Gemini Scraper

This repository is a technical, developer-focused reference for the Google Gemini API target available through the Oxylabs Web Scraper API. It explains what the target does, how requests and responses are shaped, and how the Google AI Gemini target differs from other LLM and AI scraper targets.

The gemini source lets you submit a prompt to Gemini Google and receive a parsed, structured response – including plain text and Markdown answers, plus source citations. It is intended as a single reference point you can link from other repos, docs, and discussions when working with Gemini data.

What is Google Gemini?

Google Gemini is Google's conversational AI assistant. Like other large language models, it takes a natural-language prompt and returns a generated answer, and it can reference external sources in its responses.

The Oxylabs Gemini target wraps this interaction behind a single API call. Instead of returning raw HTML that you would have to render and parse yourself, the gemini source can return a structured JSON object containing the prompt, the plain-text answer, the same answer in Markdown, and any source citations Gemini included. JavaScript rendering, proxy rotation, and server-side defense handling are managed by the API, so you provide a prompt and credentials rather than maintaining browsers and proxies. Check out our Web Scraper API documentation for in-depth knowledge about each feature.

How it works

Submit a prompt together with valid Web Scraper API credentials, then make a POST request to the /v1/queries endpoint. The example below uses the synchronous Realtime integration method.

Tip

Get a free trial of Oxylabs Web Scraper API by registering on the Dashboard.

Request sample (Python)

import requests
from pprint import pprint

# Structure payload.
payload = {
    'source': 'gemini',
    'prompt': 'best supplements for better sleep',
    'parse': True,
    'geo_location': 'United States'
}

# Get response.
response = requests.request(
    'POST',
    'https://realtime.oxylabs.io/v1/queries',
    auth=('USERNAME', 'PASSWORD'),
    json=payload,
)

# Print prettified response to stdout.
pprint(response.json())

The documentation also provides equivalent examples for cURL, Node.js, PHP, Go, C#, Java, and a plain URL/JSON payload. Find them in the Gemini documentation.

The examples use the Realtime integration method. You can also use the Proxy Endpoint or asynchronous Push-Pull methods.

Request parameters

Basic setup and customization options for the gemini source:

Parameter	Description	Default value
source (mandatory)	Sets the scraper target. Use gemini.	–
prompt (mandatory)	The prompt or question to submit to Gemini. Must be under 8000 characters.	–
parse	Set to true to return structured JSON data.	false
geo_location	Specify a country to route the request from.	–
callback_url	URL to your callback endpoint (used with Push-Pull).	–

Key Gemini features

The Gemini target exposes three capabilities worth calling out:

• Prompts. You pass your question or instruction through the prompt parameter. A single prompt must be under 8000 characters. This is the only required input beyond source.
• geo_location targeting. You can specify a country to route the request from, so responses reflect the location you care about rather than a default region. Note that Gemini's response composition may vary depending on whether the query was made from a desktop or mobile device.
• Citations. When Gemini references external sources, the parsed output includes a citations array. Each citation object contains a title, url, text, and description, so you can trace which sources informed the answer.

Google Gemini vs ChatGPT: how the targets differ

Gemini and ChatGPT are two of several LLM and AI targets available through the Web Scraper API. Both are prompt-in, structured-data-out sources built on the same request pattern: the same /v1/queries endpoint, the same authentication, the same top-level job and results[] envelope, and the same parse_status_code convention (12000 for a successful parse). Each target then defines its own content schema and its own set of parameters.

The table below is a side-by-side specification reference, not a ranking. Which target fits your use case depends on whose responses you want to observe.

Aspect	Google Gemini (gemini)	ChatGPT (chatgpt)
source value	gemini	chatgpt
Prompt parameter	prompt, within 8000 characters	prompt, within 4000 characters
Web search trigger	–	search – triggers a web search for the prompt (default true)
geo_location targeting	Supported – specify a country to route the request from	Supported – specify a country to send the prompt from
JavaScript rendering	Enabled by default; no rendering parameter needed	Enforced by default; no rendering parameter needed
parse	Set to true for structured JSON	Set to true for structured JSON
callback_url	Supported (Push-Pull)	Supported (Push-Pull)
Citations	citations array with title, url, text, description	citations array with URL and text
Batch requests	Up to 5000 requests	Up to 5000 requests
Markdown output	markdown_text	markdown_text, plus markdown_json
Plain-text output	response_text	response_text
Model identifier in output	–	llm_model
Integration methods	Realtime, Proxy Endpoint, Push-Pull	Realtime, Proxy Endpoint, Push-Pull

Compared with other LLM and AI targets, the practical distinctions are in the per-target content schema (field names and the extra data each returns) and in target-specific limits and parameters. For more information on LLM scraping targets, refer to our documentation.

Output samples

With parse set to true, the API returns a structured JSON object containing the Gemini output. A trimmed snippet looks like this:

{
  "results": [
    {
      "job_id": "7470031540181837825",
      "status_code": 200,
      "url": "https://gemini.google.com/app/ca6b069d5af5c809",
      "content": {
        "prompt": "best supplements for better sleep",
        "response_text": "If you are tossing and turning, you are definitely not alone. The supplement aisle is packed with options...",
        "markdown_text": "If you are tossing and turning, you are definitely not alone. The supplement aisle is packed with options...",
        "citations": [
          {
            "title": "Hartford HealthCare",
            "url": "https://hartfordhealthcare.org/about-us/news-press/news-detail?articleId=66180",
            "text": "Can These 3 Supplements Really Improve Your Sleep? | Hartford HealthCare | CT",
            "description": "- Valerian root. This herbal sleep option likely works by acting on GABA receptors in the brain..."
          }
        ],
        "parse_status_code": 12000
      }
    }
  ]
}

You can also work with the answer in Markdown (markdown_text) for easier integration into AI and data workflows.

Output data dictionary

All LLM targets share the same top-level job and results[] envelope. The fields below are the Gemini-specific results[].content fields:

Field	Description	Type
prompt	The submitted prompt.	string
response_text	Plain-text response from Gemini.	string
markdown_text	Complete response in Markdown.	string
citations	Source citations referenced in the response. Each object has title, url, text, description. Returned only when present in the response.	array
parse_status_code	12000 indicates success. Any other value means the parser failed to extract some or all structured fields.	integer

Notes and limitations

• JavaScript rendering is enabled by default for all LLM sources, so you do not need to include rendering parameters in the payload.
• Batch requests are not supported for the gemini source.
• Desktop vs mobile the composition of the response may vary depending on whether the query was made from a desktop or mobile device.
• The citations field is conditional and is returned only when citations are present in Gemini's response.

Related Oxylabs LLM and AI scrapers

Gemini is part of a family of LLM and AI targets. If you are researching AI scrapers more broadly, these related repositories use the same Web Scraper API patterns:

• ChatGPT Scraper – send prompts to ChatGPT and collect responses plus structured metadata.
• Perplexity Scraper – send prompts to Perplexity and collect AI-generated answers and structured metadata.

Learn more

For the full parameter list, all language examples, and additional details, see the official Gemini documentation.

Contact us

If you have questions or need support, reach out at support@oxylabs.io, or via live chat in the Oxylabs Dashboard. For enterprise inquiries, contact your dedicated account manager.