Vertex AI Search : How to query search results data introduced into Vertex AI Search

[eunheejang]

hello.

I have a question about vertex ai search.
Please understand that this is my first time using the community, so I’m not good at it.

I’d like to see the answers users search in the app, and the ai has generated summaries in the log.
Although I see some information in the ‘Analytic’ tab, only ‘Top 100 Search Queries by Search Count’ is searched, and I don’t see the entire data.

Is there a way to accumulate and check the data below?

• Search Contents
• Contents of the answer
• Whether the answer fails or not
• User feedback

[Farisds]

@eunheejang

hello.

I have a question about vertex ai search.
Please understand that this is my first time using the community, so I’m not good at it.

I’d like to see the answers users search in the app, and the ai has generated summaries in the log.
Although I see some information in the ‘Analytic’ tab, only ‘Top 100 Search Queries by Search Count’ is searched, and I don’t see the entire data.

Is there a way to accumulate and check the data below?

• Search Contents
• Contents of the answer
• Whether the answer fails or not
• User feedback

Welcome to the community! Don’t worry, your question is very clear and it is a common challenge for those moving from testing to production with Vertex AI Search.

You are correct—the ‘Analytics’ tab in the console is designed for a quick overview (Top 100), not for deep auditing. To get the full 4 items you mentioned (Queries, Summaries, Failure status, and Feedback), you should export your logs to BigQuery.

Here is the architectural approach I recommend:

1. Enable BigQuery Export: In the Vertex AI Search console, go to the “Monitoring” or “Settings” section and enable the export of logs to BigQuery. This will start accumulating every single request and response in a structured table.

2. Analyzing the Schema: Once the data is in BigQuery, you can query the following fields:

   • Search Contents: Look for the query field.

   • Contents of the Answer: Look for the answer or summary field within the JSON payload.

   • Failure Status: Check the error or status codes.

   • User Feedback: If you have implemented the feedback API (thumbs up/down) in your app, these events are also captured in the user_feedback field.

3. Data Integrity Tip: At Whitecyber Data Science Lab, we often use this raw data to perform “Hallucination Audits.” By having the full logs in BigQuery, you can run SQL scripts to identify patterns where the AI might be failing or providing low-quality summaries, which you can’t see in the Top 100 dashboard.

4. Visualization: Once the data is in BigQuery, you can easily connect it to Looker Studio to create a custom dashboard that shows 100% of your data, not just the Top 100.

Hope this helps you get the full visibility you need for your app!

Any Question so far ?