Sign in
← back
AI Mode Site Search

AI Mode Site Search

Discovery

Explore Vertex AI website search features, including Enterprise edition tools like extractive answers, image search, and advanced LLM capabilities for summaries.

Tharindu Gunawardana from Search Ministry has discovered a cool feature in Vertex AI:

Looks like we can now test and fine-tune AI Mode in Vertex AI > Vertex AI Mode, using our own website content. This also confirms the process of how AI Mode works (Simplified version )
Prepare > Retrieve > Signal > Serve

I’ll be investigating it and documenting my progress on this page.

Google Cloud console interface showing option cards for creating a new search application, including site search with AI mode.

Website search app configuration

Enterprise edition features

In addition to the standard features, you get:

  • Extractive answers – answers that are extracted verbatim from your documents
  • Image search, where you can use an image as a query
  • Website search
  • Core generative answers

Turning on Enterprise edition features is required for website search. To get higher refresh frequency, lower latency, search summaries and more features in addition to website data, you need to turn on advanced website indexing.

You can change this setting at any time.

After turning on Enterprise features, it can take up to five minutes for the features to become available.

Learn more about features and prices 

Advanced LLM features

For structured, unstructured and advanced website search, you get:

  • Search summarisation
  • Search with follow-ups
  • Advanced generative answers

Advanced LLM features are not available for basic website search.\n You can change this setting at any time.

After turning on advanced LLM features, it can take up to five minutes for the features to become available.

Learn more about features and prices 

Basic Setup Steps

Google Cloud console interface showing data source selection options for creating a new data store.
A user interface screenshot shows the Create a data store configuration page in Google Vertex AI.
A screenshot of the Google Vertex AI data store configuration interface showing settings for location and data store name.
A screenshot of the Google Cloud console showing the app creation screen with two data stores listed in a table.
A Google Cloud console interface showing the four stages of a website search application: Prepare, Retrieve, Signal, and Serve.
Google Cloud console interface showing configuration settings and a preview pane for a website search application powered by Gemini.
Search type

Select the type of search experience that the widget provides.

Search with follow-ups

A screenshot of search type configuration options in Vertex AI, highlighting the Search with follow-ups selection.
Large language models for summarisation

Try out different models to see which one works the best for youStable

Default model. Learn more about the current default model Gemini 2.0 Flash 1 new

The Vertex AI gemini-2.0-flash-001 model  with additional training for Q&A taskGemini 1.5 Flash 2

The Vertex AI gemini-1.5-flash-002 model  with additional training for Q&A taskGemini 1.5 Flash 1 V2

The Vertex AI gemini-1.5-flash-001 model  with additional training for Q&A task (version 2)Gemini 1.5 Pro preview

The Vertex AI 1M token context window model. Learn more about Gemini 1.5 Pro 

Customise the summary

Give instructions to customise the tone, style and verbosity of the generative summary

Summary language preview

Choose the language of the summary

English

Enable related questions

Controls whether the generative summary suggests related questions.

Ignore no answer summary for query

Don’t show ‘We do not have a summary for your query’ if the query returns no answers for non-product questions on the widget.

Ignore adversarial query

Prevent LLM answers on adversarial queries.

Ignore low relevant content

Prevent LLM answers with low relevant content.

Image in answers preview

Choose a source for the image returned in the answer

A website search preview interface displaying a search query with no results found and suggested follow-up questions.

No results are available until Google’s datastore indexes the content.

Dan Petrovic · Jun 04, 16:01