Module Scope

As explained on the Background page, Google recommends that websites using the Site Restricted JSON API to integrate with Google's Programmable Search Engine solution should move to Google's Vertex AI Search instead.  The vertex_ai_search module provides a replacement to the google_json_api module in support of this recommendation.

Google's Vertex AI Search solution provides features that go beyond the traditional search functionality of Google's Programmable Search Engine.  While the vertex_ai_search module will evolve, it's initial scope is meant to be comparatively narrow to the offerings of Google's Vertex AI Search. 

As a starting point, the vertex_ai_search module will provide search functionality that is commensurate with the search functionality currently provided by the google_json_api module.  There are a lot of Vertex AI Search features that will not initially be considered in deference to this initial goal.

Broadly speaking, the initial scope of the vertex_ai_search module allows for the creation of Drupal search pages that communicate with website search apps on the Google Cloud Platform (GCP).  A Service Account is the required mechanism to authenticate with a GCP project.  Features that require Advanced Website Indexing are not initially supported.  Advanced LLM features are also not initially supported by the vertex_ai_search module.

Updates to scope will be noted on this page as the module progresses.  A rough breakout of options/features/configuration provided by Google's Vertex AI Search follows.

---

App Options

App types

With Vertex AI Search, two types of apps can be created: Search apps or Recommendations apps.  

Further, there are three types of Search apps: Generic search, Media search, and Healthcare search.

Scope: The vertex_ai_search module will initially only focus on Generic Search apps.

Data stores

Each data store can contain only one of the various types of data supported.

Scope: The vertex_ai_search module will initially only focus on data stores containing website data.

Enterprise edition features

More advanced features are available when Enterprise edition features is enabled.  

Scope: Enterprise edition features is required for Search apps that use website data stores.  Although it must be enabled, not all of the features provided by Enterprise edition features will be included in the initial scope of the vertex_ai_search module.  

Advanced website indexing

The use of Advanced website indexing requires domain verification and imposes a quota on the number of pages in the index.  While Advanced website indexing does provide some benefits and advanced functionality, it is not necessary for meeting the initial goal of the vertex_ai_search app.

Scope: Advanced website indexing and the additional functionality it provides are not part of the initial scope of the vertex_ai_search module.  The initial goal of providing search functionality commensurate with that provided by the google_json_api module can be met without using Advanced website indexing.

Advanced LLM features

The Advanced LLM features add generative AI functionality at the app level.

Scope: Advanced LLM features does not need to be enabled.  Generative content is not within the initial scope of the vertex_ai_search module or the custom search pages that it creates.

---

Search form and search result page (SERP) options

Snippets

Snippets are the text extracts that are provided with each search result.  The keywords are automatically highlighted.

Scope:  Snippets are in scope and are available for presentation with each search result. 

Extractive answers

These are generally displayed at the top of a search results page.

Scope:  Out of initial scope for vertex_ai_search.  This feature would require Advanced website indexing and does not contribute to functional parity with google_json_api module.

Extractive segments

Similar, but more complete than extractive answers.

Scope:  Out of initial scope for vertex_ai_search.  This feature is only available for search apps with unstructured data stores.  

Summarizations

AI-generated summarization of search results displayed at the top of a search results page.

Scope: Out of initial scope for vertex_ai_search.  Generative content and functionality requiring Advanced LLM features are out of scope.

Results ordering

Ordering options are limited with basic indexing and will not work with advanced search indexing. 

Scope: Although, the google_json_api module provided for some ordering/sorting abilities, this capability is not being included in the vertex_ai_search module.

Autocompletion

Vertex AI Search provides different autocomplete models.  The models that can be used depends on the type of data used by a search app.

For website search apps, the default autocomplete model is 'search-history' which requires a few days' worth of real traffic to the website.

Scope: Autocomplete is in the initial scope of the vertex_ai_search module.  The 'search-history' model will be supported, but a framework will be put in place to be able to develop support for other models using plugins.

---

Search request / query options

Search with follow-ups

Scope: Out of initial scope for vertex_ai_search.  Functionality requiring Advanced LLM features and Advanced website indexing are out of scope.

Blended search

Scope: Out of initial scope for vertex_ai_search.  Features that require Advanced website indexing are out of scope.

Image search

Scope: Out of initial scope for vertex_ai_search.  Features that require Advanced website indexing are out of scope.

Filter

The filter syntax varies depending on whether or not Advanced website indexing is enabled.  Without it enabled, there is limited filtering capability and it may not be especially valuable.

Scope: Out of initial scope for vertex_ai_search.

Facets

Facets are not supported for apps with website data.

Scope: Out of initial scope for vertex_ai_search.

Serving Controls

The results of a request can be altered through the use of serving controls.  There are currently four control types: Boost, Filter, Synonyms, and Redirect control.  Only the Redirect control is applicable to website search apps.

Scope: Out of initial scope for vertex_ai_search.  While similar Redirect functionality does exist in the Programmable Search Engine, support for it was never implemented in the google_json_api module.  

Field Settings

Field settings are used for structured data stores and unstructured data stores with metadata.  These settings define how fields can be used in Vertex AI Search app results.

Scope: Out of initial scope for vertex_ai_search.  Not applicable for website search apps.

Search Tuning

Search tuning helps train the search model for better quality results.  This is only applicable to unstructured data sources.

Scope: Out of initial scope for vertex_ai_search. 

SafeSearch

Scope: In scope. Search page configuration will allow this to be enabled or disabled.

Spelling Correction

Scope: In scope.  Search page configuration will allow for automatic spell correction or for just making a suggested correction.

Promoted Search Results

This is a feature of Google's Programmable Search Engine solution that does not have a direct counterpart in Vertex AI Search. 

To support this feature, a custom plugin type (SearchResult) has been defined.  Other modules can create a SearchResult plugin to manipulate the search results that have been retrieved.

The Vertex AI Search Promoted Results module provides the PromotedSearchResult plugin that retrieves a promoted search result (based on the search keyword) and displays it at the top of the search results page.

Scope: Support for promoted results is in the initial scope for vertex_ai_search. 

---

Authentication

Service Account

For websites hosted on-premises or on a provider other than Google Cloud, credentials recognized by Google Cloud must be provided.  This can be accomplished using Workload identity federation or a Service account key. 

Scope: Service account keys are in the initial scope for the vertex_ai_search module.  It will be the only form of authentication supported initially.