Gemini Search Scraper: AI Answers, Citations, Brand Tracking
Pricing
from $2.70 / 1,000 answers
Gemini Search Scraper: AI Answers, Citations, Brand Tracking
Google Gemini search scraper for AI visibility tracking. Give prompts, get the Gemini answer text, cited sources with URL and position, and brand mention, rank and citation fields per answer. Share of voice per run. No login, no API key.
Pricing
from $2.70 / 1,000 answers
Rating
0.0
(0)
Developer
The Mine Works
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
16 days ago
Last modified
Categories
Features
Share
From The Mine Works, makers of Amazon Product Scraper and B2B Leads Finder, with over 290,000 runs across 310+ public actors.
Updated 4 Nov 2026: first release. Prompts in, one row per Gemini answer out, with the answer text, every cited source with URL and position, and brand fields you can chart over time. Bookmark this Actor on its Store page so it is one click away next time.
Why choose this actor?
- Built for tracking, not one look.
prompt_idstays the same on every run for the same prompt, country and language, so a spreadsheet or dashboard can follow one prompt month after month.answer_hashchanges when the wording changes. - Are you named, where, and is your site cited. List your brand and your competitors. Each row says whether each brand is named, how often, its rank against the others, the sentence it first appears in, and whether its website is among the sources. A share of voice report for the whole run is saved too. No extra charge.
- Pay per answer, nothing for failures. Prompts Gemini did not answer, refusals and error text are never charged. In our test runs, 40 of 40 US prompts and 19 of 20 prompts from the UK and India returned an answer. No Google account, cookies or API key.
Part of The Mine Works Marketing, SEO and reviews family: Facebook Ad Library Scraper, Google Ads Transparency Scraper, Similarweb Scraper, Google News Scraper, Trustpilot Reviews Scraper, Google Trends Scraper.
Try it in one minute
Paste this into the input (JSON view) and press Start. It asks Gemini one question and checks two brands against the answer. It took 40 seconds in our test run:
{"prompts": ["best crm for a small business"],"brands": ["HubSpot, hubspot.com", "Zoho CRM, Zoho, zoho.com"]}
Put more prompts on separate lines. Eight are asked at once, so 40 prompts took about 6 minutes in our test.
Apify's free plan includes $5 of credit every month, which covers about 1,100 answers at this actor's price.
Copy to your AI assistant
themineworks/gemini-search-scraper on Apify. Asks Google Gemini a list of prompts as a logged out visitor and returns one row per answer with answer_text, sources (position, url, domain, title, section) and brand_mentions, with no login. Call ApifyClient("TOKEN").actor("themineworks/gemini-search-scraper").call(run_input={"prompts": ["best crm for a small business"], "brands": ["HubSpot, hubspot.com"]}), then client.dataset(run["defaultDatasetId"]).iterate_items(). Rows with _type are summary, info or error rows, not answers.
Key features
- The whole answer.
answer_textis Gemini's Markdown with headings, lists and tables.answer_plain_textis the same words without the marks. - Sources with position.
sourceslists every page behind the answer. Pages tied to a sentence come first withsection: "cited"andcited_text, the sentence they support. Pages Gemini searched without tying them to a sentence follow withsection: "searched". Each hasposition,url,domain,title, and asnippetandsite_namewhen Gemini gives them. - Brand tracking fields.
brand_mentions,brands_mentioned(first named first) andbrands_cited, from plain text matching on the names and websites you list. - Share of voice. With brands listed, the
SHARE_OF_VOICErecord holds each brand's mention rate, average rank, times named first and citation rate, plus the 25 domains cited most across the run. The same report is in the summary row. - Country and language. Ask from any two letter country.
locationshows what Gemini saw, for example "London, UK". - Repeat a prompt. Gemini words each answer differently.
repeatEachPromptasks a prompt up to 10 times for a steadier share of voice than one snapshot. - Stops cleanly. If the first prompt fails all its tries, nothing else is asked. Six failed prompts in a row end the run. Each failed prompt gets a free error row saying why.
How to use it
Track your brand on the prompts that matter
{"prompts": ["best crm for a small business","best crm for a marketing agency","is hubspot worth the price"],"brands": ["Your Brand, yourbrand.com", "HubSpot, hubspot.com", "Zoho CRM, Zoho, zoho.com"],"repeatEachPrompt": 3}
Run it weekly on a schedule. Chart brands_mentioned and brands_cited by prompt_id and run_date.
Find the pages to get listed on
Filter sources on section equal to cited and count by domain. Those are the pages Gemini leans on for your category, so those are the pages to be reviewed by or quoted in. The Sources view of the dataset shows one source per line.
Compare markets
{"prompts": ["beste laufschuhe bei plattfuss"],"country": "DE","language": "de"}
Run the same prompt from two countries and compare the brands named and the domains cited. prompt_id differs by country, so the two series stay apart.
Input parameters
| Field | Type | Default | Meaning |
|---|---|---|---|
prompts | array of text | required | Questions to ask, one per line, up to 2,000 per run. A repeated prompt is asked once unless you set repeatEachPrompt. |
brands | array of text | none | One brand per line: name, other names, website. Example: Zoho CRM, Zoho, zoho.com. |
country | text | US | Two letter country code the prompt is asked from. |
language | text | en | Interface language. Gemini also follows the language of the prompt, so check answer_language. |
repeatEachPrompt | integer | 1 | How many times to ask each prompt, 1 to 10. Each answer is charged. |
What data do you get?
One row per answer.
| Field | Type | Meaning |
|---|---|---|
engine | text | Always gemini |
prompt | text | Your prompt |
prompt_id | text | Same value on every run for this prompt, country and language |
repeat | integer | Which ask of this prompt in this run |
run_date | text | Day of the run, UTC |
answer_text | text | The answer in Markdown |
answer_plain_text | text | The answer without Markdown marks |
answer_hash | text | Changes when the wording changes |
answer_complete | boolean | Gemini marked the answer as finished |
word_count | integer | Words in the answer |
sources | array | position, url, domain, title, site_name, snippet, section, cited_text |
source_count | integer | Pages in sources |
cited_source_count | integer | Pages tied to a sentence |
source_domains | array | Unique domains in sources |
cited_domains | array | Unique domains tied to a sentence |
brand_mentions | array | Per brand: brand, mentioned, mention_count, first_position, first_sentence, rank, cited, cited_position, cited_url |
brands_mentioned | array | Tracked brands named, first named first |
brands_cited | array | Tracked brands whose site is among the sources |
model | text | Model Gemini reported, for example 3.5 Flash-Lite |
country, language | text | What you set |
answer_language | text | Language Gemini reported |
location | text | Location Gemini saw |
conversation_id, response_id | text | Gemini's ids |
seconds_to_answer | number | Seconds on the successful try |
scraped_at | text | When it was read |
rank is 1 for the tracked brand named first in the answer, 2 for the next. It compares only the brands you list. cited is empty when you gave no website for that brand.
Stable fields for automations
prompt, prompt_id, run_date, answer_text, sources, brand_mentions, brands_mentioned and brands_cited keep their names. Rows with a _type field (summary, info, error) are not answers and are never charged, so filter on _type being empty.
Output examples
A shortened answer row:
{"engine": "gemini","prompt": "best crm for a small business","prompt_id": "3ef820f3dafc52e1","repeat": 1,"run_date": "2026-11-04","answer_text": "Choosing the \"best\" CRM for a small business depends heavily on your budget, technical setup, and primary goals ...","answer_hash": "6111b7e14654bd04","answer_complete": true,"word_count": 504,"sources": [{"position": 1,"url": "https://www.pcmag.com/picks/the-best-crm-software","domain": "pcmag.com","title": "The Best CRM Software We've Tested for 2026 | PCMag","site_name": "PCMag","section": "cited","cited_text": "HubSpot is famous for its intuitive interface and generous free tier."}],"source_count": 3,"cited_source_count": 1,"source_domains": ["pcmag.com", "solutionsreview.com", "g2.com"],"brand_mentions": [{"brand": "HubSpot","mentioned": true,"mention_count": 3,"first_position": 322,"first_sentence": "Best Overall for Ease of Use & Scaling: HubSpot CRM (Smart CRM)","rank": 1,"cited": false}],"brands_mentioned": ["HubSpot", "Zoho CRM"],"brands_cited": [],"model": "3.5 Flash-Lite","country": "US","language": "en","location": "New York, USA","seconds_to_answer": 39,"scraped_at": "2026-11-04T14:11:47.222Z"}
A prompt with no answer, never charged:
{"_type": "info","info_type": "no_answer","prompt": "your prompt","message": "Gemini returned no answer for this prompt after 3 tries. Not charged."}
Pricing
Pay per event: you pay for the answers you receive, plus a flat $0.003 per run, whatever memory you choose. Brand tracking and share of voice cost nothing extra. Same price for every country.
| Event | When it is charged | Free plan | Bronze (Starter) | Silver (Scale) | Gold (Business) |
|---|---|---|---|---|---|
answer-scraped | Once per answer row stored in your dataset | $4.50 per 1,000 | $4.50 per 1,000 | $3.60 per 1,000 | $2.70 per 1,000 |
run-start | Once per run, after the first answer reached your dataset | $0.003 | $0.003 | $0.003 | $0.003 |
The price of one answer is the price per 1,000 divided by 1,000.
Never charged:
- a prompt Gemini did not answer, a refusal, or an answer that is only an error message;
- a prompt that failed after its tries (it gets an error row);
- the summary, info and error rows;
- a run with no answers: the start fee is charged only after the first answer reached your dataset;
- anything past your "Max cost per run". The actor checks the budget before each answer and stops while the total is still below your limit, so the run ends SUCCEEDED with a partial result and a status message saying so.
Run it on a schedule
- Open the actor in Apify Console, set your input and click Save as a new task.
- In the task, open Schedules, click Add schedule and pick how often (every Monday at 08:00, for example).
- Compare runs on
prompt_idto see when a brand enters or leaves an answer, or when a new domain is cited. - Optional: add an integration (Google Sheets, Slack, a webhook) so each run lands where your team works.
FAQ
What does this actor read?
Gemini at gemini.google.com, as a logged out visitor. Each prompt starts a fresh conversation with no history and no personal context.
Which Gemini model answers?
The model Google gives logged out visitors, reported in model. It was 3.5 Flash-Lite on every answer in our tests. Paid Gemini models need an account, and this actor does not use one, so answers can differ from what a signed in Pro user sees.
Why do two runs of one prompt give different answers?
Gemini words each answer afresh, and the brands and sources it picks can change. In US runs location also moved between states from one prompt to the next. For a steady number, set repeatEachPrompt to 3 to 5 and read the share of voice.
Why do some rows have few or no sources?
Gemini decides per prompt whether to search the web. In our 40 prompt test, 37 answers had at least one cited source, the average was 7 sources, and 3 answers had none.
What is the difference between cited and searched?
cited pages are tied to a sentence of the answer, and cited_text shows it. searched pages are ones Gemini looked at through Google Search without tying them to a sentence. Count them apart when you report citation wins.
How long does a run take?
A prompt took 40 seconds at the median and 70 seconds at the 90th percentile in our tests, because Gemini answers at its own pace. Eight run at once, so about 7 answers a minute.
Do I need an account, cookies or a proxy?
No. The actor handles the connection, including countries outside the US, at the same price.
Can it ask follow up questions or upload files?
No. One prompt, one answer, no history.
Which formats can I download?
JSON, CSV, Excel, XML and HTML from the dataset, or through the API. sources and brand_mentions are nested; the Sources view of the dataset lays out one source per line.
Can I use it from Claude, ChatGPT or Cursor (MCP)?
Yes, through Apify's MCP server. Connector URL: https://mcp.apify.com/?tools=themineworks/gemini-search-scraper
- Claude: Settings > Connectors > Add custom connector, paste the URL, sign in with Apify.
- ChatGPT: turn on developer mode, add an MCP connector with the URL, sign in with Apify.
- Cursor or VS Code: add it as an HTTP MCP server with that URL.
- Claude Code:
claude mcp add -t http gemini-search-scraper "https://mcp.apify.com/?tools=themineworks/gemini-search-scraper"
Is it legal to scrape Google Gemini?
The actor reads answers that Gemini gives any visitor who is not logged in, and it collects no personal data. You are responsible for how you use the output, including Google's terms and the laws that apply to you. This is not legal advice.
Integrations
- Google Sheets: send each run's answers to a sheet with Apify's Google Sheets integration.
- Make, Zapier and n8n: start a run and read its dataset with the Apify apps and nodes, for example "tell Slack when our brand drops out of an answer".
- Webhooks: have Apify call your URL when a run succeeds, then fetch the dataset.
- API and SDKs: run it from Python or JavaScript with the Apify client, or with a plain HTTP call (see "Copy to your AI assistant" above).
- MCP clients: Claude, ChatGPT, Cursor and VS Code can run it as a tool through the connector URL in the FAQ.
More from The Mine Works
Marketing, SEO and reviews
- Facebook Ad Library Scraper
- Google Ads Transparency Scraper
- Similarweb Scraper
- Google News Scraper
- Trustpilot Reviews Scraper
- Google Trends Scraper
- Semrush Scraper
- App Reviews Scraper
- Trustpilot Business Search
- TripAdvisor Reviews Scraper
- Username Checker
- Backlink Building Agent
Social media and video
Leads and business directories
Real estate
Science, health and government data
Jobs and hiring
Company and business data
E-commerce and marketplaces
Food and local services
Developer and AI tools
More tools
Support
Found a prompt that comes back empty, or a field that looks wrong? Open an issue in the Issues tab with the input and the run ID, and we will look at it. For a field you need or a new source, email dmineworks@gmail.com.
Ask Google Gemini your prompts and get every answer back as one row, with the answer text, cited sources with URL and position, and brand mentions, rank and share of voice, no login.

