# Google AI Mode Scraper

Google AI Mode Scraper returns the answer Google's AI Mode tab (udm=50) shows for each query, as JSON: answer_text, structured text_blocks, resolved source URLs in references, per-block citations and inline citation chips. It costs $0.003 per Ok answer, $0.0018 on Business plans, and failed, empty or blocked rows are free. Run it from the Apify Console, the Apify API or an MCP client.

## Key facts

- Price: from $0.003 per ai mode search completed, $3.00 per 1,000, down to $0.0018 on higher Apify plans
- Pricing model: Pay per event
- Output: JSON, CSV, Excel, XML, RSS via Apify datasets
- Access: Apify Console, REST API, Python and JS clients, MCP tool google_ai_mode
- Version: Updated Nov 2026, build 0.2.22
- Usage: 65 users, 1,732 runs
- Actor: subimpact/google-ai-mode-scraper

## Who it is for

- SEO and GEO teams checking whether Google AI Mode cites their pages
- Brand marketers comparing how AI Mode describes them against competitors
- Content strategists mapping which domains AI Mode trusts in a niche
- Analysts building answer-engine visibility dashboards

## Use cases

### Track citations to your domain

Put your brand questions in queries, set gl and hl to the market, then read references and inline_citations for your domain. text_blocks attaches a citations array to each paragraph, so you can log which paragraph cites which source. Running the same set weekly shows whether you gain or lose AI Mode mentions, and search_metadata.citations_missing flags answers with no sources at all.

### Compare answers across markets

Country, language and city are inputs, so one run covers one market. Run the same queries with gl set to ae and hl to ar, then again with gl my and hl ms, and diff answer_text and references. The facts record local sources surfacing per market, for example moneysmart.sg in Singapore and ringgitplus.com in Malaysia, which is the evidence you need for localised content briefs.

### Keep only sourced answers

Google sometimes writes a full answer with no sources. Those rows come back status Ok with search_metadata.ungrounded set to true and a warning. Filter on ungrounded and citations_missing to build a clean corpus of grounded answers, and drop status Partial and Empty rows, which are not charged, before you send anything to an LLM or a dashboard.

### Monitor how answers drift

Set skipCache to true and schedule the Actor daily on a fixed query set. Cached rows are marked with cached and search_metadata.cached_at, so you can separate a real change in answer_text from a repeat served from the one hour cache. Store answer_text and references per run to chart which competitor domains enter and leave the answer over time.

### Map the sources AI Mode leans on

Run a batch of category queries and group the references array by source and link. The result is a ranked list of the forums, video channels and vendor blogs Google quotes for that topic, with the snippet it kept. That list tells a content team where a placement is worth chasing, and answer_links shows which brand names Google hyperlinked inside the prose.

## Price events

- AI Mode search completed (`search-completed`): $0.003, lowest tier $0.0018. One full Google AI Mode (udm=50) answer with citations delivered (status Ok). Partial/Empty/Blocked/Failed rows are free.

Prices from the Apify Store. Apify shows the final price before each run.

## How it compares with other Apify actors

Price per 1,000 answers at Apify Store list prices, checked 20 Nov 2026.

| Actor | Price per 1,000 answers (Free plan) | Higher plans (Gold) | Start fee per run | Platform usage | Users | Rating | Last build |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Google AI Mode Scraper (this API) | $3.00 | $1.80 | None | Included | 65 | No reviews yet | 18 Nov 2026 |
| [Google Search Results Scraper](https://apify.com/apify/google-search-scraper) | $204.50 | $3.80 | $0.00005 | Included | 239,276 | 4.7 (219 reviews) | 18 Nov 2026 |
| [Google AI Mode Scraper](https://apify.com/apify/google-ai-mode-scraper) | $200.00 | $2.00 | $0.00005 | Included | 464 | 5.0 (7 reviews) | 18 Nov 2026 |
| [Google AI Mode Scraper — Generative Answers](https://apify.com/scrape.badger/google-ai-mode-scraper) | $5.00 | $5.00 | $0.00005 | Included | 166 | No reviews yet | 10 Oct 2026 |
| [Google AI Mode Scraper](https://apify.com/lexis-solutions/google-ai-scraper) | $49.00 | $49.00 | $0.01 | Included | 106 | No reviews yet | 20 Feb 2026 |

### Where Google Search Results Scraper is stronger

- Far larger user base and a long review history on the Apify Store
- Also returns AI Overviews, ads and answers from other AI engines in the same run

### Where this API is stronger

- Output carries dedicated status and thin answer fields, so blocked or uncited answers stay visible
- No separate Actor start charge on top of the per answer price
- Answers come from the AI Mode tab itself rather than as an add on to a general SERP run

### Where Google AI Mode Scraper is stronger

- Built by Apify itself, so it sits naturally beside their other Google actors
- Only two inputs, a query list and a country, which keeps a first run quick

### Where this API is stronger

- Language can be set per query next to country, rather than country alone
- Output includes status, thin answer and warning fields for blocked or uncited answers
- Failed and empty queries are not charged

### Where Google AI Mode Scraper — Generative Answers is stronger

- Compact output of text blocks and references, quick to map into an existing schema
- Three field input covering queries, country and language, minimal to configure

### Where this API is stronger

- Output adds status, thin answer and warning fields so weak answers can be filtered
- Each paragraph citation is delivered with its resolved source URL
- No Actor start charge in addition to the per answer price

### Where Google AI Mode Scraper is stronger

- Output includes sections, sources and highlighted terms, useful for research and content reuse
- Takes a plain list of questions without country or language setup

### Where this API is stronger

- Requests the AI Mode tab directly and sets country and language per query
- Output separates status, thin answer and error so partial runs are easy to spot
- Uncited answers are flagged rather than returned as if complete

Buyers who want a whole SERP, including ads and several AI engines, are well served by Apify's general Google Search actor. If the job is only AI Mode answers, the smaller dedicated actors are simpler to set up, while this actor adds per query country and language plus status and citation fields that make answer quality easy to audit.

Prices are Apify Store list prices, checked 20 Nov 2026. Platform usage 'Paid separately' means Apify bills compute on top of the event price. Check each listing before you buy.

## Input

- `queries` (array, required): Queries to get Google AI Mode answers for. Each query is run on the real AI Mode tab (udm=50) in a real Android Chrome browser.
- `gl` (string, optional): Two-letter country code for the market whose AI Mode answers you want, e.g. ae (UAE), my (Malaysia), sg (Singapore), id (Indonesia), au (Australia), us. Works on both backends.
- `hl` (string, optional): Language code for Google's interface and answer, e.g. en, ar, ms, id, zh-CN. Write your queries in the same language. In markets where English is not the main search language (e.g. id, th) use the local language: English queries there of...
- `location` (string, optional): Google canonical location name, e.g. 'Dubai,Dubai,United Arab Emirates' or 'Kuala Lumpur,Federal Territory of Kuala Lumpur,Malaysia'. Leave empty to use the main city of the chosen country (Dubai, Kuala Lumpur, Singapore, Jakarta, Sydney...
- `skipCache` (boolean, optional): By default, a question that was answered for the same country, language and location in the last hour may be served from our cache (marked cached: true, with cached_at). Turn this on to always fetch a fresh answer from Google, for exampl...
- `failOnNoCapacity` (boolean, optional): Off by default: when our browser pool briefly runs out of eligible identities, the run waits up to 3 minutes when capacity is about to return, otherwise every query comes back as a free Blocked row with a retry hint (search_metadata.capa...
- `backend` (string, optional): android (default) = real AI Mode tab (udm=50) rendered in Android Chrome: full answer, structured blocks, resolved source URLs. http = legacy browserless SERP fetch (only returns AI answers Google embeds in the classic results page; uses...
- `concurrency` (integer, optional): Parallel queries (1-30). On the android backend this is capped automatically to the size of the browser pool.
- `udm50` (boolean, optional): [legacy http backend only] Appends udm=50. NOTE: the forced shell currently returns no answer browserlessly (Google requires a real browser for the folwr stream) — defaults OFF.
- `retryDeferred` (boolean, optional): [legacy http backend only] When the answer is deferred, re-fetch through the folsrch async endpoint. ON by default (recovery rate ~90%).
- `retryDeferredMaxFetches` (integer, optional): [legacy http backend only] Upper bound on recovery fetches per query (1-5).
- `captureRawHtml` (boolean, optional): [legacy http backend only] Save raw SERP HTML to the key-value store for every query (for corpus/debugging). Adds storage cost.

### Example input

```json
{
  "queries": [
    "how does compound interest work"
  ]
}
```

## Output fields

- `query` (string): The search query.
- `gl` (string): Country code (two letters) the query was run against.
- `hl` (string): Language code (two or three letters) the query was run against.
- `cached` (boolean/null): True when the answer was served from the internal same-market cache (skippable with skipCache). null on rows where the question never reached the cache path.
- `thin_answer` (boolean/null): True when the answer text was under 60 characters (stub below the sellable minimum). null when the answer was not thin.
- `status` (string): Ok (charged; includes ungrounded answers Google served without sources, flagged search_metadata.ungrounded) | Partial (left the AI Mode tab / parser suspect — NOT charged) | Empty (no AI Mode answer served) | Failed (error) | Blocked (CA...
- `ai_mode` (boolean/null): True when the answer was classified as the AI Mode surface.
- `ai_overview` (object/null): AI Mode answer: answer_text, text_blocks (paragraph/heading/list/table, each may carry citations), references (title/link/snippet/source), inline_citations, answer_links (array of hyperlinked phrases inside the answer prose, or null when...
- `organic_results` (array/null): Organic SERP results (optional).
- `search_parameters` (object): Echo of query params: hl, gl, udm, location.
- `search_metadata` (object): backend, total_time_ms, first_answer_s, final_url_has_udm50, surface, citations_missing + ungrounded (Google served the answer without sources; still Ok and charged), parse_suspect (http backend), capacity_blocked (true on a free Blocked...
- `warning` (string/null): Explains a Partial row, or flags an Ok row Google served without sources (ungrounded). null on clean rows.
- `error` (string): Error message when status != Ok. For a capacity-Blocked row it explains that our browser capacity is temporarily busy, that nothing was charged and after how many minutes to retry.

## Sample output

Trimmed from one of our own test runs. Personal data is redacted.

```json
[
  {
    "query": "quelle assurance pour un vélo électrique",
    "gl": "fr",
    "hl": "fr",
    "search_parameters": {
      "hl": "fr",
      "gl": "fr",
      "udm": 50
    },
    "ai_overview": {
      "answer_text": "Réponse du Mode IA pour quelle assurance pour un vélo électrique\nAssurer un vélo à assistance électrique (VAE) dépend avant tout de sa catégorie et de ce que vous souhaitez protéger (le vélo lui-même, vous-même ou les autres).\ngaya.bike\n...",
      "text_blocks": [
        {
          "type": "heading",
          "snippet": "Réponse du Mode IA pour quelle assurance pour un vélo électrique"
        },
        {
          "type": "paragraph",
          "snippet": "Assurer un vélo à assistance électrique (VAE) dépend avant tout de sa catégorie et de ce que vous souhaitez protéger (le vélo lui-même, vous-même ou les autres).",
          "citations": [
            "{...}"
          ]
        },
        {
          "type": "heading",
          "snippet": "1. L'obligation légale"
        },
        "..."
      ],
      "references": [
        {
          "title": "Assurance vélo électrique obligatoire en 2026 - Gaya",
          "link": "https://gaya.bike/journal/assurance-velo-electrique-obligatoire",
          "snippet": "Assurance VAE (19 août 2026): Garanties utiles: Responsabilité civile (via multirisque habitation), vol/casse (tous risques), domm...",
          "source": "gaya.bike",
          "index": 0
        },
        {
          "title": "Faut-il assurer un vélo électrique ? - Conseils CYCLES BLAIN",
          "link": "https://www.cycles-blain.com/fr/conseils-velo/faut-il-assurer-un-velo-electrique/",
          "snippet": "250W et 25km/ h. VAE est considéré comme un vélo classique sans assurance obligatoire une puissance égale ou inférieure à 250w ce",
          "source": "Cycles Blain",
          "index": 1
        },
        {
          "title": "Meilleure Assurance Vélo 2026 : Comparatif & Avis",
          "link": "https://www.velo101.com/meilleures-assurances-velo-2026-comparatif-et-guide-complet/",
          "snippet": "Assurances 2026 comparées (VAE 2k€ ref): * Laka: ~185€/an, 0% franchise, * Hepster: ~125€/an, 10% franchise, le moins cher. * Shar...",
          "source": "Velo 101",
          "index": 2
        },
        "..."
      ],
      "inline_citations": [
        {
          "source": "gaya.bike",
          "link": "https://gaya.bike/journal/assurance-velo-electrique-obligatoire"
        },
        {
          "source": "Cycles Blain",
          "link": "https://www.cycles-blain.com/fr/conseils-velo/faut-il-assurer-un-velo-electrique/"
        },
        {
          "source": "Velo 101",
          "link": "https://www.velo101.com/meilleures-assurances-velo-2026-comparatif-et-guide-complet/"
        },
        "..."
      ],
      "answer_links": [
        {
          "text": "Assurance Multirisque Habitation",
          "link": "https://www.service-public.gouv.fr/particuliers/vosdroits/F2697",
          "source_index": 3,
          "block_index": 7
        }
      ],
      "resolved_by": "android"
    },
    "ai_mode": true,
    "status": "Ok",
    "search_metadata": {
      "backend": "android-chromium",
      "surface": "ai_mode",
      "total_time_ms": 11809,
      "final_url_has_udm50": true,
      "first_answer_s": 3.5,
      "render_s": 10.7
    },
    "cached": false
  },
  {
    "query": "comment détartrer une bouilloire",
    "gl": "fr",
    "hl": "fr",
    "search_parameters": {
      "hl": "fr",
      "gl": "fr",
      "udm": 50
    },
    "ai_overview": {
      "answer_text": "Pour détartrer efficacement une bouilloire, utilisez du vinaigre blanc, de l'acide citrique ou du citron, quel que soit le matériau de l'appareil (inox, verre ou plastique). Il suffit de faire chauffer le produit dilué dans de l'eau, de ...",
      "text_blocks": [
        {
          "type": "paragraph",
          "snippet": "Pour détartrer efficacement une bouilloire, utilisez du vinaigre blanc, de l'acide citrique ou du citron, quel que soit le matériau de l'appareil (inox, verre ou plastique). Il suffit de faire chauffer le produit dilué dans de l'eau, de ...",
          "citations": [
            "{...}"
          ]
        },
        {
          "type": "heading",
          "snippet": "1. La méthode au vinaigre blanc (la plus économique)"
        },
        {
          "type": "paragraph",
          "snippet": "Le vinaigre blanc dissout rapidement le calcaire accumulé sur la résistance et les parois.",
          "citations": [
            "{...}"
          ]
        },
        "..."
      ],
      "references": [
        {
          "title": "Comment nettoyer et détartrer sa bouilloire électrique ?",
          "link": "https://www.darty.com/darty-et-vous/cuisine/cuisine-pratique/entretien-des-appareils/comment-detartrer-sa-bouilloire-electrique",
          "snippet": "Pour cela, il suffit de faire bouillir de l'eau avec un produit spécifique contre le tartre, puis de rincer deux à trois fois l'ap...",
          "source": "Darty",
          "index": 0
        },
        {
          "title": "Détartrer sa bouilloire : 5 méthodes naturelles et faciles",
          "link": "https://blog.lafourche.fr/comment-detartrer-sa-bouilloire-naturellement",
          "snippet": "Calcaire: caused by hard water … 5 natural descaling methods for all materials (inox, plastic, copper): - 1. White vinegar: Freque...",
          "source": "La Fourche",
          "index": 1
        },
        {
          "title": "Comment détartrer une bouilloire ? - Les 3 méthodes faciles !",
          "link": "https://www.wecasa.fr/mag/detartrer-bouilloire/",
          "snippet": "Méthode 1 (Vinaigre blanc/eau 50/50) : bouillir, agir 30-60 min, rincer. Convient à. Méthode 2 (1-2 c.à.s. acide citrique/eau) : b...",
          "source": "www.wecasa.fr",
          "index": 2
        },
        "..."
      ],
      "inline_citations": [
        {
          "source": "Darty",
          "link": "https://www.darty.com/darty-et-vous/cuisine/cuisine-pratique/entretien-des-appareils/comment-detartrer-sa-bouilloire-electrique"
        },
        {
          "source": "YouTube",
          "link": "https://www.youtube.com/watch?v=P0uQ131RaTk"
        },
        {
          "source": "Haier Europe",
          "link": "https://www.haier-europe.com/fr_FR/blogs/news/comment-detartrer-votre-bouilloire-electrique"
        },
        "..."
      ],
      "answer_links": [
        {
          "text": "Darty",
          "link": "https://www.darty.com/darty-et-vous/cuisine/cuisine-pratique/entretien-des-appareils/comment-detartrer-sa-bouilloire-electrique",
          "source_index": 0,
          "block_index": 14
        },
        {
          "text": "Wecasa",
          "link": "https://www.wecasa.fr/mag/detartrer-bouilloire/",
          "source_index": 2,
          "block_index": 14
        }
      ],
      "resolved_by": "android"
    },
    "ai_mode": true,
    "status": "Ok",
    "search_metadata": {
      "backend": "android-chromium",
      "surface": "ai_mode",
      "total_time_ms": 12645,
      "final_url_has_udm50": true,
      "first_answer_s": 2.5,
      "render_s": 7.7
    },
    "cached": false
  }
]
```

## How to run it

1. Open the Actor input and paste your questions into queries, for example ["best crm for startups", "notion vs obsidian"].
2. Set gl to the market you want, such as my or ae, and hl to the language of your queries, such as ms or ar.
3. Leave location empty to use the main city of that country, or pass a Google canonical name such as Dubai,Dubai,United Arab Emirates.
4. Set skipCache to true if every run must fetch a fresh answer rather than one cached within the hour.
5. Start the run in the Apify Console; rows stream into the dataset as each answer completes, at roughly 10 seconds per answer.
6. Read ai_overview.answer_text, text_blocks and references per row, then export the dataset as JSON, CSV or Excel.

### cURL

```bash
curl -X POST "https://api.apify.com/v2/acts/subimpact~google-ai-mode-scraper/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"queries":["how does compound interest work"]}'
```

### Python

```python
from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("subimpact/google-ai-mode-scraper").call(run_input={
      "queries": [
        "how does compound interest work"
      ]
    })

for item in client.dataset(run["default_dataset_id"]).iterate_items():
    print(item)
```

### JavaScript

```javascript
import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_APIFY_TOKEN>' });
const run = await client.actor('subimpact/google-ai-mode-scraper').call({
  "queries": [
    "how does compound interest work"
  ]
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();

console.log(items);
```

### MCP

```bash
claude mcp add subimpact-data -e APIFY_TOKEN=<YOUR_APIFY_TOKEN> -- npx -y @subimpact/data-mcp

# Then ask your agent to use the google_ai_mode tool.
```

### n8n

Use the Apify node (Run an Actor).
Actor ID: subimpact/google-ai-mode-scraper
Input JSON: paste the example input below.

## Limits

- One market per run: gl, hl and location are fixed inputs, so multi-country tracking needs one run per country.
- In Indonesia and Thailand, English queries often return generic or US answers; the facts advise writing those queries in the local language.
- Some answers are served with no sources at all; they are still charged and flagged ungrounded rather than dropped.
- Cached answers up to an hour old are returned unless skipCache is on.

## FAQ

### How much does one AI Mode answer cost?

$0.003 per Ok answer on the Free and Starter plans, falling to $0.0018 on Business and above, so $1.80 per 1,000 answers at the top tier. Platform usage is included. Rows with status Empty, Partial, Blocked or Failed are not charged.

### Do I need a Google account, cookies or a proxy?

No. Each query is rendered on the real AI Mode tab in an Android Chrome browser, and there is nothing to log in to. You pass a country and language instead of a session, and the Actor reports its own browser capacity when Google throttles it.

### Which countries and languages can I target?

Any two-letter country in gl and any language code in hl. The facts record tested answers for the UAE, Singapore, Australia and Malaysia, plus Arabic, Malay, Indonesian, Thai, Chinese, Japanese, German, French, Spanish, Portuguese, Korean, Italian, Vietnamese, Hindi and Russian.

### How many queries fit in one run?

Batches of up to 1,000 queries per run, with concurrency between 1 and 30 on the android backend, where it is capped to the browser pool size. Each answer takes around 10 seconds, and rows are billed as they are delivered, so a timed-out run keeps what it already returned.

### What does status Partial mean?

Partial means Google moved the query off the AI Mode tab or the parser was not sure it read the full answer. It is not charged. Empty means no answer was served, Blocked means a Google interstitial or exhausted browser capacity, and Failed means an error. None of those three are charged either.

### How fresh is the data?

A repeat of the same question, country, language and location within about an hour may come from the internal cache, marked cached with cached_at and cache_source. Set skipCache to true to bypass it, which is what you want when tracking how an answer changes between runs.

### Can I call it from Python, n8n or an AI agent?

Yes. The Actor runs through the Apify API with a token, so any HTTP client, n8n node or agent tool can start a run and read the dataset, or you can connect it as an MCP tool. The input is the same JSON as the Console form.

## Links

- Page: https://subimpact.net/apis/google-ai-mode-scraper/
- Category: https://subimpact.net/apis/category/search-seo/
- Run on Apify: https://apify.com/subimpact/google-ai-mode-scraper
- All data APIs: https://subimpact.net/apis/
- Catalogue JSON: https://subimpact.net/apis/catalogue.json
