Web Scraping and Publicly Available Information in Canada
Web scraping of personal information is not exempt from Canadian privacy law just because the data is public. PIPEDA's publicly available exceptions cover only five narrow classes, the Clearview AI findings rejected a broad public-data defence, and regulators expect lawful bases for AI training sets, with contract and copyright adding further limits.

Key takeaways
- Publicly accessible personal information remains subject to privacy law.
- PIPEDA's publicly available exceptions cover only five specified classes, with conditions.
- Clearview AI's scraping of social media images failed the consent and appropriate purpose tests.
- Contractual permission to scrape does not by itself make scraping lawful.
- AI training data sets built from scraped data must comply with privacy law.
- Terms of use and copyright apply even where privacy law does not.
Web scraping and the myth of free public data
Web scraping, the automated collection of content from websites and online platforms, powers price comparison, market research, lead generation, security research and, increasingly, the training of AI models. A common assumption is that anything visible on the open web is free to collect and reuse. Under Canadian privacy law, that assumption is wrong. Personal information does not lose protection because it is publicly accessible, and the exceptions for "publicly available" information under the federal Personal Information Protection and Electronic Documents Act (PIPEDA) are deliberately narrow.

This insight explains the publicly available information regulations, the Clearview AI findings, the regulators' joint statements on data scraping, the special issues raised by AI training data sets, and the contractual and copyright limits that apply even where privacy law does not. It is general information, not legal advice.
The PIPEDA framework for web scraping
PIPEDA applies to organizations that collect, use or disclose personal information in the course of commercial activities. As a general rule, it requires knowledge and consent, and it limits collection to purposes a reasonable person would consider appropriate in the circumstances under subsection 5(3). An organization that scrapes names, photos, profiles, reviews, contact details or other information about identifiable individuals is collecting personal information, so it needs either consent or an applicable exception, and it must also pass the appropriate purpose test.
PIPEDA contains exceptions to consent in paragraphs 7(1)(d), 7(2)(c.1) and 7(3)(h.1) for information that is "publicly available and is specified by the regulations". The word "specified" is critical: it is not enough that information is public in a general sense.
The publicly available information regulations
The Regulations Specifying Publicly Available Information, SOR/2001-7, list five classes of information, each subject to conditions:
| Class | Condition |
|---|---|
| Name, address and telephone number in a public telephone directory | The subscriber can refuse to appear in the directory |
| Name, title, address and telephone number in a public professional or business directory, listing or notice | Use must relate directly to the purpose for which the information appears |
| Information in a statutory registry with a legal right of public access | Use must relate directly to the purpose of the registry |
| Information in a public record or document of a judicial or quasi-judicial body | Use must relate directly to the purpose for which it appears in the record |
| Information in a publication, including a magazine, book or newspaper, in printed or electronic form | The individual provided the information |
Three features stand out. The list is closed. Most classes carry a purpose limitation, so information from a business directory can be used to contact someone in their professional capacity but not, for example, to build a consumer profile. And the publication class applies only where the individual provided the information. Scraping large volumes of social media posts, profile photos or forum content rarely fits any of these categories.
The Clearview AI findings
The leading Canadian example is the joint investigation of Clearview AI by the OPC and the commissioners of Quebec, British Columbia and Alberta (PIPEDA Findings #2021-001, 2 February 2021). Clearview scraped images of faces and associated data from publicly accessible online sources, including social media, created biometric identifiers for each image and offered a matching tool to law enforcement and other clients. Its database contained over three billion images, including images of a vast number of people in Canada, children among them.
Clearview argued that the information was publicly available and exempt from consent. The regulators rejected that argument.
Information collected from public websites, such as social media or professional profiles, and then used for an unrelated purpose, did not fall under the publicly available exceptions in PIPEDA or the Alberta and British Columbia statutes, and was not "public by law" under Quebec's private-sector law. Because biometric information is sensitive, express consent was required. The regulators also found that the purpose, mass identification and surveillance of individuals, was not one a reasonable person would consider appropriate, regardless of consent. They recommended that Clearview stop offering its service in Canada and stop collecting, and delete, images and biometric data of individuals in Canada.
Provincial orders and court challenges
Following the joint findings, the British Columbia and Alberta commissioners issued orders under their provincial statutes, which Clearview challenged in the courts of those provinces. The provincial laws have their own publicly available information regulations, worded differently from the federal ones, so court rulings on them do not automatically transfer to PIPEDA. Organizations relying on any "publication" argument should take advice on the current state of the case law in each province where they operate, as judicial interpretation of these regulations, including arguments based on freedom of expression, continues to develop.
Regulators' joint statements on data scraping
In August 2023, the OPC joined other data protection authorities in a joint statement on data scraping, followed by a concluding joint statement in October 2024. Key messages include:
- publicly accessible personal information is still subject to privacy laws in most jurisdictions;
- social media companies and website operators must protect personal data on their platforms from unlawful scraping, and mass scraping incidents can be reportable breaches;
- contractual permission to scrape does not by itself make scraping lawful, although contract terms can be an important safeguard;
- organizations that permit scraping must have a lawful basis, be transparent and obtain consent where required; and
- organizations that use scraped data sets to train AI, including large language models, must comply with privacy laws and any AI-specific laws.
Web scraping creates obligations on both sides. The scraper needs a lawful basis for collection, and the platform being scraped needs safeguards against unlawful scraping. A mass scraping incident on your own platform may have to be assessed as a breach of security safeguards.
AI training data sets
Generative AI has made web scraping a strategic issue. Training data sets often contain personal information: names, images, biographical details, opinions and contact information drawn from the open web. Under PIPEDA, an organization that builds or fine-tunes a model with scraped personal information must identify its purposes, have consent or a valid exception, and satisfy the appropriate purpose test. Using information for a purpose unrelated to the one for which it was posted, as in Clearview, is a particular concern.
Organizations that license data sets or models from vendors should ask how the training data was collected, whether personal information was filtered or de-identified, and what contractual assurances and indemnities are available. Quebec's law adds obligations around profiling technologies and privacy impact assessments that can apply to AI projects. See our insight on AI vendor contracts and our artificial intelligence practice.
Scraping non-personal data
Not all web scraping involves personal information. Collecting product prices, stock availability, weather data or aggregated statistics may fall outside privacy law entirely. But the line is not always clear. Reviews, user-generated content, seller profiles and metadata can identify individuals directly or in combination. A data protection assessment at the design stage, including filtering techniques that exclude personal information, is cheaper than remediation after a complaint.
Contractual limits: terms of use and technical barriers
Even where privacy law does not apply, website terms of use commonly prohibit automated access, scraping or commercial reuse. Whether those terms bind a scraper depends on ordinary contract principles, including whether the scraper had notice of and accepted them. Circumventing technical measures, such as rate limits, login walls or blocking tools, increases legal risk and undermines any argument that the collection was reasonable. The regulators' concluding statement encourages platforms to offer access through an application programming interface, which gives the platform more control and gives the user a clearer contractual basis.
Copyright limits
Scraped content such as articles, photographs, reviews and software code is often protected by copyright, which is separate from privacy. Canada's Copyright Act does not contain a specific exception for text and data mining. Whether reproductions made for analysis or AI training qualify as fair dealing, or fall within other exceptions, is unsettled and fact-specific. Organizations should consider the nature of the content, the amount copied, the purpose and whether outputs reproduce protected expression. Database compilations may also be protected where their selection or arrangement is original.
Quebec and provincial laws
Quebec's private-sector law requires consent unless an exception applies and does not contain a general exception for publicly accessible information; the Clearview findings applied a "public by law" concept. Quebec also requires organizations that collect personal information using technology with identification, location or profiling functions to inform individuals in advance. Alberta and British Columbia have their own private-sector laws and publicly available information regulations. A national scraping project needs a province-by-province check.
Buying or licensing scraped data
Many organizations never scrape anything themselves but buy data sets, enrichment services or lead lists that were assembled through web scraping. Accountability does not disappear with the purchase. An organization that collects personal information from a third party must be satisfied that the information was obtained lawfully and that it may be used for the intended purpose. Practical diligence includes asking vendors how the data was sourced, which sites were scraped and under what terms, what personal information is included, whether consent was obtained and how opt-outs are honoured. Contracts should contain representations about lawful collection, use restrictions, deletion obligations and indemnities, and the buyer should keep a record of the diligence performed.
A compliance checklist for web scraping
- Define the purpose and confirm whether personal information will be collected.
- If it will, identify the lawful basis: consent, a specified publicly available class or another exception.
- Apply the subsection 5(3) appropriate purpose test and document the analysis.
- Minimize collection and filter out sensitive categories, especially biometrics and children's data.
- Review terms of use and avoid circumventing technical controls.
- Assess copyright risk for the content copied.
- Set retention limits and safeguards for the scraped data set.
- For AI training, document provenance and run a privacy impact assessment.
If you operate the platform being scraped
Platforms holding personal information must protect it against unlawful scraping. The regulators expect a layered set of measures: rate limiting, bot detection, monitoring for unusual access patterns, restrictions on bulk access, contractual controls for authorized partners and enforcement action against violators. Mass scraping incidents should be assessed under breach notification rules. See our insight on breach notification.
Enforcement outlook
Federal privacy reform could change the publicly available information framework and add stronger enforcement powers; see our insight on federal privacy reform. In the meantime, regulators continue to scrutinize scraping for facial recognition, people-search services and AI training, and Quebec's CAI can impose administrative monetary penalties under Law 25.
How TRW & Co helps
We advise data, analytics and AI businesses on web scraping and the use of publicly accessible information. Our work includes lawful basis and appropriate purpose assessments, AI training data provenance reviews, privacy impact assessments, terms of use and API licensing, copyright risk analysis, and platform anti-scraping programs. See our privacy and data protection page or contact us.
Frequently asked questions
Is information on public social media profiles publicly available under PIPEDA?
Generally not. The regulations specify five narrow classes, and the Clearview AI findings held that information collected from social media or professional profiles and used for an unrelated purpose did not fall within the publicly available exceptions.
Can we scrape business contact information?
Information in a public professional or business directory, listing or notice can be used without consent where the use relates directly to the purpose for which it appears, such as contacting the person in their professional capacity. Other uses may need consent.
Does a website's permission to scrape make it lawful?
Not by itself. The regulators' 2024 concluding joint statement says contractual authorization cannot on its own render scraping of personal data lawful, although contract terms can be an important safeguard.
Can we train an AI model on scraped web data?
Only with a lawful basis for any personal information involved and an appropriate purpose under PIPEDA, along with consideration of copyright and terms of use. Provenance documentation and a privacy impact assessment are strongly recommended.
Is scraping of non-personal data regulated?
Privacy law may not apply, but terms of use, technical access controls and copyright still can. Many data sets that appear non-personal include identifiable user content, so filtering is important.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


