What the data for this category looks like
Core data for this category comes from advertising delivery platforms, landing page hosting services, product qualification filing systems, front-end code repositories, and advertising law compliance clause libraries. Daily processed documents include advertising creative files, landing page screenshots, single audit opinion records, modification iteration records, and landing page code snippets. Data updates occur daily or in real time, synchronized with adjustments to delivered creatives, updates to compliance clauses, and changes to landing page code. Single audit data includes fields such as creative ID, delivery channel, copy content, landing page URL, historical audit status, and associated code snippets. It covers full process information for a single audit.
Constraints imposed by these characteristics on multi-turn dialogue and prompt engineering
Scattered data sources and real-time updates require multi-turn dialogue contexts to link to the latest landing page URLs, creative copy, and front-end code snippets. This avoids referencing expired information. Documents with multiple version modification records require prompts to support iterative instructions by audit phase. Examples include step-by-step guidance from initial inspection to rectification review. Fixed audit data field attributes require prompts to explicitly extract designated fields such as creative ID, violation type, and code risk points. This prevents generation of irrelevant content. Real-time updates to compliance clause libraries and code repositories require dialogue contexts to load the latest content. This ensures accurate audit basis.
Configuration Settings
In FastGPT v4.17.0, adjust the following configuration items to match the business characteristics of advertising creative audit:
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Unique identifier bound to a single audit session | Isolate audit dialogue contexts for different creatives, avoid cross-session data confusion |
messages | Conversation array appended sequentially by initial inspection, rectification, and review phases | Support step-by-step guidance for multi-turn processes, retain complete audit trail |
variables | Key-value pairs passing current creative ID, landing page URL, compliance clause version, and associated code snippets | Enable prompts to accurately link to specific business and code data for the current audit |
| Maximum context length | 8000–12000 characters | Cover full dialogue and associated documents, code content for a single audit, avoid truncation of critical information |
| Maximum single-turn response length | 1500 characters | Control output format, reduce probability of JSON parsing failure caused by line breaks |
| Session timeout duration | 600 seconds | Match average processing duration of a single audit, avoid premature session termination |
Three Common Configuration Errors
- Phenomenon: Calling the dialogue interface returns the
unAuthChaterror, and the apikey parameter is correctly configured. Cause: The session permission bound invariablesis not linked to the apikey, or duplicate use of the session identifierchatIdcauses permission verification conflicts. - Phenomenon: AI dialogue response content contains a large number of line breaks, causing JSON request parameter parsing failure. Cause: The maximum single-turn response length is not restricted, and the prompt does not explicitly require output to use escaped line breaks or compact format.
- Phenomenon: An error is triggered after uploading advertising creatives or landing page screenshots, but text input dialogue functions work normally. Cause: No file upload path verification rule is configured in
variables, or the verification logic for creative file formats is not adapted.
How to Verify Correct Configuration
- Initiate a simulated advertising creative audit dialogue, check if
chatIdupdates with each session, ensure contexts for different audit tasks do not overlap. - Input a test request containing a landing page URL and code snippet into the dialogue, confirm that parameters in
variablesare correctly loaded into the prompt. - Generate a test response containing line breaks, verify that the JSON format returned by the interface can be parsed normally, adjust the single-turn response length to meet business requirements.
- Upload a test landing page screenshot, confirm that the file upload process has no errors and file information is correctly passed to the dialogue context.
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.