2026-10-26, Google · GEO · Fundamentals

Best Google AI Mode scraper APIs for 2027: fields and prices

Google AI Mode answers a search in prose, cites pages in [Label +N] chips, and adds product cards, local results and videos depending on the question. Google publishes no API for those answers, so every option here collects them from the AI Mode page. The vendors differ in which blocks they parse, whether they target a city, whether they continue a conversation, and how they bill.

Facts about each vendor come from its own pages, linked and checked on 2026-10-26; “not documented” means not found on those pages on that date. This is a buyer’s guide for 2027 budgets with facts as of October 2026; re-check prices before you commit.

Comparison

ScrapingBee also has a Google AI Mode page for its general scraper, with render_js, country_code, extract_rules and ai_extract_rules; it documents no response fields and no credit cost per request, so it is left out of the table.

Terms used below:

• Citation chip: the [Label +N] link in the answer that names one source and counts the others behind it.
• Follow-up turn: a second prompt in the same AI Mode conversation, which some vendors continue with a token from the first response.
• Query fan-out: the searches Google runs to build the answer (Google, May 20, 2025). No vendor checked returns them.

What to measure first

AI Mode sits inside Google Search, so most teams track it with the same keywords as their rankings. Pick the metric, then the field:

1. Mention rate: the answer names your brand. Any vendor’s answer text works.
2. Citation share: your domain among the sources, and whether it is the named source in a chip. Needs sources in order, ideally with chip grouping.
3. Product share: which stores hold the product slots and ads. Needs parsed shopping and ad blocks.
4. Local presence: whether your business appears in the local results, city by city. Needs local blocks and city targeting.
5. Conversation depth: how the answer changes on a follow-up. Needs a conversation token.

Metrics 1 and 2 work on most vendors. Metrics 3 to 5 narrow the list quickly.

The options

Bright Data

Bright Data’s Google AI Mode scraper runs through POST /datasets/v3/scrape with dataset gd_mcswdt6z2elth3zqr2. Inputs are a url containing udm=50, a prompt, a country, hl and index. Output fields include answer_text, answer_text_markdown, answer_html, links_attached with position, citations with title, domain and a cited flag, shopping and ads.

A synchronous call that runs past 1 minute returns HTTP 202 with a snapshot_id; large jobs go through /trigger, with delivery by webhook or to S3, GCS, Pub/Sub, Azure, Snowflake or SFTP. Pricing (checked 2026-10-26, AI scrapers page): 5K records a month free, $1.50 per 1K pay as you go, Scale $499 a month for 384,000 records then $1.30 per 1K. Failed deliveries are not charged.

It is the lowest list price per answer with shopping and ads parsed. City targeting is not documented, which rules it out for local grids.

Fits bulk country-level collection delivered to a warehouse.

DataForSEO AI Mode SERP

DataForSEO sells AI Mode as part of its SERP API. The live advanced endpoint takes a keyword (up to 700 characters), a location by name, code or coordinate, a language, device and os. The answer arrives as an ai_overview item with markdown, references and typed elements: text, video, table, expanded, code block, ai_overview_shopping (product ID, price, seller, rating) and ai_overview_paid for ads, plus refinement_chips at result level.

Pricing (checked 2026-10-26, AI Mode pricing): $0.0012 per SERP in the standard queue, $0.0024 priority, $0.004 live; calculate_rectangles doubles the price. New accounts get $1 of trial credit.

It is the cheapest option with coordinate-level location and typed shopping and ad items. The item names reuse its AI Overview model (ai_overview_*), so map them to your own record, and the lowest price is a queue.

Fits SEO platforms already on DataForSEO’s SERP API.

Oxylabs

The Oxylabs AI Mode target is source: "google_ai_mode" with a query under 400 symbols, render: "html", parse: true and geo_location. The parsed result has response_text, citations, links and prompt. Results come by push-pull (GET /v1/queries/{job_id}/results) or callback_url.

Pricing (checked 2026-10-26, AI targets page): from $1.50 per 1K results, down to $0.90 per 1K above 2.78M; Starter is $30 a month, with $5 of welcome credit.

The parsed fields stop at text and citations; shopping, local results and ads are not documented. The 400-symbol query limit is short for conversational prompts.

Fits Oxylabs customers who need AI Mode text and citations.

SearchApi

SearchApi’s AI Mode API uses engine=google_ai_mode with q, gl, hl, location or uule, an image url, and a conversation_token for follow-up turns. It returns text_blocks with reference_indexes, markdown, reference_links, local_results, web_results and inline_shopping.

Pricing (checked 2026-10-26, pricing): 100 free requests; Developer is $40 a month for 10,000 searches, down to $1 per 1K at the 5M tier. Only searches with a 200 status are charged. Async delivery is not documented.

City targeting, local results and conversation turns come in one GET.

Fits teams already on SearchApi for Google results.

SerpApi

SerpApi’s Google AI Mode API uses engine=google_ai_mode with q, location or uule, gl, hl, device, an image_url, and subsequent_request_token for follow-ups. The response has text_blocks with reference_indexes, reconstructed_markdown, references, shopping_results and related_questions; output can be JSON, HTML or Markdown. With async=true you fetch results later from the Searches Archive API; there is no webhook.

Pricing (checked 2026-10-26, pricing): 250 free searches a month; Starter $25 for 1,000, Developer $75 for 5,000, Production $150 for 15,000. Only successful searches count, and cached searches are free.

It has the widest request options of the SERP vendors. Per answer it is the most expensive list price here.

Fits teams on SerpApi with modest AI Mode volume.

Apify Google AI Mode Scraper

Apify’s own Google AI Mode Scraper takes queries and a countryCode, and returns text and sources[] with title, URL, description and a section (citations, attached or more). It costs from $2.00 per 1,000 results, pay per event.

It starts from a UI with no code. Shopping, local results and city targeting are not documented on the actor page.

Fits prototypes and one-off citation studies.

Decodo

Decodo’s AI Mode target is target: "google_ai_mode" with query, geo, parse and markdown, returning response_text, citations, links and prompt. Batches are supported, and results are fetched from GET https://data.decodo.com/v1/task/{task_id}/results; a webhook is not documented.

Its billing page prices its LLM templates, AI Mode included, at $1 per 1K requests on every plan, charged only for successful requests; plans start at $19 a month.

The per-request price is low and flat. Parsed output stops at text and citations.

Fits citation tracking at country level on a fixed budget.

Scrapeless

Scrapeless documents an AI Mode scraper, actor scraper.aimode, taking a required prompt and country, location or uule, shopping (on by default), should_fetch_share_link and webhook.url. It returns result_text, result_md, result_html, citations, search_result and products with stores, prices, reviews and variants.

With shopping on, a request costs twice the base rate. The base rate is not published on its pricing page.

Fits product research once you have a quoted rate; turn shopping off when you track citations only.

Scrapingdog

Scrapingdog’s AI Mode API is GET https://api.scrapingdog.com/google/ai_mode with query, country, language, location or uule. It returns text_blocks, references, shopping_results, inline_images and local_results, at 10 credits per successful request. Plans run from $40 to $350 a month (pricing); the page does not tie a credit count to each plan, so ask before modeling cost. New accounts get 100 free credits.

Fits local and product checks at city level.

HasData

HasData’s AI Mode API is GET https://api.hasdata.com/scrape/google/ai-mode with q, location, uule, gl, hl and subsequentRequestToken. It returns textBlocks[], typed as text, tables, code, shopping results and local results, and references[]. A request costs 10 credits, a failed request costs nothing, and new accounts get 100 free answers. Its page shows two different prices for its plans, so confirm the plan price before you budget.

Fits small AI Mode projects that need follow-up turns; it covers no other AI engine.

AnswerLine

POST /v1/monitor/aimode takes a prompt, a market (country or gl, 213 countries), hl for the interface language, location (Google’s City,Region,Country name) or uule for a city, and device (desktop or mobile).

curl -X POST https://api.answerline.dev/v1/monitor/aimode \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "best running shoes for wide feet",
    "country": "US",
    "location": "Chicago,Illinois,United States",
    "include": { "markdown": true }
  }'

Under result, per the API reference:

• text, and markdown when requested.
• sources[] with position, url, label and description; when Google served a /goto link, url is the resolved destination and redirectLink keeps Google’s redirect.
• citationPills[]: one entry per chip-and-source pair, grouped by citationPillId, so a [Label +2] chip becomes three entries.
• shoppingCards[] with numeric price.value, oldPrice, store, rating and a 0-indexed position.
• videos[] with platform, channel and duration.
• places[]: the place cards in the order the answer shows them, with title, placeId (Google’s ChIJ… Place ID), rating, reviews, type, priceLevel, address, status and the viewer link.
• map[]: the places Google pinned on the answer’s map, with gps_coordinates.
• ads when the answer shows sponsored products. inlineProducts[] is in the schema and comes back empty: Google shows an inline product mention without a name or price.

include.html links a copy of the AI Mode page and include.rawWire returns the page as rendered, both free; include.expandProducts is refused with a 400. POST /v1/monitor/google/goto resolves a redirect link for free.

An AI Mode request costs $0.00206 on every tier, with every field. A new account’s 100 free requests work on AI Mode, and failed requests are not charged (credits).

Chip grouping in citationPills, city targeting and one request contract across ChatGPT, Gemini, Perplexity, Copilot, Grok, Google Search and Google News come together, with async batches of up to 500 tasks, webhooks and idempotency keys. It has no follow-up turns.

Fits visibility products that report AI Mode citations by city next to other assistants.

Building it yourself

AI Mode is a Google Search surface: a home-grown collector needs per-market and per-city collection, a parser for chips, cards and local blocks, and upkeep each time Google changes the page. Build vs buy itemizes the work.

How to choose

1. Which blocks do you report on? Citations only, or also shopping, ads and local results. Check the exact field in each schema.
2. Do you need chip grouping? Citation share by chip needs the sources behind each [Label +N], not a flat link list.
3. Local work needs city-level targeting: location, uule or coordinates.
4. Single prompts or conversations? Follow-up turns are documented by SerpApi, SearchApi and HasData.
5. Queue or live? Queues cost less; daily tracking rarely needs live calls.
6. One contract or several? If ChatGPT, Gemini or Copilot are on the 2027 plan, count the mapping work per extra vendor.
7. Failure billing: most vendors here charge only for successful requests. Check retries and idempotency too.

Cost example

50 keywords in 4 US cities, collected daily for 30 days: 6,000 AI Mode answers a month. List prices checked 2026-10-26, before taxes and discounts. Recurring monthly free allowances (Bright Data’s 5K records) are netted; one-time trials (SearchApi’s 100 searches, Oxylabs’ $5 credit, AnswerLine’s 100 free requests) are not:

Scrapeless, Scrapingdog and HasData are left out: none ties a dollar price to an AI Mode request on its page. The cheapest rows do not target a city, which this example needs. AI monitoring cost planning covers budgeting across engines.

A daily AI Mode city grid on AnswerLine

Keywords × cities, submitted as one batch with a webhook:

import os, datetime, requests

API = "https://api.answerline.dev"
HEADERS = {"Authorization": f"Bearer {os.environ['API_KEY']}"}
today = datetime.date.today().isoformat()

keywords = ["best running shoes for wide feet", "running shoe store near me"]
cities = ["Chicago,Illinois,United States", "Phoenix,Arizona,United States"]
tasks = [
    {
        "taskType": "AIMODE",
        "payload": {"prompt": k, "country": "US", "location": c},
        "idempotencyKey": f"shoes-{i}-{c.split(',')[0].lower()}-{today}",
        "webhook": {"url": "https://example.com/hooks/aimode"},
    }
    for i, k in enumerate(keywords)
    for c in cities
]

r = requests.post(f"{API}/v1/async/task/batch", json=tasks, headers=HEADERS, timeout=30)
r.raise_for_status()
print(r.json()["summary"])

In the webhook handler, check whether your domain is the named source of a chip:

def chip_leads(result, domain="example.com"):
    first = {}
    for pill in result.get("citationPills", []):
        first.setdefault(pill["citationPillId"], pill["domain"])
    return [cid for cid, d in first.items() if d == domain or d.endswith("." + domain)]

The first entry per citationPillId is the source the chip names. Verify the webhook signature before trusting the body.

Pitfalls that skew AI Mode data

• Counting chip sources as equal. A [Label +3] chip names one page; the other three sit behind the count. Report “named in a chip” separately from “behind a chip”.
• Mixing AI Mode and AI Overview. They are separate surfaces with separate answers; see AI Mode answers as JSON.
• Building city names by hand. location expects Google’s canonical names; copy them from the Google Ads geotargets list.
• Comparing markets without hl. The interface language changes the answer; set it explicitly.
• One run per keyword. Answers vary between runs, so take several samples and report rates; see AI answer volatility.

Mentions and citations need separate metrics; see mentions vs citations. Every field is on the AI Mode engine page.

Keep reading