What the data for this category looks like
The core users of this category are ad reviewers, agency optimization specialists, and brand compliance staff. Daily processed documents include ad creative copy, landing page screenshots, review opinion sheets, and historical modification records. Data is primarily sourced from ad placement platforms, landing page backends, product qualification systems, and public advertising compliance clauses. Update frequency is daily or real-time. The data structure includes fields such as creative text, visual element annotations, violation point descriptions, and modification timeliness. There are no unified fixed units. Values adjust dynamically based on creative type and review requirements.
What constraints do these characteristics impose on the multi-turn dialogue and prompt words workflow
Long text documents involved in ad creative review cause excessive single-turn dialogue parsing overhead. Multi-turn dialogue must optimize context truncation rules to avoid losing critical compliance information. Real-time updated compliance clauses require prompt words to synchronize the latest data sources regularly. This ensures the timeliness of review standards. Multi-turn review processes track historical modification records. So variables must be bound to the current review’s creative ID and status. This ensures coherent dialogue context. The consequences of errors include ad account restrictions and administrative penalties. Prompt words must accurately mark violation points without omissions or misjudgments.
How to configure the settings
This document is based on FastGPT v4.17.0 configuration.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Maximum context length | 8000–12000 characters | Covers core content of multiple documents including ad creatives, review opinions, and modification records. Avoids truncation of critical compliance information |
| Question optimization switch | Enabled | Compliance questions for ad review are mostly decomposed queries. Optimization allows quick location of violation fields and corresponding clauses |
| Single knowledge base retrieval limit | Top 6 entries | Covers advertising law clauses and past review cases. Avoids excessive entries increasing model inference latency |
variables variable configuration | Bind review status and creative type fields | Tracks current review progress of creatives in multi-turn dialogue. Associates historical modification records and ensures coherent context |
messages context retention rule | Retain the last 15 dialogue turns | Aligns with the multi-turn modification workflow of ad reviews. Avoids redundant historical dialogue interfering with current review judgments |
chatId session isolation configuration | Create independent sessions per creative | Avoids confusion of review contexts across different creatives. Ensures each dialogue only associates with the current review task |
Three common configuration errors
- Phenomenon: Excessive time spent parsing conversations when processing large files (10+ MB Word ad creatives). Cause: Failed to adjust chunking rules for long-text creatives, resulting in oversized chunks or excessive chunk counts, increasing parsing and retrieval overhead.
- Phenomenon: Debug preview interface throws errors after enabling input guidance and configuring a custom lexicon. Cause: Incorrect custom lexicon address format, or lexicon content not adapted to compliance fields for ad creative review, leading to loading failure.
- Phenomenon: Response delay exceeds expectations after prompt word optimization. Cause: Failed to limit the number of knowledge base entries retrieved per session, resulting in overloaded context and increased model inference time.
How to confirm the configuration is correct
- Upload a single Word document of ad creative over 10 MB, observe parsing and first response latency, adjust configuration items until the latency meets scene expectations.
- Initiate a multi-turn review dialogue, verify that the
variables-bound review status field is correctly called in the dialogue, and track the current creative’s review progress. - Enable the question optimization function, enter a compliance question, check if the response quickly decomposes into specific violation points, and confirm the optimization logic is effective.
- Verify the
messagescontext retention rule: delete historical dialogue and confirm that the new dialogue only loads current session content, avoiding redundant context interference. - Create an independent session per creative, verify that
chatIdcorrectly isolates review contexts across different creatives, avoiding session data confusion.
The values provided are common starting points and should be measured 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.