Determining generative search results document for queries using generative artificial intelligence models and arbitration models
A generative document system creates generative search results documents using generative artificial intelligence (AI) models and dynamically determines which one of the generative search results documents to provide in response to a search query. For example, in response to receiving a search query, the generative document system obtains search link results (e.g., website links and corresponding grounding information) for the search query and utilizes this information with multiple generative AI models to generate various types of generative search results documents. Additionally, the generative document system generates and utilizes a generative document arbitration model to determine, based on the search link results, which of the generative search results documents to provide in response to the search query.
1 . A computer-implemented method for generating one or more search results documents using one or more artificial intelligence (AI) models, comprising:
in response to receiving a search query, obtaining search link results, including website links and grounding information corresponding to the website links, from a search system;
generating multiple generative search results document prompts to provide to multiple generative AI models for generating multiple generative search results documents concurrently from the search query and the search link results;
selecting a selected generative search results document from the multiple generative search results documents utilizing a generative search results document arbitration model; and
providing the selected generative search results document in response to the search query.
2 . The computer-implemented method of claim 1 , wherein the generative search results document arbitration model includes a lightweight machine learning model trained to determine a presentation quality score for generative search results documents based on corresponding search queries.
3 . The computer-implemented method of claim 2 , wherein selecting the selected generative search results document from the multiple generative search results documents includes:
generating a first presentation quality score for a comprehensive summary generative search results document using the generative search results document arbitration model based on the search query;
generating a second presentation quality score for a visual digest generative search results document using the generative search results document arbitration model based on the search query;
selecting the comprehensive summary generative search results document as the selected generative search results document when the first presentation quality score is more favorable than the second presentation quality score; and
selecting the visual digest generative search results document as the selected generative search results document when the second presentation quality score is more favorable than the first presentation quality score.
4 . The computer-implemented method of claim 3 , further comprising removing a generative search results document of the multiple generative search results documents from selection consideration based on determining that the generative search results document has a presentation quality score that does not meet a presentation quality threshold.
5 . The computer-implemented method of claim 1 , wherein the generative search results document arbitration model includes a generative AI model generated to determine a presentation quality score for generative search results documents based on corresponding search queries.
6 . The computer-implemented method of claim 1 , wherein:
the generative search results document arbitration model includes a set of heuristics for selecting the selected generative search results document; and
the set of heuristics includes a ranking of generative search results documents according to available generative search results documents.
7 . The computer-implemented method of claim 1 , further comprising determining that a cached generated document corresponding to the search query and the search link results is not available, wherein cached generative search results documents include previously generative search results documents for search queries previously generated by a generative AI model to populate curated content formatted in a visually enhanced layout.
8 . The computer-implemented method of claim 7 , further comprising determining that an additional cached generated document for an additional search query is available by:
querying a generative search results document store with the additional search query and additional search link results corresponding to the additional search query;
determining a correlation score between the additional search query and the additional search link results with a set of stored generative search results documents; and
selecting a cached generative search results document from the set of stored generative search results documents based on the cached generative search results document having a highest correlation score that also satisfies a correlation threshold.
9 . The computer-implemented method of claim 1 , wherein:
a first generative AI model of the multiple generative AI models generates a comprehensive summary generative search results document;
a second generative AI model of the multiple generative AI models generates a visual digest generative search results document; and
a third generative AI model of the multiple generative AI models generates a direct answer generative search results document, wherein the multiple generative search results documents differ from each other.
10 . The computer-implemented method of claim 1 , further comprising storing the selected generative search results document in a generative search results document store with the search query and the search link results.
11 . The computer-implemented method of claim 1 , further comprising:
obtaining additional search link results in response to receiving an additional search query;
generating multiple additional generative search results document prompts to provide to the multiple generative AI models for generating multiple additional generative search results documents from the additional search query and the additional search link results;
based on determining, by the generative search results document arbitration model, that the multiple additional generative search results documents do not meet a presentation quality threshold, generating a visually-based search results document from the search link results; and
providing the visually-based search results document in response to the additional search query.
12 . The computer-implemented method of claim 1 , further comprising:
obtaining additional search link results in response to receiving an additional search query;
determining that an additional cached generated document corresponding to the additional search query and the additional search link results is available; and
providing the additional cached generated document in response to the search query without generating multiple additional generated documents using the multiple generative AI models.
13 . A system comprising:
a processing system; and
a computer memory comprising instructions that, when executed by the processing system, cause the system to perform operations of:
in response to receiving a search query, obtaining search link results, including website links and grounding information corresponding to the website links, from a search system;
generating multiple generative search results documents by:
generating a first generative search results document prompt to provide to a first generative AI model for generating a comprehensive summary generative search results document from the search query and the search link results;
generating a second generative search results document prompt to provide to a second generative AI model for generating a visual digest generative search results document from the search query and the search link results; and
generating a third generative search results document prompt to provide to a third generative AI model for generating a direct answer generative search results document from the search query and the search link results, wherein the first generative search results document prompt, the second generative search results document prompt, and the third generative search results document prompt are provided concurrently;
selecting a selected generative search results document from the multiple generative search results documents utilizing a generative search results document arbitration model; and
providing the selected generative search results document in response to the search query.
14 . The system of claim 13 , wherein the selected generative search results document includes:
a set of suggested follow-up search queries corresponding to content included in the selected generative search results document; and
a follow-up search query field for receiving a follow-up search query corresponding to the selected generative search results document.
15 . The system of claim 13 , further comprising:
storing the search query and context information corresponding to the selected generative search results document;
receiving a follow-up search query from the selected generative search results document;
generating multiple additional generative search results document prompts to provide to multiple generative AI models for generating multiple additional generative search results documents concurrently from the follow-up search query and additional search link results, wherein the multiple additional generative search results document prompts also include the search query and the context information corresponding to the selected generative search results document from a previous search query request;
selecting an additional selected generative search results document from the multiple additional generative search results documents utilizing the generative search results document arbitration model; and
providing the additional selected generative search results document in response to the follow-up search query.
16 . The system of claim 15 , wherein the context information includes information corresponding to the search query, the search link results, the website links, the grounding information, and content of the selected generative search results document.
17 . The system of claim 15 , wherein providing the additional selected generative search results document includes appending the additional selected generative search results document to the selected generative search results document from the previous search query request.
18 . The system of claim 16 , wherein the additional selected generative search results document is a different generative search results document type from the selected generative search results document.
19 . A computer-implemented method for generating one or more search results documents using one or more artificial intelligence (AI) models, comprising:
in response to receiving a search query, obtaining search link results, including website links and grounding information corresponding to the website links, from a search system;
determining that a cached generated document corresponding to the search query and the search link results is not available by:
querying a generative search results document store with the search query and the search link results; and
receiving a negative response regarding an available matching cached generated document;
based on the cached generated document being unavailable, generating multiple generative search results document prompts to provide to multiple generative AI models for generating multiple generative search results documents concurrently from the search query and the search link results;
selecting a selected generative search results document from the multiple generative search results documents utilizing a generative search results document arbitration model that determines a presentation quality score for each of the multiple generative search results documents; and
providing the selected generative search results document in response to the search query.
20 . The computer-implemented method of claim 19 , further comprising:
receiving the search query from a client device;
determining the search link results using a search index; and
providing the selected generative search results document within a webpage to the client device in response to the search query.