2026-09-17, Google · SERP API · Tutorials

Google AI Overview API: get AI Overviews as JSON in 2026

A Google AI Overview is the generated summary Google shows above the results for some queries, with links to the sources it draws on. An AI Overview API returns that summary and its sources as structured data. On vendor pages checked 2026-09-17, nine APIs document this. They differ on whether the overview comes back in the same request, which fields you get (especially the citation mapping), and how it is billed. After the comparison, the last sections show requests to AnswerLine with include.aioverview.

Comparison table

Not included: Serper does not document AI Overview on its public pages, and ScraperAPI’s documented Google SERP output has no AI Overview field. Zenserp’s pricing page lists “AI Overview Access” on every plan, but its documentation lists AI Search as “Coming Soon” and its API reference has no AI Overview parameter (docs), so confirm with them before relying on it.

Fields each vendor documents

For tracking, the citations carry the value: which sources the overview used and which sentence cites which source.

There are two data models. Most vendors return ordered text blocks with reference indexes, which suits re-rendering the overview. AnswerLine returns text or markdown plus citation pills, one entry per chip-and-source pair sharing a citationPillId, which suits counting how often a domain is cited inline versus only listed in the source rail. Pick the one that matches your report.

Vendor notes

SerpApi

The overview usually comes back inline in engine=google results. When Google loads it separately, the response carries page_token and serpapi_link for a follow-up call to the google_ai_overview engine. Its docs give two lifetimes for the token: “within 4 minutes” on the AI Overview results page and “within 1 minute” on the engine page. They also state that the AI Overview block is “only seen for English searches (hl=en) with a limited range of countries (gl)”. Neither page says whether the follow-up counts toward monthly searches.

Fits teams already using SerpApi for other engines that want the overview in the same integration.

SearchApi

Also inline, with a google_ai_overview engine for deferred overviews. Its docs say page tokens “expire in less than 1 minute”. The follow-up engine adds a markdown rendering. It bills only HTTP 200 responses; whether the follow-up is billed as a separate search is not stated.

Fits teams wanting GET-style integrations at lower per-search prices.

DataForSEO

The overview is an item in Google organic results. The load_async_ai_overview parameter loads overviews Google delivers asynchronously and adds one base price, $0.0006 in the standard queue (organic pricing). A separate AI Mode SERP endpoint costs $1.20 per 1,000 standard.

Fits high-volume, queue-tolerant AI Overview tracking at the lowest per-SERP price.

Bright Data

Add brd_ai_overview=2 to a Google request. Its docs say this “increases the likelihood” of receiving the overview and adds about 5 to 10 seconds. If no overview is returned, a warning comes back in a response header (query parameters). No separate overview price was published on the pages checked.

Fits teams on Bright Data that need many countries and cities.

Oxylabs

Use the google_search source with render: "html", a user_agent_type and parse: true. Results come back in an ai_overviews array, which can hold several blocks when overviews appear in other SERP sections too. Rendering is required, so the rendered rate applies: $1.35 per 1,000 on Micro down to $0.95 on Corporate.

Fits multi-site scraping pipelines with batch delivery.

ScrapingBee

The overview comes back in ai_overviews on regular requests only, because it requires JavaScript rendering. A regular Google request costs 15 credits instead of 10.

Fits smaller volumes where a general scraping account is already in place.

Scrape.do

The ai_overview field has a status: complete, deferred, unavailable or null. A complete overview costs nothing extra on the 10-credit search. A deferred one returns a session_key for GET /plugin/google/search/ai-overview, which costs 5 credits, works once and expires after 90 seconds.

Fits cost-sensitive teams that can make the follow-up call immediately.

Apify

The standalone Google AI Overviews Scraper is priced per captured overview, from $1.00 per 1,000 on higher tiers ($2.00 on Starter), and returns text and sources. The Google Search Results Scraper captures overviews as an add-on event on top of the page event. Both charge the overview event only when an overview is captured.

Fits scheduled, lower-volume jobs and no-code workflows.

AnswerLine

The overview is requested as part of the Google Search call and returned next to organic results, People Also Ask and ads, so one response shows whether there is an overview, whether it cites you, and where you rank organically.

• Request: include.aioverview: {} or { "markdown": true }.
• Response: result.aioverview with text, markdown (when requested), sources, citationPills, relatedLinks, videos and ads. It is null when Google showed no overview.
• People Also Ask: include.paaAioverview: true also fills AI-Overview-type People Also Ask items with markdown and sources, at no credit cost beyond the AI Overview add-on; responses take longer.
• Cost: +2 credits once per request that asks for it, on top of 3 credits per async task. A synchronous call adds 2 more, and extra result pages add 2 credits each (credits).
• Targeting: country (or gl), hl, location or uule, and device.
• Scope: Google web search only; the Google News endpoint has no overview option. Credits reset monthly.

Fits AI-visibility and SEO products that report AI Overview citations next to organic rank and next to ChatGPT, Perplexity and AI Mode answers.

Requesting an AI Overview from AnswerLine

Synchronous request

curl -X POST https://api.answerline.dev/v1/monitor/google \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "how to lower a credit card interest rate",
    "country": "US",
    "hl": "en",
    "device": "desktop",
    "include": { "aioverview": { "markdown": true } }
  }'

The response shape, trimmed, with placeholder values:

{
  "success": true,
  "result": {
    "organicResults": [{ "position": 1, "title": "…", "link": "https://example.com/…", "page": 1 }],
    "aioverview": {
      "text": "…",
      "markdown": "…",
      "sources": [{ "position": 1, "url": "https://example.com/guide", "label": "…", "description": "…" }],
      "citationPills": [{ "citationPillId": 1, "label": "…", "url": "https://example.com/guide", "domain": "example.com", "position": 1 }],
      "relatedLinks": []
    }
  }
}

The X-Credits-Charged response header shows what the call cost.

Python: citation share for a domain

import os
import requests

API = "https://api.answerline.dev/v1/monitor/google"
HEADERS = {"Authorization": f"Bearer {os.environ['API_KEY']}"}

def overview(query: str, country: str = "US") -> dict | None:
    body = {"query": query, "country": country, "include": {"aioverview": {}}}
    resp = requests.post(API, headers=HEADERS, json=body, timeout=320)
    resp.raise_for_status()
    return resp.json()["result"].get("aioverview")

def cites(aio: dict | None, domain: str) -> dict:
    if aio is None:
        return {"has_overview": False, "in_sources": False, "inline_pills": 0}
    hosts = [s["url"].split("/")[2] for s in aio.get("sources", [])]
    pills = {p["citationPillId"] for p in aio.get("citationPills", []) if p["domain"].endswith(domain)}
    return {"has_overview": True, "in_sources": any(h.endswith(domain) for h in hosts), "inline_pills": len(pills)}

print(cites(overview("how to lower a credit card interest rate"), "example.com"))

inline_pills counts distinct citation chips that point at the domain, which separates “cited in the answer” from “listed in the source rail”. Metrics built on this are covered in monitoring Google AI Overviews and mentions vs citations.

Async batch with a webhook, for keyword lists

For hundreds or thousands of keywords, submit tasks instead of synchronous calls. Tasks cost the base price, wait for a concurrency slot instead of failing, and post the result to your endpoint.

import { Client } from "@answerline/sdk";

const client = new Client({ apiKey: process.env.API_KEY! });
const keywords = ["how to lower a credit card interest rate", "balance transfer vs personal loan"];
const day = "2026-09-17";

const batch = await client.createBatch(
  keywords.map((query) => ({
    taskType: "GOOGLE" as const,
    payload: { query, country: "US", include: { aioverview: { markdown: true } } },
    idempotencyKey: `aio-${day}-US-${query}`,
    webhook: { url: "https://your-app.example/hooks/answers" },
  })),
);
for (const item of batch.results) if (!item.success) console.warn(item.index, item.error.code);

Batches take 1 to 500 tasks. Each webhook delivery is signed; verify it before trusting the body (webhooks, verifying signatures). A full scheduled pipeline is in tracking AI Overviews with async batches and webhooks.

Selection criteria

1. Citation granularity. If your report says “cited in the answer”, you need a sentence-to-source mapping (reference indexes or citation pills), not only a source list.
2. Same-request delivery. Follow-up tokens that expire in one to four minutes, or 90 seconds, force the follow-up into the same worker as the search. Check that your queue design allows it.
3. Billing when there is no overview. Many queries have none. Apify bills the overview event only on capture; DataForSEO’s load_async_ai_overview and AnswerLine’s add-on are priced per request that asks for it. Measure your keyword set’s overview rate on a sample before budgeting.
4. Rendering price. Rendering-based vendors bill the rendered rate on every request that needs the overview.
5. Markets and languages. Test the countries and languages you report on; SerpApi’s docs note limited coverage outside English.
6. Organic context. Overview data is most useful next to organic rank from the same request, so prefer vendors that return both in one response.
7. Delivery at volume. Callbacks or webhooks, idempotent task submission and batch limits decide how much glue code you write.

Cost example

20,000 keywords a month, first page plus AI Overview on every request. List prices checked 2026-09-17.

The “or more” and “up to” cells are open questions to settle in a trial. Run a sample of your keywords on free tiers and compare the billing logs.

Pitfalls

• Treating null as a failure. No overview is a valid result and the denominator of your presence rate, so store it.
• Comparing overviews across devices and locations. A mobile overview in one city and a desktop overview with no location are different observations; keep them in separate series.
• Grouping by redirect URLs. Google sometimes serves links through redirects. AnswerLine returns the destination url and, when present, the redirectLink separately, so group by url or domain.
• Expecting stability. Overviews change between runs for the same query. Sample repeatedly before drawing conclusions; see AI answer volatility.

For AI Mode, Google’s separate conversational surface, see capturing Google AI Mode answers. Everything the Google endpoint returns is on the Google Search engine page.

Keep reading