Can you license a dataset you scraped from the web? Four checks, and which route will take it
Quick answer, Nov 16, 2026As of 16 November 2026, a dataset you scraped can be licensed only as far as four checks allow. They are how you reached the pages, the site terms you accepted, whether the fields are facts or someone else's expression, and whether any personal data in it was lawfully public. Each route that might take it reads them differently.
A company that has spent years collecting listings, prices, filings or reviews from public websites often holds a large, clean dataset. The first question a buyer asks about it is not size. It is how the records were collected, and whether you can pass on rights you may never have held.
Four checks before anyone prices it
| Check | What the law or the terms say | Source | What to keep on file |
|---|---|---|---|
| How you reached the pages | Federal law makes it an offence to access a computer "without authorization or exceeds authorized access" and obtain "information from any protected computer", and lets a person who suffers damage or loss sue, in the cases § 1030(g) lists. The Supreme Court reads both clauses as "a gates-up-or-down inquiry" and left open whether limits "contained in contracts or policies" count. The Ninth Circuit concluded that "the concept of 'without authorization' does not apply to public websites". | 18 U.S.C. § 1030(a)(2)(C) and (g); Van Buren v. United States, No. 19-783 (U.S. 3 June 2021), slip op. at 13 and n.8; hiQ Labs v. LinkedIn, No. 17-16783 (9th Cir. 18 April 2022), slip op. at 36 | A collection log showing that no login, password or other access control was passed |
| The terms you accepted | The same court said that where the federal statute does not apply, "breach of contract" is among the claims that "may also lie", alongside copyright infringement, misappropriation, unjust enrichment, conversion and breach of privacy. Platform terms can bar collection outright: LinkedIn members agree not to use "software, devices, scripts, robots or any other means or processes" to "scrape or copy the Services". | hiQ Labs v. LinkedIn, slip op. at 41; LinkedIn User Agreement, § 8.2 "Don'ts" (effective 3 November 2025) | Which accounts, if any, collected each source, and a dated copy of each site's terms as they stood when you collected |
| What you copied | The Supreme Court wrote that "facts are not copyrightable" and that "the copyright in a factual compilation is thin". Protection never extends "to any idea, procedure, process, system, method of operation, concept, principle, or discovery", and a compilation copyright "extends only to the material contributed by the author". | Feist Publications v. Rural Telephone Service, 499 U.S. 340, 344 and 349 (1991); 17 U.S.C. § 102(b); 17 U.S.C. § 103(b) | A field list that separates facts, such as a price or a date, from expression, such as a review, an article or a photo, with the source site for each |
| The people in the rows | Privacy regulators from several countries wrote that "Individuals and companies that scrape such personal information are therefore responsible for ensuring that they comply with these and other applicable laws", and later that terms authorizing scraping "cannot, in and of themselves, render such scraping lawful". Illinois bars collecting a "voiceprint" or a "scan of hand or face geometry" without first informing the person in writing and receiving "a written release", and bars anyone holding one from selling it. Photographs alone are not biometric identifiers there. | Joint statement on data scraping and the protection of privacy, 24 August 2023, "Introduction"; Concluding joint statement, 28 October 2024, "Key takeaways"; 740 ILCS 14/10 ("Biometric identifier"), 14/15(b)(1)–(3) and 14/15(c) | Which fields identify a person, where each came from, and whether anything biometric was ever derived from images or audio |
Every quotation was read on the page it links to on 16 November 2026. The hiQ decision came at the preliminary-injunction stage and binds only the Ninth Circuit. This page is not legal advice, and whether a given collection was lawful is a question for your counsel.
Which route will take it
The four checks do not change by route, but each route leans on a different one. The side-by-side of the routes themselves is in a data program, a direct licence or a marketplace. Here is how each reads a dataset collected from the web.
A buyer program. The published programs describe a company's own operating records, not collected web data. micro1 asks for "mature business operations, and substantial historical operational data" (micro1). Mercor asks you to "Choose from 100+ supported tools and link your accounts", then "extracts and de-identifies data into an asset" (Mercor), and each entry on its listings page names records the company produced or owns, such as "complete files from closed matters and completed deals" or "original video you shot and own" (both read 16 November 2026). If web-collected fields sit inside the records you would hand over, expect the program to ask about them.
A lab directly. What a lab's own partnership page asks is set out in the route guide. Read that row with the four checks above in hand.
A data marketplace. The marketplace turns the fourth check into a written promise. On AWS Data Exchange you "represent and warrant" that identifying data "has already lawfully been made available to the general public, such as via governmental records, widely distributed media, or legally required public disclosures" (AWS Service Terms, § 73.5, last updated 5 November 2026). Its publishing guidelines define Publicly Available Information as information lawfully made available through "federal, state, local government records, open court records, or public company filings", or "lawfully made available to the general public by the data subject" (AWS Data Exchange publishing guidelines). Sort each identifying field into one of those categories, or remove it, before you list.
An advisor-led sale or licence. This route fits when the collected data is mixed with records your business produced, or when the checks decide whether there is anything to license at all. The work takes the records, the rights question, the buyer and the terms together, and a later investor, lender or buyer of the company reads the same answers in diligence.
What to have ready
- The collection log. Dates, sources, and whether any login or account was involved.
- The terms file. A copy of each source site's terms from the period you collected.
- The field map. Facts, expression and personal data, marked field by field.
- The licensed-in parts. Anything you bought or were given rather than collected follows its own grant. Can you relicense data you licensed covers that chain.
- Your own records. Where they sit beside the collected data, the rights checklist covers customer contracts and employee privacy.
Where Frontier Desk fits
In the Cybernetic Age a dataset is priced as an asset, and an asset whose collection cannot be explained prices badly or not at all. Frontier Desk is the company's advisor on a sale or licence of its data. The data advisory examines "the records, the rights question, the buyer and the proposed terms alongside your plans for the business". The buyers and marketplaces named here are destinations a dataset can go to, and the buyer index shows what each published program states. The data room guide shows where the collection log and terms file belong. Start with a first look.
Questions
The pages were public. Doesn't that settle it? No. Public access speaks to the first check only. The court that read the federal statute narrowly for public websites still listed breach of contract, copyright infringement and breach of privacy as claims that may lie (hiQ, slip op. at 41).
What is a scraped dataset worth? UNKNOWN. No public source read today states a price for a web-collected dataset, and the ranges that programs publish are for operating records. Price follows the four checks: a buyer cannot value rights you cannot show.
Can we clean the dataset after the fact? Partly. Dropping expressive fields and personal data narrows the third and fourth checks. It does not change the terms you accepted while collecting, and it does not undo how you reached the pages.
Does it matter that we never logged in? It matters for the first check, and the collection log is how you show it. It does not decide the second: whether a site's terms bind a visitor who never signed in turns on those terms and on your counsel's reading of them.