Writers, you now have more control over how AI companies use your stories

An update to securing consent, credit, and compensation for writers when AI companies train on your stories

When you publish a story on Medium, you probably do want it to be seen by as many readers as possible, and you probably don’t want your text gobbled up to train an AI tool in a way that doesn’t benefit you.

This was the premise that made us originally decide on our default stance on AI training: AI models should not train on your stories, unless we can secure consent (you say it’s OK), credit (you get a citation), and/or compensation (you earn money if your story is used) for our writers. We refer to these as the 3Cs.

While we continue working to make those 3Cs a reality, our current policy is that we optimize for reach. We request that AI companies not use your stories for training, unless that impacts the reach of your story.

As the AI landscape grows more complex, and as writers develop stronger opinions on what they are and aren’t comfortable with, we wanted to give you more options on how AI companies use your Medium stories.

Today, we’re adding an even stricter option that allows you to request that AI companies not train on your stories, even if it limits your reach. Check your settings to select the option you prefer, and read more details in the Help Center.

Your two options

The first option is what was the default case for all stories on Medium before today. The second is the new option that’s now available to you in the settings.

• Prioritize maximum reach (the default, unchanged from today): Ask AI companies not to train on your stories, unless doing so would also stop them from directly crediting your work and sending readers to it. Your stories still appear in traditional search and AI-assisted search features. The default option additionally means that AI products can directly credit you and link to your Medium stories. Your stories should not appear in training data unless that would prevent you from receiving credit and traffic via their AI products.
• Minimize third-party training: Ask all AI companies not to train on your stories, even if it means they may no longer directly credit your work and send readers to it. Your stories still appear in traditional search and AI-assisted search features. This option, however, means that some AI products (which don’t separate training consent from citation) are unlikely to directly credit you and link to your Medium stories. Your stories should not appear in training data.

What these options don’t control

These options don’t impact how your story shows up in search engines, on social media platforms, or on Medium. This preference is only about how third-party AI companies access your stories for training, and the impact that it has on your reach via AI products. Your stories may still appear in AI-assisted search features in traditional search engines.

Looking deeper at what each option means

For something as nuanced and important to writers as understanding AI training, we want to make every effort to communicate clearly about what we are doing, so you can make an informed decision about how we request Large Language Models (LLMs) treat your writing on Medium.

Imagine you’ve published a story about growing potatoes in buckets on Medium. When someone types “how can I grow potatoes in buckets” into an LLM chatbot, like Anthropic’s Claude AI chatbot, it could theoretically access your story on Medium in two ways. First, it will consult its training data to see what the statistically likely answer is. Second, it will also pull up relevant web results and cite your story as a place to learn more. A curious gardener could read Claude’s summary and then click into your story to read more details.

Some AI companies differentiate between those two kinds of use. We have the ability to request, “Do cite this user’s stories in your results, but don’t use them to train your model.” Today, the biggest companies that fall into this category are OpenAI (ChatGPT) and Anthropic (Claude).

Other AI companies don’t differentiate between the two kinds of use. We can only ask, “Do not train on this user’s Medium stories,” but that means some of them won’t directly cite your stories when users ask chatbots relevant queries. Microsoft (Copilot) and Google (Gemini) are the two biggest companies that currently don’t differentiate between these two kinds of use.

In the latter cases, we default to optimizing for reach. Medium writers have told us again and again that they come to Medium to reach readers — not just on Medium, but through search and increasingly, AI chatbots like Claude, Gemini, ChatGPT, and Copilot.

You now have the option to ask that those AI companies not use your text for training, even if that means they won’t directly credit your work and send readers to it.

What you can do

If you want to continue reaching as many readers as possible, you don’t need to do anything. If you’d prefer to take a stronger stance on preventing AI companies from training on your writing even if it limits your reach, update your preferences here.

And, as always, please let us know what you think. We make these policy changes for you, so if you love them, hate them, or feel any kind of way about them, we want to know. Leave a reply or raise a support ticket with your feedback.

Update:

(added March 24, and posted as a response below)

Thank you all for the responses to this post. We’re reading them and hearing your feedback, and want to address a few questions and concerns here in one place.

How content on the internet is discovered has changed dramatically, and we’re trying to make sure that your stories are still able to be seen by people searching for information.

As a public platform (and like other public platforms) we’ve always provided search companies access to content so that your stories can be show up in search results, including paywalled content (though anyone coming to a paywalled story from search would have to become a member to view it). Search crawlers have been a staple of the open internet since Yahoo! Search back in 1995, and Medium’s strong SEO authority has been a major reason writers have chosen to write on our platform since our launch in 2012.

And for years, traditional internet search was the primary way readers off the platform found Medium stories, but that changed a few years ago. With the rise of AI, like all internet platforms, we saw a decline in traffic to your stories from those traditional search companies. We have, however, seen a rise in traffic from AI search (think: when you ask a question in an LLM and it provides an answer with a link to the page it found the information on.) People are using AI products as search engines, and we want your stories to show up there as well, and that means ensuring that AI search crawlers (not training crawlers) can see stories in order to include them in their results. We don’t want to unilaterally limit your potential reach by default.

Some companies have intertwined their search crawlers with AI training though, so it’s not as easy as saying yes to search, no to training. Since we fundamentally believe in an open, searchable internet, we wanted to default to discoverability when we could. We’ve now added controls so that you have the power to decide how your content is seen by the rest of the internet.

This is a tough challenge. We recognize this update won’t satisfy everyone’s concerns. But we also want Medium to be a part of what the internet is evolving into, so this is our attempt at walking that line. Thank you for your feedback as we navigate this with you.

Regarding stories in the Partner Program: When AI products like ChatGPT try to access paywalled stories directly, they’ll see the same limited preview that a person would, and they can link to the post so that people can read it directly (assuming that they’re a paying Medium member). These products often consult search indexes also, but they’ll receive indexed snippets, not full post content, and — just like search — can use that to link to the most relevant stories.

A quick note: We’ve seen questions asking about compensation from AI companies, and that is something we continue to work towards. We’ve joined other companies in supporting an industry standard AI training — this announcement is one of many steps we’re taking to attempt to get our writers consent, credit, and compensation for their work.