The GDPR's Recital 26 sets the test. To decide whether a person is identifiable, account is taken of all the means reasonably likely to be used by the controller or by anyone else, considering cost, time and available technology. If identification is not reasonably likely under that test, the information is anonymous and the regulation does not apply to it.
In practice the bar is high. Removing names is not enough if a record can be singled out by the other details it contains, linked to another dataset, or used to infer something about a known person. Aggregated statistics often meet the test. Record-level data, especially free text, often does not. The difference from pseudonymization is the route back. Pseudonymized data can be re-linked with separately held information, so it remains personal data. Anonymization aims to make that impossible for everyone. Sellers should use the word only when the test has been assessed, since calling a dataset anonymous is a legal claim.
Anonymity is also judged over time. New datasets and better matching techniques can make identification possible where it was not before, so an assessment made today should be revisited when the data is released again.