AI Training Opt-Out for Websites – TDM, Copyright and Evidence
A practical guide for creators and website owners: reserving copyright for text and data mining, understanding robots.txt and keeping an auditable rights record without blocking search engines.
Can you stop AI companies from training on photographs, articles or other protected works simply by adding 'AI training prohibited' to your website? An express copyright reservation can affect the general text and data mining (TDM) exception. But a rights reservation, ordinary website notice, robot access setting and research exception are different matters. No single file or label guarantees technical protection against every kind of copying.
Short answer: Sweden's Copyright Act contains Sections 15a–15c. A rightholder may reserve the rights concerned under Section 15a in an appropriate manner. For publicly accessible online works, the EU DSM Directive highlights machine-readable reservations. First establish which rights you actually own or control, choose clear technical signals, preserve dated evidence and avoid accidentally blocking ordinary search indexing. A reservation does not replace a licence agreement or individual infringement analysis.
1. What is text and data mining (TDM)?
Under Section 15c of Sweden's Copyright Act, TDM means automated analysis of digital text and data to generate information, such as patterns, relationships or statistics. AI development may use TDM, but not every instance of AI training or web collection automatically has identical legal treatment. Examine the act and the exclusive rights involved. Online availability does not remove copyright.
Under Section 15a, making copies for TDM is permitted under specified conditions when the user has lawful access. Copies may not be retained longer than necessary or used for other purposes under the provision. The rightholder may appropriately reserve the rights. Uses outside this exception's conditions require their own legal basis.
2. Separate exceptions: Sections 15a and 15b
Section 15a is the wider lawful-access TDM exception but does not apply when the relevant right has been appropriately reserved. It implements Article 4 of Directive (EU) 2019/790. Section 15b applies to certain research organisations, publicly accessible libraries and museums, archives and audiovisual heritage institutions undertaking research TDM under particular conditions. It implements Article 3; it requires safeguards for storage and security, and contractual terms restricting the statutory Section 15b right are invalid.
A reservation under Section 15a therefore does not automatically bar research TDM lawful under Section 15b. Separate licences, other rights and other exceptions may also apply. Do not suggest a single 'no AI' line stops all data-related activities.
3. Who is entitled to reserve rights?
Create a rights ownership register: work name, creator, URL, relevant copyright holder, licence terms, image-agency terms, commissioned work and authority to issue a reservation. A retailer publishing customers' photos or licensed images may lack the authority to reserve all rights. Different rights may exist in translations, joint works, archived content, illustrations and product images.
A reservation may describe specific works, catalogues or site areas where the scope is clear. Do not assert exclusive copyright over public-domain material or content whose licence permits the intended activity.
4. How can an online reservation be machine-readable?
Article 4(3) of the DSM Directive requires an express reservation in an appropriate manner, such as machine-readable means for online-public content. The European Commission discusses robots.txt and other machine-readable TDM protocols. Standards and interpretations are still developing. A footer notice alone should not be presented as a reliably machine-readable opt-out. Check which technically implemented methods fit the content and rights.
An illustrative visible statement could say: 'The rightholder reserves text and data mining rights for the identified protected works to the extent provided by Section 15a of the Swedish Copyright Act and Article 4(3) of Directive (EU) 2019/790.' This wording alone does not establish that an effective machine-readable reservation exists. Check that machine-readable settings, metadata, contractual licences and the visible statement agree.
5. robots.txt is not the same as search indexing or an all-purpose legal ban
robots.txt is a machine-readable protocol used to guide crawler access to paths. Bot-specific settings can affect which crawlers fetch content, but bot names and actual compliance vary. Blocking one crawler is not proof that all uses for AI training are legally prohibited; conversely, a rights notice cannot force malicious technology to obey.
For Shopify and other publishing sites: do not add a site-wide Disallow: / simply to stop AI. That can block ordinary search engines and harm discoverability. Assess targeted methods and their legal consequences, back up settings and test relevant paths and user agents. This article does not make changes to the website's robots.txt.
6. Rights and evidence control matrix
| Check | Evidence to keep | Purpose |
|---|---|---|
| Chain of title | Original files, authors, licences and commission terms | Shows who may reserve rights |
| Works and URLs | Titles, file hashes, addresses and dates | Identifies covered material |
| Reservation scope | Exact text, mechanism, works and versions | Avoids ambiguity |
| Machine readability | Technical signal and dated response record | Shows what a crawler can access |
| Search visibility | robots.txt review and critical indexing routes | Prevents accidental SEO blocking |
| Licence exceptions | Permissions granted to partners or customers | Prevents contradictory instructions |
| Incidents and records | Time-stamped settings, logs and correspondence | Supports later legal review |
7. Eight practical steps for creators and website operators
- Inventory ownership. Identify your works, photographers, co-authors, image-agency licences and platform terms.
- Define coverage. List URLs and files included and excluded.
- Write the reservation. Clearly state the relevant TDM rights, authorised holder and works, without overclaiming.
- Select a technical signal. Review an appropriate machine-readable method or bot-specific rule for online works.
- Test SEO independently. Check normal search crawlers and important pages are not blocked accidentally.
- Preserve before-and-after evidence. Archive the visible statement, technical response, version, date and existing licences.
- Notify relevant partners. Inform the website administrator, platform and licensees where appropriate.
- Review changes. Monitor new URLs, image catalogues, protocol developments and potential misuse.
8. What does the AI Act require from model providers?
Article 53(1)(c) of Regulation (EU) 2024/1689 requires providers of general-purpose AI (GPAI) models to implement a policy respecting EU copyright and related rights, particularly to identify and comply with rights reservations under DSM Directive Article 4(3), including state-of-the-art technologies. The Commission's voluntary GPAI Code of Practice describes measures involving robots.txt and further machine-readable protocols. The Article 53 duty applies to GPAI model providers, not automatically to everyone using an AI chatbot.
The Commission describes enforcement of GPAI obligations from 2 August 2026, subject to applicable transitions. That does not prove that any one provider used your work, honoured your reservation or owes automatic compensation. Rights, collection, dates and use require separate evidence.
9. Two practical situations
Photographer publishes an online portfolio
A photographer owns their original images and wishes to reserve TDM rights. They record image sets and URLs, communicate an express reservation and investigate machine-readable methods. A screenshot of footer text alone does not show what a crawler was technically able to read; archive technical responses and licences.
Store owner blocks the entire site through robots.txt
A merchant adds a blanket robot block to prevent AI training and accidentally blocks search crawlers from products and guides. Restore the incorrect configuration using a verified backup, consider narrower options and assess the legal TDM reservation separately. SEO access controls and copyright rights are distinct issues.
10. Frequently asked questions
Is 'AI training prohibited' in the footer enough?
Not as a reliably complete solution. Online rights reservations should be expressed clearly and in an appropriate machine-readable way for the technical context. A footer can add explanatory information.
Does robots.txt stop all AI training?
No. It influences crawlers that choose to respect it and covers their specified access. It cannot prove that older copies are not used.
Can I exclude the research exception in Section 15b?
A Section 15a reservation does not universally override Section 15b research TDM. That section has its own requirements and disallows contracts restricting the statutory permission.
Can I reserve rights in images that belong to someone else?
Not without relevant authority. Check authorship and licences; permission to display an image is not necessarily permission to control all exclusive rights.
11. Official legal sources and related guides
- Swedish Copyright Act (1960:729), Sections 15a–15c
- EU Directive 2019/790, Articles 3–4
- European Commission – AI Act Article 53 and copyright
- European Commission – TDM opt-out protocols
- European Commission – GPAI provider obligations
- Mallbutiken – intellectual property licensing agreements
- Mallbutiken – AI policy for companies
Legal sources reviewed 19 November 2026. A Section 15a reservation is not a ban on all AI activity or lawful research TDM. Technical standards are evolving; examples are not guarantees of enforceability. No robots.txt, Shopify theme or product was changed.
Explore the topic
Business guides – contracts, HR, AI & complianceLicense Agreement – exclusivity, royalty, IP and key terms and conditions
Guide to licensing agreements for software, trademarks, patents, and know-how: exclusivity, territory, royalty, audit, sublicensing, infringement, ...
Business guides – contracts, HR, AI & complianceAI policy for companies – what should it include and when is it needed?
What should an AI policy include? A practical guide to permitted AI tools, data, human oversight, AI literacy, GDPR, security, and incidents.

