Google AI Mode Scraper — Answers & Citations

What is Google AI Mode Scraper?

Send one or more natural-language prompts to Google AI Mode and receive a structured dataset with the answer text, text blocks, cited sources, related questions, image/product modules, provenance, and optional multi-turn follow-ups.

The Actor retrieves answers through a managed search data route rather than a browser. Google AI Mode is a conversational search surface that can fan out a question into related searches and return supporting web links; the managed route preserves those answer and citation fields. Direct browser access to Google is challenge-prone in automated environments, so the managed route is the reliability boundary for this build.

Best fit

• GEO/AEO teams auditing which sources Google AI Mode cites for a prompt
• Researchers exporting cited answers and related questions for analysis
• AI agents that need one narrow prompt-in, structured answer-out tool
• Conversation tests that compare an initial prompt with ordered follow-ups

Workflow story

An SEO analyst starts with a fixed prompt set, chooses a country and language, and runs the Actor. Each answer row keeps the prompt, turn, text blocks, citations, related questions, provider search ID, and fetch time together. The analyst then exports the citations view to review source coverage, or sends the answer rows to a GEO report, alert, or research notebook. With followUpQueries, the same workflow continues from the initial response through each ordered question while conversationId keeps separate prompts isolated.

Input

{
  "queries": [
    "Compare the best project management tools for a five-person team."
  ],
  "followUpQueries": ["Which option is best for a bootstrapped team?"],
  "countryCode": "us",
  "languageCode": "en",
  "device": "desktop",
  "continuable": true
}

queries is required and accepts up to 25 prompts. followUpQueries is optional and accepts up to five ordered prompts; the same sequence is applied to every initial prompt. The Actor uses the continuation token returned by Google AI Mode for each next turn. location, imageUrl, noCache, and maxRetries are optional. For image prompts, provide a public HTTPS image URL.

Authentication is handled by the Actor's owner-managed secret, so you do not need an account or key. An optional bring-your-own-key input (advanced) overrides the managed secret for testing; never place a key in a public task, README example, source file, or dataset.

Output

One record is written for each successful prompt turn:

Field	Meaning
query	Prompt sent to Google AI Mode
conversationId / turn	Run-scoped conversation and one-based turn number
responseText	Normalized response text, preferring reconstructed Markdown
textBlocks	Ordered answer blocks with citation reference indexes
citations	Ordered title, URL, domain, source, snippet, and reference index records
relatedQuestions	Suggested Google AI Mode follow-up questions
subsequentRequestToken	Opaque continuation state for the next turn

{
  "query": "Compare the best project management tools for a five-person team.",
  "conversationId": "conversation-1",
  "turn": 1,
  "isFollowUp": false,
  "responseText": "...",
  "reconstructedMarkdown": "... [0] ...",
  "textBlocks": [
    { "type": "paragraph", "text": "...", "referenceIndexes": [0, 1] }
  ],
  "citations": [
    {
      "position": 1,
      "title": "Example guide",
      "url": "https://example.com/guide",
      "domain": "example.com",
      "source": "Example",
      "snippet": "...",
      "referenceIndex": 0
    }
  ],
  "relatedQuestions": [{ "question": "Which tool is best for a small team?" }],
  "referenceCount": 1,
  "subsequentRequestToken": "opaque-provider-token",
  "fetchedAt": "2026-08-09T00:00:00.000Z",
  "warnings": []
}

The opaque continuation token is retained for provenance and debugging. Treat it as transient provider state, not as a credential. Run-level SUMMARY, OUTPUT, and RUN_SUMMARY records include the terminal outcome and charged event counts.

Pricing and usage

• Pay per event: apify-actor-start is charged once per run, and ai-mode-query is charged once per persisted prompt turn (initial prompts and follow-ups each count as one turn). Valid empty answers are saved and charged as a turn.
• Example: a run with 4 initial prompts and 1 follow-up per prompt writes up to 8 turns, so up to 8 ai-mode-query events plus one apify-actor-start event.
• Event prices are shown on the Pricing tab, plus Apify platform usage.

The prompt event is charged only after a provider response has been normalized and immediately before it is saved. A provider issue is reported in the run summary and receives no successful-result charge.

Free Apify plan

Runs started from a free Apify plan return up to 25 results per run, with every input mode and output field available. Paid Apify plans have no per-run limit, so your usual result settings apply.

API example

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("khadinakbar/google-ai-mode-scraper").call(run_input={
    "queries": ["How does Google AI Mode differ from AI Overviews?"],
    "countryCode": "us",
    "languageCode": "en"
})

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["query"], item["referenceCount"])

MCP and agent prompt card

Connect the Actor through Apify MCP and give an agent this prompt:

Use Google AI Mode Scraper for a current Google AI Mode answer to a specific prompt. Return the answer text, citation URLs and domains, related questions, fetchedAt, and the run outcome. Use followUpQueries when the next question depends on the previous turn. Treat generated text as an answer to inspect and citations as provenance links to read back.

Agents can read the dataset for row-level answers and RUN_SUMMARY for completion, partial-result, valid-empty, configuration, or upstream outcomes. Each row includes the routing options listed in the input schema for provenance and freshness.

Practical guidance

• Use informational, comparison, and research prompts when you want rich answer text and citations.
• Keep localization fixed when comparing runs; changing countryCode, languageCode, location, or device can change the response.
• Use noCache: true only when freshness matters enough to justify a new provider request.
• Keep follow-ups concise and dependent on the initial question so the conversation boundary remains clear.
• A valid empty answer is preserved as a successful row with responseText: null and a warning; an upstream/configuration outcome is reported separately in RUN_SUMMARY.

Builder's note

I built the normalization layer around Google AI Mode's structured response shape: text blocks keep their reference indexes, citations derive their domain from the URL, and continuation tokens stay attached to the conversation row. This keeps the dataset useful when the answer includes prose, related questions, shopping modules, or a follow-up turn in the same run.

Best results and responsible use

Google AI Mode responses, citations, availability, and localization can change. A managed search API notes that some languages may return an incomplete or unexpected response shape. Use public, authorized search workflows and comply with Google's terms, a managed search API's terms, Apify policies, and applicable law.

Support

Use the Apify Actor Issues tab for reproducible problems. Include the prompt shape, country/language/device, run ID, and warning fields; never include API keys or continuation tokens.

Continue the workflow

• Then use Google AI Overviews Scraper — AIO Citations & Brand Visibility to extend Google AI Mode Scraper — Answers & Citations research with a complementary content contract.
• Then use Google SERP All-in-One Scraper — AIO, PAA & Knowledge Panel to extend Google AI Mode Scraper — Answers & Citations with a neighboring search visibility work source when the brief calls for Google Serp data.