What is RSL (Really Simple Licensing)?

RSL (Really Simple Licensing) solves a very specific problem: robots.txt lets you block or allow a crawler's access, but says nothing about licensing or price. A publisher could either keep an AI bot out or let it in with no conditions. RSL adds a third option: a License directive that points to a document with the exact terms under which the content may be used, from a free licence with attribution to payment per crawl or per inference.

The standard extends the Robots Exclusion Protocol (REP) rather than replacing it. Allow and Disallow still exist; RSL adds License as a layer on top. It is worth distinguishing it from the TDM Rights Reservation (TDMRep): TDMRep expresses an opt-out right for text and data mining recognised under EU law. RSL is not a new legal right, it is a format for negotiating and expressing licensing terms, including a price, and it works regardless of which country's law applies.

For a publisher, the practical difference is this: TDMRep says «I don't want you to use this to train models». RSL says «you can use it, under these terms, at this price».

RSL is published as an XML vocabulary linked from robots.txt. A publisher adds a License line pointing to the URL of its licence document; a crawler that respects the standard reads it before downloading content, so it knows the applicable terms without having to ask. The licence document can describe several models at once: a free licence with mandatory attribution, a subscription, payment per crawl, or payment per inference, and terms can vary by declared use, meaning training, search, or an agent answering in real time.

The standard is developed by the RSL Technical Steering Committee, with Eckart Walther (RSS co-author) and Doug Leeds (former Ask.com CEO) as founders of the RSL Collective. The Collective launched on September 10, 2025, with Reddit, Yahoo, and Medium among the first publishers to adopt it. The RSL 1.0 specification, with Recommendation status, was published on December 10, 2025. Declared supporters include Cloudflare, Akamai, Fastly, Creative Commons, O'Reilly Media, and Ziff Davis, a broad coalition of infrastructure providers and publishers.

RSL does not require any crawler to pay or to respect the directive: like robots.txt, it is a statement of intent that depends on the crawler reading and honouring it. Its strength comes not from a technical blocking mechanism but from documenting the terms under which use was, or was not, authorised, which matters if a case ends up in court. In that sense it resembles a sign listing conditions at the entrance of a shop more than a lock: it does not physically stop entry, but it records the terms under which entry was granted.

Until now, a publisher had only two buttons for an AI crawler: let it through, or block it with robots.txt. Neither allowed saying «you can come in, but pay» or «you can come in, but credit me». RSL turns that binary decision into a conditional one, at a moment when several large infrastructure providers (Cloudflare, Akamai, Fastly) have already publicly backed the standard.

For a publisher client of ZDS, the question is no longer just «do I block AI bots or not», but «under what terms do I authorise each use». That includes deciding whether to charge, how much, and whether attribution is required when the content is quoted in an AI assistant's answer.

It is a young standard: it launched in 2025, and how widely it is actually adopted, both by publishers and by the AI companies that are supposed to respect it, remains to be seen. There is, as of this writing, no public, measured data on how many AI crawlers actually honour the License directive in practice. That does not invalidate it, but it does mean treating it for what it is: a standard proposal with significant backing, not a legal obligation. Which reading proves right will depend on how many AI companies actually adopt it over the coming year, something no one can honestly claim to know yet.

• Publish the licence document at a stable URL and link it from robots.txt with the License directive, before announcing payment terms anywhere else.
• Decide the business model first (free with attribution, subscription, payment per crawl) and only then draft the document; doing it the other way round ends up bending the business to fit the format.
• Distinguish between use for training, for search, and for an agent answering in real time: these are different cases and can carry different terms.
• Check which crawlers declare that they honour RSL before assuming the directive alone stops anything; today it still depends on the crawler respecting it.
• If you already have a robots.txt block or a TDMRep reservation, keep both: RSL adds to them, it does not replace either.
• Review the specification every few months; with RSL 1.0 published in December 2025, the standard is still in an early phase of change.

• Confusing RSL with a technical blocking mechanism: it does not prevent access, it only attaches conditions to it on paper.
• Presenting RSL as if it replaced TDMRep: these are different mechanisms, one is an opt-out right, the other a licensing format.
• Setting a price without checking which AI crawlers actually respect it, then expecting revenue that never arrives.
• Adding the License directive without keeping robots.txt up to date: if the two rules contradict each other, a crawler may follow the more permissive one.
• Treating RSL as a signed legal obligation, when it is a voluntarily adopted standard.