serp is a Google search results service for application development and information retrieval, not a conversational generative model. It accepts keywords or query statements and returns program-friendly JSON, covering web, image, news, video, place, and map searches. Combined with controls for region, language, time range, and pagination, it is suitable for building research discovery, keyword monitoring, and AI assistant retrieval workflows.
Clarify capacity, inputs and outputs, and calling methods before choosing a model.
The above are the calling specifications for the serp search endpoint, not the context or output capacity of a generative model; the actual number of results and additional information vary by query.
Learn what serp can bring to your work.
A single entry point lets you switch between web, images, news, video, places, and maps, without assigning every task to general web search. Use search to look up company information, news to find coverage, and places to discover local businesses, making your input goal better match the result category.
Search results are returned as JSON, making it easy to extract titles, links, snippets, and ranking positions, then continue with deduplication, display, or storage. General search may also return related questions, related searches, and a knowledge graph, helping applications expand query directions instead of receiving only a block of text that is difficult to separate.
With country, language, and range, you can organize searches around target markets and time windows. For example, set the country and language separately for the same brand, then query information from the past day or week. page and number control pagination and request volume, making it easier to collect results according to task scale.
Start with specific tasks to find where the model can be effective.
Enter brand terms, product terms, or industry keywords, and save the titles, links, and positions from web results to create search records for comparison. When querying regularly, keep region, language, and pagination conditions consistent to help monitor competitor pages and keyword performance; the application organizes the recording and comparison logic itself.
Use a company name or event topic as the query, select news, and set a time range to organize returned titles, snippets, links, and available dates into a list of coverage leads. Suitable for editorial planning and brand information monitoring; you still need to open the original articles to verify details rather than treating short snippets as complete reports.
Have the assistant first turn the user's question into a query, then provide the search results to a language model for summarization and citation organization. MCP-supported clients can also use categorized search tools through SerpMCP. serp is responsible for discovering information and returning links; the assistant workflow handles final answers, web page reading, and multi-step execution.
Choose based on task complexity, input materials, and expected results.
If a task requires internet links, search snippets, result positions, or related queries, serp is a better fit than simply asking a generative model to answer based on existing knowledge. If the deliverable is a long-form article, explanation, or comprehensive analysis, you can first use serp to obtain materials, then pass them to a language model for processing; serp itself does not generate articles, nor does it automatically read all matched web pages.
The key to choosing serp is the search mode, not a version upgrade of a generative model. For general information discovery, use search; for recent reports, use news; for image and video discovery, use images and videos respectively; for local businesses, use places or maps. The direct API is suitable for embedding in programs, while MCP is suitable for invocation by compatible clients; neither represents a different model generation.
From a small-scale task to full integration.
Define the objective, required inputs, and output requirements, and use real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm the billing rules on the Pricing page.
Example inputs and verification methods to help you design your first trial.
Start with a clear query, such as “vector search API documentation.” Choose language and region conditions based on the audience, review a small number of results first, then decide whether to expand the query. When you need results such as news or images, choose the search type supported by the API rather than mixing different result formats.
Check whether result titles, links, and snippets match the query intent, and open important sources to confirm their content and dates. When retrieving paginated results, use pagination parameters as specified in the documentation; applications can further deduplicate and sort results, but search results themselves are not the same as verified facts.
Before formal use, understand the output quality and scope of capabilities.
Answers to common questions about using serp.
No. serp is a Google search results service that returns structured search data after a query is entered, rather than generative conversation. It is suitable for finding web pages, news, or other resources; when explanations, writing, and integrated judgment are needed, search results can be passed to a language model for further processing.
Send a POST request to /serp/google, use a Bearer Token, and provide query in the JSON request body; no model is required. The default type is search, the page number is 1, and the requested quantity is 10; the query cannot contain only whitespace characters, and undefined parameters should not be added.
Set type to news, fill query with an event or brand keyword, and use d or qdr:d for range. When targeting a specific market, you can also set country and language. Returned content is for discovering recent coverage; specific publication times and event details should still be confirmed against the original text.
No, serp searches existing images and does not generate new ones. When type is images, use image_size to set large, medium, icon, or supported mp filter levels; other search types cannot include this field. Image links in the results also do not mean usage authorization has been obtained.
A single search request will not automatically complete web page reading or multi-step research. serp returns links, summaries, and available additional information, which applications can use to choose subsequent reading targets. When used through MCP, search is only one of the assistant's tools; the complete task is still organized and executed by the client.
Model information · Updated: 2026-10-01. For request parameters and billing rules, see the API and pricing sections.
Start with a clear goal and see how it fits your work with real results.
