Frequently Asked Questions
This section captures common questions and operational guidance for Oracle Trusted Answer Search.
Why not just use a generic LLM Agent?
While LLMs are flexible for open-ended Q&A, they can hallucinate, return inconsistent answers, and exhibit longer response times. Trusted Answer Search focuses on deterministic, vetted matches: it maps questions to curated descriptions and returns the top search targets. Your application then renders the correct report or initiates the correct action with confidence.
What benefits can I expect?
- Faster access for your end-users to the right page, report, or action via natural language.
- Consistent outcomes because a deterministic matching algorithm is used, not LLMs.
- Stronger security posture: search targets are vetted, and your application retains full control over authorization.
- Lower operational risk: approvals, versioning, and regression reports make language changes auditable and predictable.
What does Trusted Answer Search return—an answer or something else?
Trusted Answer Search returns a structured response in JSON format to your application. The JSON document includes the search target’s URL or SQL along with the target inputs. Your application remains in control of how the JSON response is used and what is rendered to the user.
How can I use Trusted Answer Search?
Trusted Answer Search can be managed and integrated into your enterprise application in different ways, depending on how much UI you want to build yourself.
Administration
- Trusted Answer Search Administrator (APEX app): used by administrators and Search Space experts to curate targets, manage versions, and control access.
Integration with your Application
- Trusted Answer Search Portal (APEX app): a ready-to-use end-user experience for issuing natural language queries.
- Using the API directly: Developers who need a custom UI can call the Trusted Answer Search PL/SQL API. The core
DBMS_TRUSTED_SEARCH.SEARCH()PL/SQL procedure accepts a natural language question and returns a structured response in JSON format.
How does Trusted Answer Search handle value-specific questions (categories, names, etc.) present in the question?
When configured correctly by a Search Space Expert, Trusted Answer Search can infer parameters mentioned in the natural language query and include them in the returned response. For example, if a manager asks an HCM app “What is my current PTO balance?”, the response from Trusted Answer Search could include a deep link such as:
https://hcm-app.my_enterprise.com/deeplink?objType=ABSENCE_BALANCEHere ABSENCE_BALANCE is a parameter value derived from the user’s mention of “PTO”.
Authorization is always enforced by your application—if the user lacks permission to view PTO balances, the HCM app should block access even if the Trusted Answer Search response contains the link.
How does Trusted Answer Search improve over time?
Trusted Answer Search records end-user query history, match outcomes, and user feedback. Search Space Experts can:
- Review incorrect matches and take corrective action: add new descriptions or sample queries to targets, mark a different target as the right match, or create a new target for this match.
- Upvote good matches and downvote bad matches.
- Preview how proposed description edits affect past queries to target matches before you publish the change.
This continuous feedback loop keeps the curated language up to date and reduces regressions.