What the Data for This Category Looks Like
The data for this category covers four core document types: ad creative copy, landing page screenshots, review opinions, and modification records. Data sources include advertising platforms, landing page servers, product qualification filing systems, and advertising law compliance clause databases. The data update frequency is daily or real-time. Fields associated with a single creative include creative ID, delivery channel, compliance status, and review node records. There is no unified statistical unit; association and traceability must be completed per single creative or single batch review task.
Constraints on the Citation Sources and Traceability Link
The core traceability requirement for ad creative review is to bind a single creative ID with compliance judgment basis, so the citation link must accurately match the creative with corresponding qualification and clause data. Real-time or daily updated data requires the traceability chain to support incremental recall, to avoid citing expired review opinions or qualification documents. In scenarios with mixed document types, different citation formats for creative copy and landing page screenshots must be distinguished, to ensure traceable content can correspond to specific delivery nodes. At the same time, the consequences of errors are directly linked to advertising delivery compliance risks, so citation sources must be traceable back to the original data source, and vague or irrelevant citation entries must be avoided.
Configuration Settings
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
quoteMaxToken | 600–1500 token | Covers the total length of a single ad creative copy and associated compliance clauses |
responseData | Enabled | Retains source identifiers for raw data, used to match creative IDs and review records |
detail | Enabled | Displays recalled raw document snippets, to facilitate verification of compliance basis accuracy |
| Single Search Limit | Top 3–6 entries | Matches the volume of qualification, clause, and landing page data that needs to be associated for single creative review |
| Maximum Context Length | 1800–2500 token | Adapts to the integrated length of multiple document types, to avoid overflow |
| Citation Trigger Threshold | Calibrated via actual testing | Distinguishes similarity matching standards between ad creatives and compliance clauses |
Three Common Configuration Mistakes
- Phenomenon: After increasing
quoteMaxToken, the large model returns content with no associated citation sources. Cause: The configuration did not synchronize updated knowledge base recall association parameters, resulting in recalled content exceeding the context without binding traceability tags. - Phenomenon: In a local deployment scenario, the citation field is empty when called via external channels. Cause: The
responseDataconfiguration item is not enabled, or the release channel did not synchronize the knowledge base traceability permission settings. - Phenomenon: The returned citation content is irrelevant to the answer and does not match the ad creative’s compliance basis. Cause: The similarity threshold setting does not adapt to the matching logic between ad copy and compliance clauses, resulting in recall of irrelevant data.
How to Confirm Proper Configuration
- System logs can be reviewed to confirm that each recalled document snippet is associated with the exclusive identifier of the ad creative.
- During test calls, returned results can be checked for the presence of original source information displayed in the
detailfield. - After adjusting the
quoteMaxTokenvalue, the length of cited content can be verified to cover core compliance basis. - Results from local testing and external channel calls can be compared to confirm consistency of traceability logic.
The values provided are common starting points and should be measured against applicable local test samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.