When a person whose public data you collected writes to ask what you hold about them, or asks you to delete it or stop using it, you generally have to answer, and in many cases you have to act, within a deadline set by the privacy law that applies. The common assumption in scraping projects is that rights requests belong to companies with customer accounts, and that information copied from public pages is out of their reach. That assumption breaks in a specific place: the GDPR attaches rights to the person, not to the source of the data, and it has no exception for information that was public when you found it. The CCPA does carve out certain publicly available information, but the carve-out has conditions and does not cover everything a scraper collects. The practical consequence is that any project holding names, usernames, profile URLs or other identifiers of real people needs a way to find a person in its data and a procedure for responding before the first request arrives.
Which rights usually reach scraped data
Under the GDPR, the rights that most often apply to scraped personal data are access (what do you hold about me, where did you get it, what do you do with it), rectification, erasure, restriction, objection and, less often, portability. Under the CCPA, the closest equivalents are the rights to know, delete and correct, plus the right to opt out of sale or sharing. Other jurisdictions have their own lists, and the details differ, so the safest reading is that at least access, deletion and objection should be treated as live possibilities wherever your project reaches people.
A useful way to classify what you hold is by how directly it points at a person. A record with a profile URL and a display name points directly. A record with a review text and a star rating, with the reviewer removed, may still point at the person if the text is distinctive. Aggregates, such as counts of posts per week, usually do not. The classification matters because the rights attach to data that relates to an identifiable person, and because your ability to find that person's records depends on the identifiers you kept.
Why "it was public" does not end the conversation
The GDPR applies to processing of personal data regardless of where the data came from. Collecting it from a public page is processing, storing it is processing, and analyzing it is processing. The fact that the person published the information affects the balancing test for legitimate interests and the person's reasonable expectations, but it does not switch off the rights. The guide to legitimate interests and web data collection covers how public availability enters that balancing.
The CCPA takes a different route. Its definition of personal information excludes "publicly available" information, and that term covers, in general terms, information lawfully made available from government records, information a business reasonably believes the person made available to the general public, and information from widely distributed media. The conditions matter: a public post on a platform may qualify, but data the person shared with a limited audience generally does not, and the exclusion is read against how the business uses the information. Whether a given dataset falls inside or outside this exclusion is exactly the kind of question to put to a lawyer once, for the project as a whole, rather than guess at per request.
Being able to find a person is the prerequisite
No procedure works if you cannot locate a person's records. Suppose a hypothetical dataset of two million public product reviews, each with a reviewer display name, a profile URL and the review text, collected over three years and stored as monthly snapshots. A request arrives from someone who gives their full name and says they wrote reviews on two of the sites you cover. To answer, you need to search by display name across every snapshot, then by profile URL once you have identified the right account, then in any derived tables, such as a reviewer-level summary, and in backups and in copies held by a vendor or shared with a client.
The concrete check is to run that search once, before any request exists, with a fabricated identifier that you know appears in a handful of places, and to count how many stores you had to open to find all of them. If the answer involves opening files by hand, the project needs an index of identifiers to storage locations. The post on keeping a history of scraped data explains why snapshots multiply the places a record can live.
Verifying who is asking without collecting more data
You may ask for enough information to be reasonably sure the requester is the person the data is about, and no more. For scraped public data, this usually means asking which account or profile the request concerns, because a name alone may match several people in the dataset. A proportionate approach is to ask the requester to point to the public profile and, where the stakes justify it, to send a short message from that account or through the platform's contact feature. Demanding a copy of an identity document to delete a public username is usually disproportionate and creates a new store of sensitive data that you then have to protect.
The mistake that usually breaks this step is the opposite one: treating any email as sufficient and deleting records about a third party at the request of someone who is not that person. Keep verification light but real, and record what you relied on.
Answering an access request
An access response has two parts: the data itself and the context around it. The context generally includes the purposes you process the data for, where you obtained it (the specific sites or platforms), who you have shared it with or who may receive it, how long you keep it, and a reminder of the person's other rights. For scraped data the source line matters more than usual, because the person may not know you exist until your response arrives, and the response is often their first notice.
The data part should be readable. A raw export with internal column names, nested JSON and snapshot timestamps answers the letter of the request and fails its purpose. In the hypothetical review dataset, a sensible response lists each review found, the site it came from, the date it was collected and whether it is still present on the live site, plus the reviewer-level summary fields if you computed any. Both the GDPR and the CCPA set a short statutory deadline for responding, which varies by law and can be extended in some cases; confirm the current figures with counsel and build the tracking around them rather than relying on memory.
Erasure and objection: when you stop and when you may continue
An objection to processing based on legitimate interests obliges you to stop unless you can show compelling legitimate grounds that override the person's interests, or the processing is needed for legal claims. An objection to direct marketing use is absolute. An erasure request succeeds on several grounds, including that the data is no longer necessary, that the person objected, or that the processing was unlawful, and it has exceptions, including freedom of expression, legal obligations and certain research or statistical purposes with safeguards. Which of these fits a particular dataset is a legal judgment, but a working default for most commercial web data projects is to honor erasure and objection requests about identifiable records, because the compelling-grounds argument is hard to make for data that was collected opportunistically from public pages.
The practical difficulty is scope. Erasure means the record, every snapshot of it, every derived row that can be traced back to the person, and copies at third parties, not just the row in the current table. Aggregates that no longer identify anyone can usually stay. The hypothetical review dataset would need the review rows removed from all thirty-six monthly snapshots and from the reviewer summary, and the client who received a quarterly export would need to be told.
Suppression lists and the risk of collecting the person again
Deleting a person's records does not stop the next run from collecting them again from the same public page. The standard answer is a suppression list: a minimal store, usually only the identifier such as the profile URL or a one-way hash of it, checked by the scraper before it stores a record. Holding that list is itself processing of personal data, done to honor the person's request, and it needs its own entry in your records of processing with that purpose stated.
The check is simple: after an erasure, run the collection against a saved copy of a page containing that identifier and confirm the record is dropped before storage, not filtered out later. The post on testing a scraper against saved pages describes how to run such checks without sending requests to the live site.
Vendors, clients and copies you do not hold yourself
When a provider collects the data for you, the request still lands on your desk as the controller; the provider's role is to help you locate and remove the records. A provider such as ScrapingIsNotACrime acting as a processor would be expected to assist with that search and deletion, and the contract should say how. When you deliver data to clients, you generally have to tell each recipient about an erasure or rectification, unless that proves impossible or involves disproportionate effort, and the person can ask you who those recipients were. The roles involved are explained in who answers for scraped personal data.
Setting up the procedure before the first request
A workable procedure for a small team has five pieces. First, a visible way to reach you, ideally the same contact route that your scraper's information page already advertises to website owners. Second, an intake log with the date received, the right invoked, the verification relied on, the deadline and the outcome. Third, a written search procedure that lists every store to check, including backups and vendor copies, so the search does not depend on one person's memory. Fourth, response templates for access, erasure and objection, with the context paragraphs already drafted. Fifth, a decision record for refusals, because a refusal has to be explained to the person and may have to be defended later.
Suppose that same hypothetical team receives three requests in a quarter: one access, two erasures. With the index of identifiers in place, each search takes minutes; without it, the first request alone consumes days of opening snapshots. The cost of the index is paid once; the cost of not having it is paid on every request.
Mistakes that turn a request into a complaint
Ignoring a request because the data was public is the most common mistake and the one most likely to reach a regulator. Over-verification, asking for identity documents to remove a username, is the second. The next three are operational: forgetting backups and vendor copies, deleting the suppression list along with everything else so the person reappears in the next run, and answering an access request with an unreadable dump. The last is procedural: treating an informal message as something other than a request. A request does not have to cite a law or use the right words; if a person asks what you hold or asks you to delete it, the clock has started.
Handling rights requests about scraped personal data
The project that handles these requests well is the one that decided, before collecting anything, which identifiers it would keep and where every copy of a record would live. From that foundation, access becomes a lookup plus a template, erasure becomes a lookup plus a deletion across known stores plus a suppression entry, and objection becomes a judgment that is usually resolved in the person's favor. Have the lawyer confirm the jurisdictional questions once, for the project: whether the CCPA exclusion applies, which GDPR exceptions you could ever rely on, and what the current deadlines are. After that, the work is clerical, and clerical work is what a documented procedure is for.