Query Generation logs

Every call to a Query Generator and its results are logged in the Query Generation Log [sn_query_gen_log] table.

Using the logs

Go to Query Generation > Administration > Logs to inspect failed or incorrect queries.

1. Find the entry by user, date, or utterance.
2. Check Semantic Entities/Dimensions/Segments Count
3. Open Query Constitutor Output to see what was selected.
4. Compare with Semantic Filter Result for raw semantic matches.

Table 1. AI Data Explorer troubleshooting guide

Symptom	Likely cause	Fix
"Unable to understand" response	System cannot find a matching table	Improve entity description; rephrase with table name; verify AI search is running; verify that all Indexed Sources are fully processed.
Wrong table selected	Similar table names, no disambiguation	Add semantic usage instructions to differentiate.
Wrong table selected (permissions)	User lacks read access to the intended table or fields	Verify and grant table and field read ACLs for the intended data.
Missing field in results	Dimension turned off, unclear label, or field not in semantic layer	Check dimension is active; update label.
Related fields not pulled in	Referenced table missing from semantic layer or reference dimension turned off	Add and enable the referenced table entity; enable the reference dimension.
Wrong filter applied	Bad segment match	Fix segment name and description; disable the segment.
No filter when expected	No matching segment; phrasing too vague	Create a manual segment; rephrase question.
Segment from report is wrong	Auto-generated segment has stale or irrelevant filter	Disable the report segment source or the specific segment.

Table 2. Query Generation log columns

Column	Description	Shown by default?
Created	Timestamp of the call to Query Generation	Yes
User	The user from the User [sys_user] table who made the call. If this is blank, try Created by.	Yes
Utterance	The plain language question asked by the user	Yes
Status	The success or otherwise of the call	Yes
Error message	If the status is "error," this is the error message that is returned to the user.	Yes
Total time	Time in milliseconds it took for the AI to process the query and return the response	Yes
"X" time	Time in milliseconds each component X spent processing the query	Yes
Semantic Filter Results	Fields in the semantic layer tables that the query references, in JSON	Yes
Utterance Bank Results	Questions similar to the user's question (Utterance) that were found in the Utterance Bank and used to interpret the user's question, in JSON	Yes
Response	The JSON response to the query	Yes
Entities	The records in the Entities table that are referenced by the query. Each record corresponds to a facts table.	Yes
Query Constitutor Output	JSON of the query that is generated	Yes
Source	The application from which the question was asked. If the question was asked from the Now Assist panel when it was open in a page that is not part of an application, the UUID of that page is shown instead.	Yes
Response	The response that the LLM returns	Yes
Generative AI Log	The LLM used for the query	Yes
Conversation	Unique chat conversation ID, such as in a Now Assist Panel to Query Generation conversation	Yes
ACL Time	Time it took to validate the Access or Security of the user for the underlying table	No
Additional info	Start time, total calls, and other metadata about the AI Search run	No
Created by	Person who created the query. Includes ServiceNow® staff members logged on as maint.	No
Domain	Domain in which the query was created	No
Dot walked	Whether the query included dot walking between tables (true/false)	No
Entity has usage score	Whether the entity has a usage score (true/false)	No
Entity source	Whether the Source was in ENTITIES, USER SENSE, or SEGMENTS	No
Error code	The error code, if there was an error	No
Follow Up Generated	The number of follow-up questions that the LLM generated in response to the query	No
Follow Up Returned	How many of the generated follow-up questions are actually returned. The questions must refer to tables that the querying user can access and are in the semantic layer, for example.	No
LLM Time	Time spent by the large language model (LLM) on the query	No
Query Constructor Time	Time spent by the query constructor on the query	No
RAG time	Time spent by the retrieval-augmented generation (RAG) on the query	No
Semantic Dimensions count	Number of table records (dimensions) examined for the query	No
Semantic Entities Count	Number of tables (entities) examined for the query	No
Semantic Filter Results	Results of running the semantic filter	No
Semantic Filter Time	Time spent by the semantic filter	No
Semantic Segments Count	Number of segments examined for the query	No
Source Table	Name of table corresponding to Source Id	No
Tags	Any tags that you add to the log	No
Updated	Date stamp when log entry was last updated	No
Updated by	User responsible for last log entry update	No
Updates	Count of updates to the log entry	No
Utterance bank results	Similar examples extracted from utterance bank	No
Utterance bank time	Amount of time spent searching the utterance bank	No

Related concepts

• Common AI Data Explorer issues and fixes