What the Data for This Use Case Looks Like
Four sources provide core data for ad creative review: ad platform published creative copy, landing page screenshots, product qualification documents, and general advertising law compliance clauses. Daily processed documents include single creative text content, transcribed landing page text, violation rectification opinions issued by reviewers, and historical modification operation records. Data updates occur daily or in real time. Product qualifications and latest regulatory clauses sync in real time per regulatory requirements. Creatives and review records update with campaign progress. No unified fixed fields apply across all documents. Creatives include placement channel and size requirements. Review opinions include violation type and rectification requirements. Modification records include operator and operation time.
Constraints Imposed on Context and Token Configuration
Data characteristics of ad creative review impose multiple constraints on context and token configuration. First, multi-source heterogeneous data structures require context to cover text, transcribed text, and other formats. Token allocation must account for length differences across document types. Second, real-time updated qualifications and regulations require context to support dynamic loading of the latest compliance references, to avoid using expired data. Third, a single review involves many associated documents. Context recall scope must cover sufficient historical review records and compliance clauses, while controlling total token usage to stay within model limits. Fourth, errors directly affect ad delivery compliance. Context must prioritize high-priority compliance references, preventing critical information loss from token truncation.
Configuration Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Max Context | 10000–15000 token | Ad creative review requires recalling multiple historical review opinions, latest product qualifications, and advertising law clauses. This range covers complete context requirements for batch reviews. |
| Max Token per Single Reference | 1800–2200 token | The combined length of single ad creative copy, extracted landing page text, and corresponding review opinions falls within this range, avoiding truncation of critical violation determination content. |
| Number of Context Recall Entries | Top 6–8 | Matches core associated data for ad creative review. Too many entries use excessive tokens, too few miss key rectification references. |
| Context Update Strategy | Real-time Sync | Matches the real-time update rhythm of ad qualifications and regulations, ensuring timeliness of review references. |
| Token Overflow Handling Rule | Prioritize Retaining Compliance References | Prioritize advertising law clauses and historical compliance modification records, truncate redundant text from landing page screenshots. |
Three Common Configuration Mistakes
- Phenomenon: After calling the knowledge base creation or collection creation interface, chat context fails to load as expected. Cause: No associated matching field for context recall is configured, so the system cannot link ad creatives to historical review opinions.
- Phenomenon: Increasing chat history does not show context memory, or search knowledge interferes with context recall results. Cause: Context recall entries are set too low, or max token per single reference is set too small, leading to forced truncation of critical historical review records.
- Phenomenon: Knowledge base capacity calculation results do not match actual document usage. Cause: Calculations do not account for actual token usage of different document types such as ad creatives and review opinions. Redundant text from landing page screenshot transcriptions is mistakenly included in valid capacity.
How to Verify Proper Configuration
- Upload a single ad creative and its corresponding historical review opinions, confirm that context recall content includes at least 3 related review records.
- Adjust the max token per single reference value, verify that no forced truncation of advertising law clauses or rectification requirements occurs in single-reference text.
- Simulate uploading a real-time updated product qualification document, confirm that the context automatically loads the latest compliance references.
- Check system operation logs to confirm there are no context token overflow error records.
The values provided are common starting points, and should be evaluated against the reader's own 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.