Everything you have published is being read by machines. AI companies crawl it, models are trained on it, and assistants summarise it for people who never visit your page. For most creators this has been happening for years, at no charge, with no record of permission having been asked or given.
During 2025 and 2026 the tools to charge for it arrived. Almost no individual creator has set them up, partly because the subject is discussed as though it only concerns large publishers. Setting it up takes an afternoon. Whether it pays you anything soon is a separate question. Probably not much.
This page covers both halves: the AI licensing part, which is mostly about position rather than income, and the older kinds of licensing that pay money now and that most people give away for free without noticing.
The size of the AI licensing market#
Content licensing deals with AI companies went from none in 2022 to twelve in 2023, twenty-eight in 2024 and around thirty-six in 2026. The largest publicly reported figure is OpenAI’s arrangement with News Corp at roughly $250 million over five years. Reddit earns about $130 million a year licensing to Google and OpenAI. OpenAI has by far the most agreements, around twenty-four publicly announced.
A pattern worth noting: 2026 deals increasingly rent a live, continuously updated feed rather than selling a static archive once. AI companies want ongoing access with attribution and links, which turns licensing into a subscription rather than an asset sale.
None of those deals are available to you. They go to organisations with large archives, legal departments and something a lab actually needs. Anyone implying an individual creator can negotiate with a frontier lab is selling something, and the something is usually a course.
Sources: Media and the Machine, LLM Pulse deal map.
Decide this before you touch anything#
There is a trade-off at the centre of this and it is easy to get wrong by reflex.
Blocking AI crawlers protects your work from being trained on. It also removes you from AI-generated answers, which is now a real discovery channel: around 41% of people rely on AI summaries rather than clicking through, and among 18 to 24 year olds, 66% use ChatGPT to find information. If assistants cannot read you, they cannot cite you.
So there are three defensible positions and you should pick one deliberately:
Open with attribution. Crawlers may use your work provided they credit and link. You keep the discovery channel and you have stated terms rather than granting silence. For most creators building an audience, this is the right answer.
Open for answering, closed for training. RSL 1.0 distinguishes between usage categories, training input, indexing and general AI use so you can permit the kind that sends people to you and refuse the kind that absorbs you. This is the position most working creators actually want, and it only became expressible recently.
Closed and priced. No access without payment. Appropriate if you have a substantial archive and no need for AI-mediated discovery. Costly if you are still growing.
Having no position at all is the default, and in practice it reads as consent.
Setting it up#
RSL, Really Simple Licensing#
An open standard launched in September 2025, managed by the RSL Collective, a non-profit whose founders include one of the co-creators of RSS. It lets you publish machine-readable licence terms alongside your content: free, attribution required, subscription, pay-per-crawl, or pay-per-inference. Reddit, Yahoo, Medium and Quora have backed it.
The mechanism is deliberately mundane.
One. Put a file called license.xml in your site root. It is an XML document using the <rsl> root element and the RSL namespace, declaring what may be done with your content and on what terms. The specification includes usage categories such as ai-all, ai-input and ai-index, which is how you separate “may be used to answer questions about me” from “may be used as training data”. Templates for the common positions, prohibit AI training, require attribution, require a subscription, are published with the spec rather than something you need to draft.
Two. Add a line to your robots.txt pointing at it:
License: https://yoursite.com/license.xmlThe directive is global, must be an absolute URL including the protocol, and instructs crawlers to fetch and comply with the terms before processing anything on the site.
Three, optional. RSL also integrates with HTTP headers, RSS feeds and HTML elements, which matters if your work is distributed as a feed rather than only as pages.
On WordPress there is a community plugin that generates and serves the file for you, which is the shortest route if you would rather not edit XML by hand.
The limitation: compliance is voluntary. RSL states your terms clearly and a crawler that ignores them faces no technical obstacle. What it does establish is a dated, machine-readable record that permission was never granted, which is what matters if the question is ever litigated, and litigation in this area is no longer hypothetical.
Full documentation, including the file format and robots.txt directive, is at rslstandard.org.
Pay-per-crawl#
Cloudflare’s approach, which enforces rather than declares. A crawler requesting your content receives an HTTP 402 Payment Required response with your price attached, and either pays or is blocked. It does not rely on robots.txt and it does not rely on good behaviour. Stack Overflow adopted it during 2026.
If your site already sits behind Cloudflare, this is a setting rather than a project. If it does not, moving a site behind Cloudflare for this reason alone is probably premature unless you have an archive worth defending.
The two approaches are complementary rather than competing: RSL states the terms, pay-per-crawl enforces a price. Robots.txt on its own does neither, it is a request and always was.
Sources: RSL getting started guide, RSL robots.txt directive, RSL overview, The Register, Stack Overflow on pay-per-crawl. Checked 7 September 2026.
What the AI part pays#
Very little, for an individual, at the moment. There is no clearing house paying small creators per crawl and the buyers are pursuing large archives. Better to say so than imply a revenue stream that does not exist.
The reasons to do it anyway are cumulative. You put on record that terms were stated and not agreed to. You keep the option open, because a licensing market that develops will need people who already have terms published rather than people starting from scratch. And there is a collective argument: standards like this acquire force only if enough sites adopt them, which is why RSL is run by a non-profit rather than a broker.
An afternoon, once, for a position you would want to have.
The licensing that pays now#
This part is older, less discussed, and considerably more likely to put money in your account this year. Most of it is money creators already have available and decline without realising.
Permission requests. A company wants to use your video in internal training, a conference deck, an onboarding course, a client presentation. This happens more often than people expect and the reflex answer is a delighted yes. The correct answer is a fee. A single-use internal licence for a corporate audience is not a favour you owe anyone, and companies have budgets for exactly this, they are asking because their legal team told them to.
Syndication and reprints. A publication wants to republish something you wrote, or re-edit it for their audience. Standard practice is a fee, and the fee is negotiable upwards if they want exclusivity or a byline arrangement. “We’ll credit you and link back” is payment only if you are early enough that the link is worth more than the money, which does happen, and lasts a shorter time than most people think.
Stock footage and photography. Material you shot anyway, sitting on a drive. Not large money and it is entirely passive once uploaded.
Music and sound. If you produce audio, sample packs and library placement work the same way.
Translation rights. Someone wants to run your guide in another language. Same structure as syndication.
None of this is transformative income. It is the kind that arrives a few times a year from work already done, and the reason it does not arrive currently is usually that there is nowhere on your site indicating it is possible.
Put up a licensing page#
Put one up. It costs nothing and it converts requests you are currently answering ad hoc into a priced conversation. Something close to:
Licensing and permissions
Most of what is here can be quoted freely with a link and attribution. If you want to do more than that, the terms are below and I answer quickly.
Republishing a full article — from $150 for a non-commercial publication, from $400 commercial. Includes a canonical link back.
Translation — priced per piece, generally the same as republishing.
Internal corporate use (training materials, decks, onboarding), from $250 per asset, single organisation, two years.
Footage or images — from $80 per clip for editorial use; commercial and advertising use quoted separately.
Teaching and non-profit use — free. Just tell me where it is going.Automated access and AI usage terms are published in machine-readable form at /license.xml.
Requests: [email]
The numbers matter less than their existence. A page like this does three things: it makes the free uses free and unambiguous, it establishes that other uses have a price, and it means a company that asks receives an invoice rather than an apology.
The legal backdrop#
Whether training on published work without permission is lawful is being litigated at scale. Trackers follow more than 160 active AI copyright cases across US, UK and EU courts. Anthropic settled with authors for $1.5 billion. In music, Warner settled with Suno and Universal with Udio, establishing an early licensing template, while Sony continues to litigate the core fair use question. In July 2026 a Munich court rejected a fair use defence in GEMA’s case against Suno.
Two things follow for you. Courts have increasingly focused on how data was obtained, whether it was pirated, whether contractual terms were breached, rather than only on the abstract fairness of training. That makes published, machine-readable terms more useful than they were, because they turn an ambiguous silence into an explicit refusal.
And the direction of travel is towards permission and payment rather than away from it. Being on the right side of that shift costs you an afternoon.
Court outcomes change and this describes the position as at 7 September 2026. It is not legal advice. If you have an actual dispute you need a solicitor, not a website.
Where people go wrong#
Blocking everything by reflex. The most common mistake, and it quietly costs you a discovery channel that is growing while search shrinks. Decide the position deliberately.
Treating robots.txt as a licence. It is a request with no legal weight and no enforcement. RSL states terms; pay-per-crawl enforces a price; robots.txt does neither.
Granting AI training rights inside a brand contract. Contracts increasingly include the right to use delivered content as training data, which is a licence you are giving away for free inside a deal about something else entirely. Contract clauses that cost you money has the wording to look for.
Saying yes for nothing. When a company asks permission, the answer can be a number. Most creators do not realise they are in a negotiation.
Having no archive of your own. If everything you have made lives inside someone else’s app, none of this applies. You cannot set crawler terms on YouTube’s behalf, and you have no page for anyone to license from.
Who this suits#
People with a substantial written or visual archive on their own domain. People who already receive permission requests and have been granting them free. People with footage, images or audio sitting unused. And anyone who wants their terms on record regardless of whether it pays this year.
Who it does not#
Anyone looking for near-term income. This is a slow, small stream. Anyone whose work exists only on platforms. And anyone expecting the publisher-scale deals to become available to individuals, which they are not going to.
Setting it up#
- Publish somewhere you control. Without your own domain, the rest of this does not apply, and that is the real first step for a lot of people.
- Decide your position: open with attribution, open for answering but closed for training, or closed and priced. Do not default into silence.
- Add
license.xmlto your site root using a template from the RSL spec, and add theLicense:line torobots.txt. On WordPress, use the plugin instead if you prefer. - If your site is behind Cloudflare and you want enforcement rather than declaration, turn on pay-per-crawl and set a price.
- Put up a licensing page with real numbers, including the free uses.
- Stop granting permission for nothing.
If you are unsure whether to allow AI crawlers at all, being found by AI assistants covers the other side of that decision, including how little is known about it.