Google AI Mode Scraper: the conversational answer and its citations

This Google AI Mode scraper returns the answer behind Google's AI tab as structured JSON, for AI Mode tracking, SEO and research. Ask a question the way you would ask a person and you get the full answer text, the same answer broken into headings, paragraphs and lists with citations attached per block, and source cards when Google includes them in the answer response.

AI Mode is a separate surface from the AI Overview above the search results. It answers at length, follows the question rather than the keywords, and cites a different, usually wider set of pages. There is no official API for it. This actor returns one dataset row per question.

What the Google AI Mode scraper returns

• The complete answer as one plain string (answer_text), which is what you want in a spreadsheet cell or as context in a prompt.
• The answer with its structure intact (answer_blocks): headings stay headings, lists stay lists, and each block carries its own citations. That is what makes it possible to attribute one claim to one source.
• Available cited source cards (sources) with site name, domain, page title, the snippet Google used, the original site URL, favicon and thumbnail. Where available, page_url and ordered citations identify the cited page.
• The distinct domains (source_domains) plus a count, so "who does Google cite for this question" is one field, not a parse.
• Links inside the answer (links) with Google's redirect already resolved to the real destination.
• The thread identifier (thread_id) Google assigned to the answer.
• An honest answered flag (answered). Questions that came back without an answer are returned marked, and are not billed.

Why scrape Google AI Mode

AI Mode changes what a search result is. Instead of ten links it returns one composed answer built from several sources at once, and the sources it picks are not the ten that would have ranked. A page can sit outside the first page of ordinary results and still be quoted in the answer, and a page that ranks first can be absent from it entirely. Neither fact shows up anywhere in a rank report.

That makes citation share the thing worth measuring. If Google's answer to a question in your category quotes three competitors and not you, that is a concrete gap with a concrete fix, and it is invisible to every tool that measures position. Running a fixed question set on a schedule turns it into a trend you can act on.

The answers are also useful as raw material. They are Google's own summary of what a good answer to a question looks like, assembled from the sources it trusts most. answer_blocks shows the structure it chose: what it led with, which subheadings it used, what it decided belonged in a list. For anyone writing the page that should have been cited, that is a better brief than a keyword tool.

Input

Field	Type	Required	Default	Description
queries	array of strings	yes	—	One question per line.
country	string	no	us	Two-letter country code. Sets the market the answer is written for.
language	string	no	en	Two-letter language code for the answer.
retries	integer	no	3	Extra rounds to spend when an answer does not render. Range 0 to 5.

Phrase inputs as questions. AI Mode is built for conversational input, and a bare keyword string produces a noticeably thinner answer than the same subject asked as a question.

How to use the Google AI Mode scraper

1. Open the actor in Apify Console and paste your questions into queries, one per line.
2. Set country and language for the market you want the answer for, or keep the defaults us and en.
3. Start the run. Each question becomes one row in the dataset, which you can export as JSON, CSV or Excel, or read through the Apify API.

Google AI Mode API: call it from your own code

Google offers no public API for AI Mode. Through the Apify API this actor works as one: start a run with the same input and read the dataset when it finishes.

curl -X POST "https://api.apify.com/v2/acts/s-r~google-ai-mode-scraper/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"queries": ["is the lego 10313 worth buying"], "country": "us", "language": "en"}'

The response is the list of rows described under Output. The same call works from the Apify Python and JavaScript clients, from schedules and from integrations such as Make, Zapier and n8n.

Output

One row per question.

The following is a literal field selection from an actual response. The full row also includes answer_blocks, sources, links and ordered citations.

{
  "query": "is the lego 10313 worth buying",
  "country": "us",
  "language": "en",
  "answered": true,
  "answer_text": "The LEGO Icons Wildflower Bouquet (10313) is worth buying if you want a relaxing build, vibrant home decor that never dies, or a unique gift. Retailing for around $59.99 with 939 pieces, it offers strong value for money compared to other display sets.\n\nTo help you decide, here is a breakdown of the set's key pros and cons:\n\nKey Pros\n\n- Stunning Visuals: Features 8 different wildflower species (including lavender, lupins, and poppies) across 16 adjustable stems that look surprisingly lifelike from a distance.\n- Relaxing Build Experience: Rated for adults (18+), the assembly process is engaging, creative, and serves as a great mindful activity.\n- Customizable Display: You can alter the height of the individual stems to fit neatly into any personal vase.\n- Expandable: It can be combined with the original LEGO Flower Bouquet (10280) to build an even larger, fuller arrangement.\n\nKey Cons\n\n- No Vase Included: You will need to supply your own vase to display the finished bouquet.\n- Dust Magnet: Because of the complex shapes and open layout, cleaning dust off the plastic petals requires gentle care (like a compressed air duster).\n- Static Display: Unlike traditional LEGO sets with play features, this is strictly a static model meant to sit on a shelf or table.\n\nYou can view more details and purchase the set directly from the Official LEGO Shop US.\n\nIf you'd like, let me know:\n\n- Do you plan to display it on its own or combine it with another set?\n- Are you buying this as a gift or for yourself?\n\nI can help you figure out the best way to arrange or display it!",
  "answer_block_count": 10,
  "source_count": 1,
  "source_domains": [
    "lego.com"
  ],
  "products": [
    {
      "title": "LEGO Icons Wildflower Bouquet (10313)",
      "url": "https://www.google.com/search?prds=pvt:hg,pvo:29,mid:576462788227172071,imageDocid:15385124669211416508,gpcid:14746354957568825417,headlineOfferDocid:16156216421252268267,catalogid:9131157491774088421,productDocid:14228807214483735920,rds:PC_14746354957568825417%7CPROD_PC_14746354957568825417,oapvfc:EswCCowCQUppVDR0S2UweEFiZEh1Mk9Sa2NsbzlwSWVseWkxclZOVkYxMXBxazdXbnd4YkhBd3NTYWc1UTVOc0ZrVUNNdjlIaDJ4WEROd2x6dmVjM1VDczdUOGFYTjZETmVmMUxMOVJJQ0NJTzhhUzdGSFZ0SDZiX193cW9wR1BOXzRqYTVHSUJIS09ubjFlNjNEZHg4OFZDb1lDSzBXaUpzQ3VVVVNqWmR0WXVpX2I2eWVzOGVESFZoQ05SdXpnNE12UTJLQlNHTl84VGlUYU5fdlVEWWtWaU93MWREbnFuUFZxVGNHd3I1eGRpbmhBTEtoQzBucjUwdlNHRV9zRHJHaXd2ckgwejNSX3IzYlEtQhIXaWpUQmFydVFJZU9DOXU4UHNfVHNrQWcaIkFEc3I5ZlQ1YUk0V1NHTG1XQWJWM1pfajNuMGNUdmlaUEE&ibp=oshop&q=product",
      "image": "",
      "catalog_id": "9131157491774088421",
      "product_id": "14228807214483735920",
      "merchant_id": "576462788227172071",
      "cluster_id": "14746354957568825417"
    },
    {
      "title": "LEGO Wildflower Bouquet",
      "url": "https://www.google.com/search?q=product&prds=pvt:hg,productid:14997873278647843262,catalogid:1083505197665243410,gpcid:14746354957568825417,mid:576462788227172071&ibp=oshop",
      "image": "https://encrypted-tbn2.gstatic.com/images?q=tbn:ANd9GcRo5bkWn4PessSWHCnw28LFRTQupTUC5aD1Px6tXE3AhYay5SAi1yXqjgUFCx_VEKR9IaCNVv_m6oGcYZI",
      "catalog_id": "1083505197665243410",
      "product_id": "",
      "merchant_id": "576462788227172071",
      "cluster_id": "14746354957568825417"
    }
  ],
  "thread_id": "jTTBaoeHFfbi7_UPtdyPsAc",
  "attempts": 1,
  "error": null,
  "fetched_at": "2026-10-03T16:59:58Z",
  "duration_seconds": 7.45,
  "response_bytes": 334384
}

Use cases

Google AI Mode tracking

Measuring citation share in AI answers. Fix a list of questions your buyers actually ask, run it weekly, and count how often each domain appears in source_domains. That count is the metric: it tells you whether Google considers you an authority on the question, which is now a separate outcome from ranking for it.

Research and content

Finding the competitors that rank reports miss. AI Mode cites pages that do not appear on page one. Run your category's questions and the domain frequency list will surface sites you were not tracking, because the surface that introduced them to your buyers is not the one you were watching.

Briefing content from Google's own structure. Before writing the page that should be cited, pull the current answer and read answer_blocks. The subheadings, the ordering and the list items are Google's own decomposition of the question, taken from the system doing the deciding.

Feeding a retrieval pipeline. Each row is an answer with its sources already attached and deduplicated. That is a usable summarisation layer for a question set: the prose for context, sources for provenance, so the answer you pass downstream carries its own citations.

How it compares

	This actor	Typical alternative
Answer structure	Headings, paragraphs and lists preserved, citations per block	Answer as one flat string
Sources	Full card: name, domain, title, snippet, URL, favicon, thumbnail	Domain or URL list
Unanswered questions	Returned and marked, not billed	Often a failed run
Browser required	No	Several alternatives drive a headless browser
Start fee	None	$0.01 per run on some listings

Pricing

$0.01 per AI Mode answer returned. There is no actor start fee. Questions that come back unanswered are returned in the dataset and are not charged. All pricing is pay-per-event, so you only pay for results you receive. There are no per-compute-unit charges.

Limits and gotchas

• AI Mode answers vary between runs. The same question asked twice returns the same substance in different words. Track changes in source_domains, rather than string equality on answer_text, which is not.
• Some answers cite nothing at all. A short conversational reply with an empty sources array is a real answer, not a parse failure. answer_block_count tells you how substantial it was.
• Set country and language together. They control the market and the language of the answer, and mismatching them gives you a result that suits neither audience.
• Questions run a few at a time inside the run. A list of many questions can take minutes. Split large sets into batches so they fit the actor run timeout.
• Free Apify plans are capped at 10 rows per run. Split larger lists across runs or upgrade to remove the cap.
• products contains product mentions when Google includes them, with catalog identifiers and links. It stays an empty list for answers without products.
• An empty sources list can occur even in a detailed answer. Google sometimes provides the answer separately from its source cards. Such an answer is still billed because answer_text is present.

FAQ

Can I scrape Google AI Mode without an API key? Yes. Run it from the Apify Store or call it through the Apify API. No Google account, key or quota is involved.

Is AI Mode the same as the AI Overview above the search results? No. They are separate surfaces and they return different answers citing different pages. For the overview above the blue links use the Google AI Overview Scraper below.

Which sites does Google cite in AI Mode? source_domains gives the distinct list per question; sources gives the full card per cited page including the title and the snippet used.

Why does the same question return slightly different text each time? The answer is generated per request. Text and cited sources can both change. Compare repeated runs and keep the question, market and language consistent when measuring citation presence.

Is there an official Google AI Mode API? No. Google does not publish an API for AI Mode. This actor gives you the same answer as structured JSON through the Apify API, billed per answered question.

Which countries and languages does it support? Any two-letter country and language code Google accepts. The values are passed on as given and echoed in every row, so you can mix markets across runs.

Can I use this to track whether my domain is cited? Yes. Run a fixed question list on a schedule and check for your domain in source_domains per run. Presence over time is the metric.

Related Actors

• Google AI Overview Scraper for the AI summary above the ordinary search results.
• Google AI Product Answers for AI answers about a specific product, from a barcode or model number.
• Google Search Scraper for the ordinary organic results.