AI scraping is not automatically illegal or copyright infringement. The answer depends on access, copying, reuse, personal data, contracts, and the laws that apply to the project.
AI web scraping means using automated tools to collect website content, sometimes with a model extracting or interpreting it. Collecting prices, republishing articles, and building a model-training dataset involve different uses and legal questions.
Is scraping a public website legal?
In the US, the Ninth Circuit's 2022 hiQ v. LinkedIn opinion affirmed a preliminary injunction in a dispute over public profiles and the Computer Fraud and Abuse Act (CFAA). It was a decision about a specific access claim at that stage of litigation. It did not give scrapers blanket immunity from copyright, contract, privacy, or other claims.
Login requirements, paywalls, technical barriers, and the terms governing your access can materially change the analysis.
Is AI scraping copyright infringement?
It can be. Copying protected articles, photographs, or other original expression can implicate copyright even when the material is freely viewable.
In US law, facts themselves are not protected by copyright, although a sufficiently original selection or arrangement may be. The Copyright Office explains that distinction in Circular 14. Extracting a factual price and republishing an entire product photograph are different uses requiring different analysis.
Fair use is case-specific. US courts consider the purpose and character of the use, the nature of the work, the amount taken, and the effect on the work's market. Commercial use is not automatically disqualifying, and describing a use as “transformative” does not decide the outcome. See the Copyright Office's fair use guidance.
Crediting the source does not by itself authorize copying. Neither does paraphrasing through an AI tool establish that the underlying collection or reuse was lawful.
Do not treat AI training as a universal fair use exemption. The Copyright Office's May 2025 pre-publication report on generative AI training analyzes different uses and circumstances. It is policy analysis, not a court judgment granting permission for a particular dataset.
What changes in the EU and UK?
The EU's Copyright Directive, Articles 3 and 4, provides text and data mining exceptions with conditions. Article 4 concerns lawfully accessible material and applies where rights have not been expressly reserved in an appropriate manner. Article 3 has separate rules for qualifying scientific research. Neither is an unrestricted right to republish the source material. National implementation also matters.
Separate EU database rights can also restrict extraction or reuse of all or a substantial part of a qualifying database. Individual facts being outside copyright does not settle that question.
The UK's section 29A permits qualifying computational analysis for non-commercial research by someone with lawful access, subject to conditions including sufficient acknowledgement where practicable. Do not assume that exception covers commercial scraping or commercial AI training.
Does public personal data remain protected?
Yes. Privacy obligations can apply to publicly accessible personal information. The ICO and other privacy regulators specifically address this issue in their joint work on scraping. A visible email address is not unrestricted permission to build and sell a personal database.
Where the EU's General Data Protection Regulation (GDPR) applies, processing personal data needs a lawful basis under Article 6. Consent is one possible basis, not the only one. Public availability does not remove the other obligations under data protection law.
What is the best way to reduce legal uncertainty?
Start with your own data, a licensed dataset, or an API whose terms permit your intended use. Record the permission, collect only what the task needs, and separate permission to analyze from permission to redistribute.
Before collecting, answer these questions:
- Access: what pages may you retrieve, and under which account terms or license?
- Content: are you collecting facts, protected expression, personal information, or a database subject to separate rights?
- Use: will you analyze privately, publish, sell, or train a model? Does your permission cover that use?
- Jurisdiction: which laws and contractual obligations apply to the operator, source, and data?
Respect access controls and crawl instructions. Robots.txt is not access authorization, and an allowed crawl does not grant a copyright license.
For a commercial project involving protected content or personal data, get jurisdiction-specific legal advice before committing to the dataset.
For the technical workflow, see how ChatGPT extracts website data.