TDM Rights Reservation

A TDM rights reservation is a statement by a rightsholder that their content may not be used for text and data mining (TDM) without permission. The term comes from EU copyright law: Article 4 of the 2019 Copyright in the Digital Single Market Directive allows anyone with lawful access to mine content, including for AI training, unless the rightsholder has expressly reserved that use, and for content made publicly available online the reservation must be made by machine-readable means. A reservation does not apply to the separate exception for scientific research organizations in Article 3.

The EU AI Act turned this into an obligation for model developers. Article 53 requires providers of general-purpose AI models to put in place a policy to identify and comply with TDM reservations, and those obligations have applied since August 2025. The General-Purpose AI Code of Practice, a voluntary code providers can sign to show compliance, spells out how in its copyright chapter: signatories commit to crawlers that follow robots.txt as specified in RFC 9309, and to honoring other machine-readable opt-out protocols that are standardized or generally agreed through an EU-facilitated process (Code of Practice, copyright chapter). The European Commission has run a stakeholder consultation to identify those protocols; candidates discussed include TDMRep, C2PA's TDM assertions and ai.txt.

What counts as machine-readable is being settled in court. On December 10, 2025, the Hamburg Higher Regional Court ruled in Kneschke v. LAION that a reservation written in natural language in a website's terms did not meet the requirement, because it could not be reliably interpreted by automated processes at the time of the 2021 download. The court allowed an appeal to Germany's Federal Court of Justice, so the question remains open at the highest level. Outside the EU, rules differ: the UK government said in March 2026 that a broad TDM exception with an opt-out was no longer its preferred approach and that it would not introduce a new exception for now.

There are several technical ways to express a reservation, and they do different things:

• Robots.txt blocks specific crawlers by name. It is the one mechanism the Code of Practice names explicitly, but it controls access, not use, and has to be set crawler by crawler. See robots.txt.
• TDMRep, a W3C Community Group specification (not a W3C standard), sets a tdm-reservation flag and an optional link to a licensing policy, via a /.well-known/tdmrep.json file, HTTP headers or HTML metadata (TDMRep spec). Adoption is low: a February 2026 crawl of the top million domains found it on 37 of them.
• Usage-preference signals such as Cloudflare's content signals separate search, AI input and AI training in robots.txt. The IETF AI Preferences working group is drafting a standard vocabulary for this, still at Internet-Draft stage in September 2026.
• Contracts and licenses, including publisher licensing deals and Really Simple Licensing, grant permission on stated terms.

For AI search visibility, the key is to reserve training rights without cutting off retrieval. A blanket reservation or a block on every AI user agent can also remove a site from AI search answers, because search and training often use different crawlers with different rules. A typical setup reserves TDM for training through TDMRep or content signals and blocks training-only crawlers such as GPTBot and ClaudeBot, while allowing search crawlers such as OAI-SearchBot, Claude-SearchBot and PerplexityBot. For Google, Google-Extended covers Gemini training and grounding, and the Search Console generative AI control covers AI Overviews and AI Mode. Keep terms of service, technical signals and licensing pages consistent, and take legal advice for your jurisdiction. Related: citation crawlers vs. training crawlers and AI regulation.