What the data for this category looks like
The data for this category mainly comes from advertising placement platforms, landing page resources, product qualification documents, and advertising law compliance clauses. Daily processed documents include single ad creative content, landing page screenshots, modification opinions output by reviewers, and creative iteration records. Data updates occur daily or in real time, with no fixed statistical cycle. Document fields include creative copy, landing page URL, violation type tags, time limit requirements for modifications, and more. There are no unified standard units. Fields must be matched according to a single review task.
What constraints these characteristics impose on multi-turn dialogue and prompt engineering
Ad creative review data sources are scattered and updated in real time. Multi-turn dialogue must support real-time retrieval of the latest advertising platform compliance rules and creative status. Daily processed documents include unstructured creative copy and landing page screenshots. Prompts must adapt to multi-modal input parsing logic. Review tasks process single creatives one at a time. Multi-turn dialogue must limit context association scope to avoid cross-task interference. Review opinions and modification records must be saved as dialogue history. These records must bind to exclusive parameters for the current review task. This ensures each dialogue’s context accurately matches the current review item.
The following configurations are based on FastGPT v4.17.0.
How to Set Configurations
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind to the unique ID of the current review task | Ensure multi-turn dialogue context only associates with a single creative review task, avoiding cross-task data confusion |
messages | Retain the most recent 15–20 turns of dialogue | Cover the complete process of creative modification and opinion feedback, while controlling context resource usage |
variables | Pass in creative copy, landing page URL, compliance clause version | Provide exclusive current review parameters for prompts, adapting to real-time updated compliance rules |
| Single search limit | Top 3 results | Match the lightweight needs of single creative review, avoiding redundant information interference |
| Maximum context | 8000–12000 token | Calibrated based on actual testing, adapting to document length and historical dialogue for ad creative review |
| Dialogue timeout | 600 seconds | Calibrated based on actual testing, adapting to the processing duration of single creative review |
Three Common Misconfigurations
- Phenomenon: The front-end page displays dialogue content for multiple review tasks simultaneously, and cannot accurately associate with the currently processed creative. Cause:
chatIdis not bound as the unique identifier of the current review task, resulting in mixed multi-task context. - Phenomenon: The prompt fails to identify landing page consistency or copy violation issues as required, and review results do not meet expectations. Cause: Landing page URL or compliance clause version parameters are not passed in
variables, so the prompt cannot obtain corresponding verification data. - Phenomenon: The released dialogue interface only displays raw LaTeX code, while rendering works normally during the debugging preview phase. Cause: The format rendering rules of the front-end dialogue module are not configured, resulting in LaTeX parsing not being enabled after release.
How to Verify Successful Configuration
- Submit a test review task. Check whether the dialogue context only includes modification records and review opinions of the current creative, with no historical content from other tasks.
- Pass test creative, landing page URL, and compliance clause version. Trigger multi-turn dialogue, and confirm that the prompt can correctly call corresponding parameters for review.
- Insert review opinions in LaTeX format. Check whether they render normally on the released dialogue interface with no raw code remaining.
- Initiate multiple test dialogues. Confirm that the system conversation process can proceed normally without persistent lag or timeout situations.
The values listed above are common starting points and should be evaluated against 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.