first. That’s still true.
In 2024, NaNoWriMo announced it would not condemn AI-assisted writing, saying that doing so had classist and ableist undertones. I was a NaNoWriMo municipal liaison at the time, so I resigned in protest and started The Order of the Written Word, a writing challenge for stories written start to finish by human hands, without a single word lifted from a machine trained on stolen books.
When the NaNoWriMo nonprofit announced in March 2025 that it was shutting down, citing years of financial trouble after months of backlash over its AI stance and its handling of forum moderation, I wasn’t surprised, and I doubt anyone else was either.
Since then, the Order has grown into a Discord community of more than 780 members who cheer each other on and keep each other writing. This year’s AI-free challenge starts November 1, so if you want to spend the month writing alongside people who care about human storytelling, come join us.
Human stories carry lived experience and cultural knowledge, along with the emotional weight that comes from actually living a life. You can’t get any of that from training data.
Taking Action on Your Creative Work
Your stories hold a piece of your life, and nobody should get to take them without your permission. Working through this guide is how you let AI companies know writers are paying attention.
Short on time this week? Start with these:
Paste the bot list above into your robots.txt file.
Put a “No AI training” notice in your next book or post.
Go through the AI settings on each platform you publish on.
Look up your books in the LibGen database and the Anthropic settlement works list.
At Stygian Media, we’re here for writers who won’t let their work be treated as free training data. You spent years on your book. An afternoon protecting it is a pretty good trade, and if you’d like a hand with any of this, get in touch.
Holly Rhiannon is the founder and CEO of Stygian Media, a Montreal horror media company that publishes Canadian dark fiction and produces the Haunted Montreal paranormal series. She writes horror and fantasy, hosts on camera, and will defend a good B-movie to anyone who asks.
Interested in writing for The Stygian Blog? We welcome submissions on any aspect of horror, from fiction and filmmaking to art, theory, and commentary on the darker corners of imagination. If you have an idea or a piece you’d like to share, email us at minion@stygiansociety.com.
Comments