China Privacy Law PIPL automated decision-making and personalized recommendations
PIPL checks for automated decisions, personalized recommendations, marketing, differential treatment, explanations, refusal rights, and impact assessments.
Article 24 applies when personal information is used to make decisions automatically. Check information pushes and marketing, transaction terms, explanations for decisions with a major effect, and the right to refuse a major decision made solely through automation.
If a system uses personal information to analyze or assess behavior, habits, interests, finances, health, credit, or another characteristic and then makes decisions automatically, treat it as under PIPL. Article 24 requires transparency and fair, impartial results, prohibits unreasonable differential treatment in transaction terms such as price, and gives additional choices for recommendations and marketing. People can request an explanation of decisions with a major effect on their rights or interests and refuse such decisions made solely through automation.
1
Section 1
Classify the feature before choosing the control
Inventory any feature that profiles, scores, ranks, recommends, targets, prices, approves, rejects, flags, or allocates using personal information. Examples can include a personalized product feed, a credit or fraud score, a price or discount rule, or an eligibility decision; the actual inputs and decision path determine coverage. Record the decision made, the input categories, the people affected, whether a person meaningfully participates, and the likely effect on rights or interests.
Article 24 contains three related rules. All must be transparent and produce fair and impartial results. Automated information pushes and commercial marketing must also offer either a non-personalized option or a convenient refusal method. When an automated decision has a major effect on rights or interests, the individual can request an explanation. The individual can also refuse such a decision made solely through automated decision-making.
Recommendation or marketing owner: provide the alternative or refusal where the output is an information push or commercial marketing. Test that the control is easy to find and changes subsequent processing.
Transaction owner: test whether similarly situated people receive different prices or other transaction terms, and document why any difference is reasonable and connected to a legitimate processing purpose.
High-impact decision owner: determine whether the decision materially affects eligibility, access, finances, work, health, safety, or another right or interest and whether it is made only through automation.
Human-review owner: a nominal approval is not enough evidence of meaningful human involvement. Record what information the reviewer sees, what authority the reviewer has, and whether the reviewer can change the outcome.
Articles 24 and 73 define automated decision-making and establish transparency, fairness, differential-treatment, non-personalized-option or refusal, explanation, and solely automated decision-refusal rules.
2
Section 2
Apply the ordinary PIPL duties too
Article 24 does not replace the rest of PIPL. Choose an Article 13 processing condition, give the Article 17 notice, collect only what is directly related and necessary, keep it for the shortest necessary period, support individual rights, and apply Article 51 security measures.
Complete a personal information protection impact assessment (PIPIA) before using personal information for . The assessment must address legality, legitimacy, necessity, effects on individual rights, security risks, and whether safeguards are lawful, effective, and proportionate. Keep the PIPIA and processing record for at least three years. PIPL has applied since 1 November 2021 and does not provide a general grandfathering exception for an automated feature that already existed on that date.
Data owner: record the source, accuracy, relevance, and sensitive-information status of each input. Sensitive personal information requires the separate Article 28-30 analysis.
Product owner: explain the purpose and practical effect in clear language. Do not describe the system so generally that a person cannot understand what choice or consequence is involved.
Rights owner: define intake, verification, response, explanation, refusal, complaint, and escalation steps. PIPL does not set one universal explanation format or response deadline for Article 24 requests.
Vendor owner: if a vendor processes personal information on instructions, include Article 21 terms and supervision; if the vendor independently decides purpose and method, assess the separate-recipient duties.
Monitoring owner: retest inputs, outputs, refusal behavior, non-personalized options, overrides, complaints, and differential outcomes after material model, rule, data, purpose, or population changes.
Articles 5-8, 13, 17, 21, 28-30, 44-56, and 59 establish processing conditions, notice, data quality, sensitive-information, rights, security, vendor, and PIPIA duties that also apply to automated decisions.
3
Section 3
Evidence for launch and ongoing review
A reviewer should be able to reconstruct the feature, legal analysis, implemented control, and observed outcome. Keep source facts and test results together so a later change does not leave the PIPIA or notice describing an old system.
Feature record: purpose, owner, affected population, inputs, data sources, output, decision path, human role, recipients, retention, and consequences.
Article 24 record: transparency copy, non-personalized option or refusal design, explanation procedure, solely automated decision-refusal procedure, and differential-treatment testing.
PIPIA record: risks to rights and interests, security risks, selected safeguards, residual risk, approver, date, and at least three years of retention.
Operating record: refusal and option tests, explanation requests, reviewer decisions, overrides, complaints, incidents, fairness monitoring, and material changes.
Boundary: this page cannot decide whether an effect is major or whether human involvement is meaningful without the product's facts, decision consequences, and actual reviewer authority.
Articles 24, 50, 55, and 56 support the Article 24 control record, convenient rights mechanism, pre-processing PIPIA, and three-year minimum retention.
Put the requirement into practice
Prepare the automated decision-making evidence file
Assign owners, implement the Article 24 controls, and keep the feature, PIPIA, tests, requests, and changes connected.
Binding source for the automated decision-making definition, Article 24 controls, ordinary processing duties, rights, PIPIA, and record retention.
Articles 24 and 73 define automated decision-making and establish transparency, fairness, differential-treatment, non-personalized-option or refusal, explanation, and solely automated decision-refusal rules.
Articles 5-8, 13, 17, 21, 28-30, 44-56, and 59 establish processing conditions, notice, data quality, sensitive-information, rights, security, vendor, and PIPIA duties that also apply to automated decisions.
Show 1 more
Articles 24, 50, 55, and 56 support the Article 24 control record, convenient rights mechanism, pre-processing PIPIA, and three-year minimum retention.