Information date: 29 September 2026 — Two distinct routes exist. Algorithm filing applies to services that use recommendation, ranking, sorting or search algorithms to push information to users, and is filed with the provincial CAC office. Generative AI, where the service produces text, images, audio or video to the public in China, sits under a separate regime that combines security assessment with filing and adds content-labelling and training-data duties. Knowing that statement is not enough for an operating, research or compliance decision. The team must first establish who and what it applies to, how the effect reaches the real process, and which evidence would justify action.
Verified facts and scope
Two distinct routes exist. Algorithm filing applies to services that use recommendation, ranking, sorting or search algorithms to push information to users, and is filed with the provincial CAC office. Generative AI, where the service produces text, images, audio or video to the public in China, sits under a separate regime that combines security assessment with filing and adds content-labelling and training-data duties.
Decide first who provides the service to the Chinese public: a local entity, a joint venture, a licensed partner, or an overseas app accessible from China. Then map model hosting location, whether user prompts leave China, whether output is published to the public or only to a closed enterprise customer, and which entity holds the ICP filing and user data.
How the effect reaches operations
The classification turns on the function, not on the label. If users in China receive algorithmically ranked or generated content, the filing duties attach to the domestic service provider; a foreign SaaS vendor alone cannot file. Enterprise-only internal tools with no public dissemination often fall outside, which is why the hosting and user-population facts decide the route.
Assuming that an enterprise B2B contract removes the obligation, or that an API behind a local partner transfers all duties, is a frequent misjudgement. Also risky: launching a Chinese-language public interface before filings are complete, training on data without lawful basis, and omitting output labels or complaint channels.
For “Generative AI Filing vs Algorithm Filing in China: A Comparison for Overseas SaaS Providers”, official rules or published findings, direct evidence from the relevant product or process, and assumptions that remain untested should be recorded separately. A broad source defines the external boundary; it does not replace batch records, protocols, contracts, labels or direct observations.
Decision
If the service is enterprise-only and hosted outside China with no public access, document that position and monitor. If it reaches the Chinese public with generated output, plan for the generative AI route with security assessment first. If it only ranks or recommends third-party content, the algorithm filing route is usually the fit.
Implementation checklist
- Classify the feature by output type and audience, not by product name.
- Confirm which Chinese entity will be the filing applicant.
- Sequence security assessment before public launch, not after.
- Assign one decision owner, one implementation owner and a dated review point for “Generative AI Filing vs Algorithm Filing in China: A Comparison for Overseas SaaS Providers”.
- For “Generative AI Filing vs Algorithm Filing in China: A Comparison for Overseas SaaS Providers”, archive the source page, access date, applicable population or entity, and internal evidence both supporting and opposing the current decision.
- When a rule, formulation, supplier, protocol or observed result changes, reopen only the affected question in “Generative AI Filing vs Algorithm Filing in China: A Comparison for Overseas SaaS Providers”.
Evidence and review
For “Generative AI Filing vs Algorithm Filing in China: A Comparison for Overseas SaaS Providers”, start with one real case rather than an abstract checklist. Record the input version, responsible owner, start time, observed result and stop condition. If the team cannot complete “Classify the feature by output type and audience, not by product name.” with current evidence, it should not expand the process to more products, patients, suppliers or markets. The first review should focus only on facts capable of changing the decision.
The second control follows “Confirm which Chinese entity will be the filing applicant.”. Keep the source date, applicable population or entity, deadline, cost effect and owner in the same evidence file. A wording preference does not justify a new version. A repeated discrepancy, an unsupported health claim or a regulatory mismatch does: correct that point and hold release until the evidence is available.
After “Sequence security assessment before public launch, not after.”, compare the intended outcome with what actually happened. Apply the same success criteria to each later expansion. If only one number, date or responsibility changes, update that field and the affected conclusion instead of recreating evidence that remains valid. This keeps the decision traceable without turning review into an open-ended rewrite cycle.
Limits of the conclusion
This comparison is informational and not a regulatory determination; filing scope depends on the actual service, hosting and user base, and should be confirmed with the competent CAC authority.
