Roundup

November 2026 · 9 min read

Best Google AI Overview and AI Mode APIs: 3 Tested and Compared

Google's AI Overview and AI Mode answers, with the sources they cite, as JSON you can store or pass to a model. This guide shows developers and SEO teams which API to use for each, so you can pick one and make a first call. We sent the same three queries through three APIs live and measured what came back and how fast.

Adam Ben-Ayoun

Adam Ben-Ayoun

CTO, OpenWeb Ninja

Google AI Overview API · Google AI Mode API · SERP API · API Comparison

Best Google AI Overview and AI Mode APIs, tested and compared

Is There an Official Google AI Overview API?

No. Google's developer docs offer two related products, and neither returns the AI Overview or AI Mode answer people see in Search:

  • Custom Search JSON API: Google's overview page says it "is closed to new customers", and existing customers have until January 1, 2027 to move to an alternative. Its documented response is a list of results (title, link, snippet and similar fields) with no AI Overview field.
  • Grounding with Google Search: the Gemini API docs say it "connects the Gemini model to real-time web content". You get Gemini's own answer with source citations, not Google Search's AI Overview.

Google's own AI Overview, asked in November 2026, says the same. To get these answers as data, developers use a SERP API, and the three below are the ones we ran live.

Quick Answer: Google AI Overview and AI Mode APIs at a Glance

OpenWeb Ninja, SerpApi and SearchApi all return Google's AI Overview and AI Mode answers as JSON. All 18 of our November 2026 test requests returned an answer. SerpApi was fastest for AI Overviews (median 0.73 s), OpenWeb Ninja for AI Mode (2.02 s). Lowest entry price: OpenWeb Ninja for AI Overviews ($2.50 per 1,000), SearchApi for AI Mode ($4.00).

OptionData coveredPrice per 1,000 requestsFree tierOur live run (3 queries)
OpenWeb NinjaAI Overview text and reference links; AI Mode answer, reference links and a session token for follow-upsAI Overview $2.50 on Pro ($25 for 10,000). AI Mode $5.00 on Pro ($25 for 5,000)50 requests/mo on each API, no cardAI Overview: 3 of 3, 7 to 8 blocks, 9 to 12 sources, median 2.61 s. AI Mode: 3 of 3, 4 to 10 blocks, 0 to 3 sources, median 2.02 s
SerpApiAI Overview inside the Google results (or by page_token); AI Mode text blocks, references and markdown$25.00 on Starter ($25 for 1,000 searches)250 searches/mo, 50 an hourAI Overview: 3 of 3, 9 to 15 blocks, 9 to 12 sources, median 0.73 s. AI Mode: 3 of 3, 8 to 9 blocks, 0 to 7 sources, median 4.60 s
SearchApiAI Overview inside the Google results (or by page_token); AI Mode text blocks, reference links, markdown and a conversation token$4.00 on Developer ($40 for 10,000 requests)100 requests on sign-up, no cardAI Overview: 3 of 3, 9 to 14 blocks, 9 to 12 sources, median 3.36 s. AI Mode: 3 of 3, 8 to 16 blocks, 0 to 8 sources, median 5.28 s

Blocks are the text blocks each API parsed the answer into; sources are the cited links. Every request in our run cost 1 request or search on each account. The order follows the combined median time for one AI Overview plus one AI Mode answer (4.63 s, 5.33 s, 8.64 s). All three were tested live with our own keys. We build OpenWeb Ninja (#1), so read its row with that in mind.

How We Tested

  • Three queries that show an AI Overview: "how does a heat pump work", "what is retrieval augmented generation" and "how to remove a red wine stain".
  • Both endpoints on each API: one AI Overview request and one AI Mode request per query, 18 requests in all, in one session from one machine in November 2026. Default settings (US, English), and no_cache=true on SerpApi.
  • The code below is what ran: each vendor's sample was run as shown, with our key and the query swapped in. Response time is the round trip of each request, timed around fetch. The outputs and trimmed responses under the OpenWeb Ninja and SerpApi samples come from a later run of that code, so their counts can differ from the tables here.
  • Same answer check: for each query, all three APIs returned the same AI Overview, with the same opening sentence. They returned the same number of sources on two queries and 10 vs 11 on the third. Block counts differ because each API splits lists differently.
  • AI Mode: answers are generated per request, so the wording differs between calls, even for the same question. We compare what each response contains, not the wording.
  • Sample size: three queries, one run each. Treat it as a smoke test, not a benchmark.
  • Pricing: read from each provider's own pricing page in November 2026.
AI Overview queryOpenWeb NinjaSerpApiSearchApi
how does a heat pump work7 blocks, 12 sources, 2.75 s9 blocks, 12 sources, 0.69 s9 blocks, 12 sources, 4.57 s
what is retrieval augmented generation7 blocks, 9 sources, 2.61 s9 blocks, 9 sources, 1.00 s9 blocks, 9 sources, 3.27 s
how to remove a red wine stain8 blocks, 10 sources, 2.56 s15 blocks, 11 sources, 0.73 s14 blocks, 11 sources, 3.36 s
AI Mode queryOpenWeb NinjaSerpApiSearchApi
how does a heat pump work4 blocks, 0 sources, 2.02 s9 blocks, 0 sources, 4.60 s8 blocks, 0 sources, 5.89 s
what is retrieval augmented generation4 blocks, 3 sources, 1.86 s8 blocks, 7 sources, 1.91 s8 blocks, 7 sources, 5.28 s
how to remove a red wine stain10 blocks, 3 sources, 2.05 s8 blocks, 0 sources, 6.87 s16 blocks, 8 sources, 3.64 s
#1

OpenWeb Ninja AI Overviews and Google AI Mode APIs

this site

Best for: the fastest AI Mode answers in our run, follow-up questions, and the lowest entry price per AI Overview

OpenWeb Ninja has two GET APIs on one key: the AI Overviews API takes a search query as q, and the Google AI Mode API takes a prompt. The code below asks for an AI Overview, an AI Mode answer, and a follow-up question in the same AI Mode conversation.

# openwebninja_ai_answers.mjs

const headers = { "x-api-key": "YOUR_API_KEY" };
const q = "how does a heat pump work";

// AI Overview for a Google search
const aio = await fetch(
  `https://api.openwebninja.com/ai-overviews/ai-overviews?${new URLSearchParams({ q })}`,
  { headers }
).then((r) => r.json());
console.log(aio.data.search_returned_ai_overviews, aio.data.text_parts.length, aio.data.reference_links.length);

// AI Mode answer, then a follow-up in the same conversation
const mode = await fetch(
  `https://api.openwebninja.com/google-ai-mode/ai-mode?${new URLSearchParams({ prompt: q })}`,
  { headers }
).then((r) => r.json());
console.log(mode.data.reply_parts[0].text);

const followUp = await fetch(
  `https://api.openwebninja.com/google-ai-mode/ai-mode?${new URLSearchParams({
    prompt: "Which one is cheaper to run?",
    session_token: mode.data.session_token,
  })}`,
  { headers }
).then((r) => r.json());
console.log(followUp.data.reply_parts[0].text);

Output of this code

true 6 9
A heat pump doesn't actually "create" heat. Instead, it works like a refrigerator in reverse, using electricity to transfer heat from one place to another.
When it comes to running costs, a heat pump is almost always cheaper to run for heating than traditional systems like oil, propane, or standard electric resistance heaters (like baseboard heat).

The first line is the AI Overview flag, the number of text parts and the number of reference links. The responses below are from the same run of this code, a separate call from the tables above: this time the AI Overview had 6 text parts and 9 links, and the AI Mode answer had 8 reply parts and no sources.

AI Overview response from this code run, trimmed

{
  "status": "OK",
  "request_id": "7460ba0b-75c1-4e4d-9c9c-5e16d7f975e5",
  "parameters": {
    "q": "how does a heat pump work",
    "nfpr": "0",
    "gl": "us",
    "hl": "en",
    "device": "desktop"
  },
  "data": {
    "text_parts": [
      {
        "type": "paragraph",
        "text": "A heat pump works by moving heat from one place to another rather than creating it by burning fuel or using pure resistance, functioning essentially like a two-way refrigerator or air conditioner."
      },
      ... 5 more
    ],
    "reference_links": [
      {
        "title": null,
        "link": "https://www.youtube.com/watch?v=QykwWs3L1W8&t=4",
        "snippet": null,
        "source": "Carrier",
        "date": null
      },
      ... 8 more
    ],
    "search_returned_ai_overviews": true
  }
}

AI Mode response from this code run, trimmed

{
  "status": "OK",
  "request_id": "cd2ebd3d-e330-4c75-9d04-1c40490e86d2",
  "parameters": {
    "prompt": "how does a heat pump work",
    "gl": "us",
    "hl": "en"
  },
  "data": {
    "reply_parts": [
      {
        "type": "paragraph",
        "text": "A heat pump doesn't actually \"create\" heat. Instead, it works like a refrigerator in reverse, using electricity to transfer heat from one place to another."
      },
      ... 7 more
    ],
    "reference_links": [],
    "session_token": "Q21vd1lUWkpSVEl5Y2pkdlIyODNOWGxK..."
  }
}
  • Parameters: AI Overviews: q, location, gl, hl. AI Mode: prompt, session_token, location or uule, gl, hl, and PDF attachments.
  • Free tier: 50 requests a month on each API, no credit card.
  • Paid plans: AI Overviews: Pro $25 for 10,000 requests, Ultra $75 for 50,000, Mega $150 for 200,000 ($0.75 per 1,000), Pay As You Go $0.005 per request. AI Mode: Pro $25 for 5,000, Ultra $75 for 25,000, Mega $150 for 75,000 ($2.00 per 1,000), Pay As You Go $0.008 per request.
  • Our run: all 6 requests returned an answer, 1 request each on our usage meter. AI Overview median 2.61 s, AI Mode median 2.02 s, responses 3 to 6 KB. Our first follow-up request returned a 503 "Temporary error"; it was not counted on the meter, and a re-run worked.
  • AI Mode sources: our AI Mode answers cited fewer sources than the other two. For "what is retrieval augmented generation" we returned 3 sources, against 7 on both SerpApi and SearchApi. Across the 3 queries: 6 sources on OpenWeb Ninja, 7 on SerpApi, 15 on SearchApi.
  • Sources: 28 of the 31 AI Overview links were direct page URLs. The other 3 were Google redirect links (google.com/goto).
  • For agents: both APIs have an llms.txt (AI Overviews, AI Mode), and the MCP server exposes the APIs as tools.
#2

SerpApi Google AI Overview and AI Mode APIs

monthly plans

Best for: the fastest AI Overviews in our run, inside a full Google results page

SerpApi returns the AI Overview as ai_overview in its Google search response (engine=google). Its AI Overview docs describe a second request with engine=google_ai_overview and a page_token that "expires within 1 minute of the search". AI Mode is engine=google_ai_mode. The code below handles both cases and ran as shown with our key.

# serpapi_ai_answers.mjs

const apiKey = "YOUR_API_KEY";
const q = "how does a heat pump work";
const serp = (params) =>
  fetch(`https://serpapi.com/search.json?${new URLSearchParams({ ...params, api_key: apiKey })}`).then((r) => r.json());

// AI Overview: inline in the Google results, or one more call with page_token
let { ai_overview } = await serp({ engine: "google", q, no_cache: "true" });
if (ai_overview?.page_token && !ai_overview.text_blocks) {
  ({ ai_overview } = await serp({ engine: "google_ai_overview", page_token: ai_overview.page_token }));
}
if (!ai_overview) {
  console.log("No AI Overview for this query");
} else {
  console.log(ai_overview.text_blocks.length, ai_overview.references.length);
}

// AI Mode answer
const mode = await serp({ engine: "google_ai_mode", q, no_cache: "true" });
console.log(mode.text_blocks[0].snippet);

Output of this code

12 11
A heat pump doesn't actually "create" heat. Instead, it works like a refrigerant-based conveyor belt that moves heat from one place to another.

Output with q set to "facebook login", a query with no AI Overview

No AI Overview for this query
Access your account directly through the official Facebook Login page.

The responses below are from the same run of this code, a separate call from the tables above: 12 AI Overview text blocks and 11 references, and 9 AI Mode text blocks.

AI Overview part of the Google response from this code run, trimmed

{
  "ai_overview": {
    "text_blocks": [
      {
        "type": "paragraph",
        "snippet": "A heat pump does not generate heat by burning fuel; instead, it works like a refrigerator in reverse to move existing heat from one place to another using electricity. Even when ...",
        "reference_indexes": [1, 2, 3, 4]
      },
      ... 11 more
    ],
    "references": [
      {
        "index": 0,
        "title": "How Does a Heat Pump Work? Simple Animated Explanation (Under 2 ...",
        "link": "https://www.youtube.com/watch?v=psMkHcUddwk&t=1",
        "source": "YouTube"
      },
      ... 10 more
    ]
  }
}

AI Mode response from this code run, trimmed

{
  "search_parameters": {
    "engine": "google_ai_mode",
    "q": "how does a heat pump work",
    "device": "desktop"
  },
  "text_blocks": [
    {
      "type": "paragraph",
      "snippet": "A heat pump doesn't actually \"create\" heat. Instead, it works like a refrigerant-based conveyor belt that moves heat from one place to another."
    },
    ... 8 more
  ],
  "reconstructed_markdown": "A heat pump doesn't actually \"create\" heat. Instead, it works like a r..."
}
  • Parameters: AI Mode: q, location or uule, gl, hl, continuable and subsequent_request_token for follow-ups, and no_cache, per its AI Mode docs.
  • Free tier: 250 searches a month, 50 an hour.
  • Paid plans: Starter $25 for 1,000 searches, Developer $75 for 5,000, Production $150 for 15,000 ($10.00 per 1,000). Only successful searches count toward the plan.
  • Our run: all 6 requests returned an answer, 1 search each on our account. All 3 AI Overviews came inside the first Google response, with no page_token request. AI Overview median 0.73 s, AI Mode median 4.60 s.
  • Response size: the AI Overview comes with the whole results page, 82 to 86 KB per response in our run, including organic results and related questions.
  • Sources: all 32 AI Overview links were direct page URLs.
#3

SearchApi Google AI Overview and AI Mode APIs

monthly plans

Best for: the lowest entry price per AI Mode answer, with markdown and a conversation token

SearchApi works like SerpApi: the AI Overview comes inside its Google search response (engine=google), and its AI Overview docs describe a google_ai_overview request with a page_token that expires "in less than 1 minute". AI Mode is engine=google_ai_mode, with the key in an Authorization header. The code below is the code from our test, plus the check for a missing ai_overview, which we added afterwards and have not run on SearchApi: our trial credits ran out.

# searchapi_ai_answers.mjs

const headers = { Authorization: "Bearer YOUR_API_KEY" };
const q = "how does a heat pump work";
const search = (params) =>
  fetch(`https://www.searchapi.io/api/v1/search?${new URLSearchParams(params)}`, { headers }).then((r) => r.json());

// AI Overview: inline in the Google results, or one more call with page_token
let { ai_overview } = await search({ engine: "google", q });
if (ai_overview?.page_token && !ai_overview.text_blocks) {
  ai_overview = await search({ engine: "google_ai_overview", page_token: ai_overview.page_token });
}
if (!ai_overview) {
  console.log("No AI Overview for this query");
} else {
  console.log(ai_overview.text_blocks.length, ai_overview.reference_links.length);
}

// AI Mode answer
const mode = await search({ engine: "google_ai_mode", q });
console.log(mode.text_blocks[0].answer);

Output of this code

9 12
A heat pump works a bit like a refrigerator in reverse. Instead of creating heat, it simply moves heat from one place to another.

AI Overview part of the Google response, trimmed

{
  "ai_overview": {
    "text_blocks": [
      {
        "type": "paragraph",
        "answer": "A heat pump works by moving heat from one place to another instead of burning fuel to create it, acting like a two-way air conditioner that can both heat and cool your home.",
        "reference_indexes": [0, 1, 2]
      },
      ... 8 more
    ],
    "markdown": "A heat pump works by moving heat from one place to another instead of ...",
    "reference_links": [
      {
        "index": 0,
        "title": "What Is A Heat Pump And How Does It Work? - Carrier",
        "link": "https://www.google.com/goto?url=CAESpgEBxCJF1-zqiKeZqJB5MbEn...",
        "snippet": "What Is a Heat Pump And How Does a Heat Pump Work?...",
        "source": "Carrier",
        "favicon": "https://encrypted-tbn3.gstatic.com/favic...",
        "thumbnail": "https://encrypted-tbn3.gstatic.com/image..."
      },
      ... 11 more
    ]
  }
}

AI Mode response, trimmed

{
  "search_parameters": {
    "engine": "google_ai_mode",
    "q": "how does a heat pump work",
    "gl": "us",
    "hl": "en"
  },
  "text_blocks": [
    {
      "type": "paragraph",
      "answer": "A heat pump works a bit like a refrigerator in reverse. Instead of creating heat, it simply moves heat from one place to another."
    },
    ... 7 more
  ],
  "markdown": "A heat pump works a bit like a refrigerator in reverse. Instead of cre...",
  "pagination": {
    "conversation_token": "eyJtc3RrIjoiQVV0RXhmRDBUbDQ5eDhv..."
  }
}
  • Parameters: AI Mode: q, location or uule, gl, hl, and conversation_token for follow-ups, per its AI Mode docs. The docs say a conversation "cannot be continued forever" and returns a 400 error once it ends.
  • Free tier: 100 free requests on sign-up, no credit card.
  • Paid plans: Developer $40 for 10,000 requests, Production $100 for 35,000 ($2.86 per 1,000), BigData $250 for 100,000 ($2.50). Failed requests are not billed.
  • Our run: all 6 requests returned an answer, 1 credit each on our account. All 3 AI Overviews came inside the first Google response. AI Overview median 3.36 s, AI Mode median 5.28 s.
  • Sources: 30 of the 32 AI Overview links were Google redirect links (google.com/goto). The docs say link=resolved returns "the direct destination URL instead". Our run used the default.

AI Overview or AI Mode: Which Endpoint Fits?

AI OverviewAI Mode
InputA search query, as typed into GoogleA question or prompt
When you get an answerOnly for queries where Google shows an AI OverviewFor every prompt in our run (9 of 9)
Follow-up questionsNoYes, with a session or conversation token
Response time in our runMedian 0.73 to 3.36 sMedian 2.02 to 5.28 s
A common useChecking which pages Google cites for keywords you already trackChecking how Google answers questions about a brand or topic, and research sessions with follow-ups

To find out which of your keywords show an AI Overview before you fetch it, the Real-Time Web Search API returns a has_ai_overviews flag with the organic results.

Queries With No AI Overview

Google shows an AI Overview for some queries and not others, so a keyword list will hit both. We sent "facebook login" to each API in November 2026, a few seconds apart.

OptionWhat came backBilled
OpenWeb Ninjasearch_returned_ai_overviews: false, with empty text_parts and reference_links1 request, checked on our usage meter
SerpApiNo ai_overview key. The organic results (7) and related searches still came back1 search on our account. Its pricing FAQ says "Only successful searches are counted" and that empty result sets also count as 1 search
SearchApiGoogle showed an AI Overview on SearchApi's search for this query, so we could not see its no-overview response1 credit on our account. Its pricing page says "Only successful searches with a 200 status code incur charges"

The SerpApi and SearchApi samples above check for a missing ai_overview before reading it, so a keyword list does not stop on the first query without one.

Which API Fits Your Use Case

  • Tracking which pages Google cites: you store the cited URL per keyword per day, so direct page URLs save a step. SerpApi returned direct URLs for all 32 AI Overview links, OpenWeb Ninja for 28 of 31, and SearchApi returns them with link=resolved. Our AI Mode brand tracking guide shows a full tracker.
  • You already track organic rankings and want the AI Overview too: SerpApi and SearchApi return the AI Overview inside the full results page, 82 to 86 KB per SerpApi response in our run with organic results and related questions included. On OpenWeb Ninja, the Real-Time Web Search API does the same in one call with fetch_ai_overviews=true on /search: 8 organic results and the AI Overview in about 9.6 KB in our check. Its docs say a request uses two credits instead of one when an AI Overview is available.
  • Many queries a day: monthly cost at volume decides it. For 100,000 requests a month on published plans: OpenWeb Ninja $150 for AI Overviews (Mega, 200,000 included) or $200 for AI Mode (Mega, 75,000 included, then $0.002 each), SearchApi $250 (BigData) and SerpApi $725 (Searcher). The SerpApi and SearchApi prices buy the whole results page, organic results included, so they are not like-for-like with a dedicated AI Overviews API.
  • Follow-up questions in an LLM tool: all three support AI Mode follow-ups. OpenWeb Ninja AI Mode was the fastest in our run (median 2.02 s) and puts ChatGPT, Gemini and Copilot APIs on the same key, but it cited fewer sources (3 against 7 on both competitors for "what is retrieval augmented generation"). If the sources matter more than speed, compare all three on your own prompts. For wiring answers into a model, see Give Your AI Agent Live Web Search.
  • Answers for a specific city: AI Mode answers about local businesses, weather or events depend on where the search comes from. All three take location for both AI Overviews and AI Mode: OpenWeb Ninja (AI Overviews and AI Mode docs), SerpApi (Google Search and AI Mode docs) and SearchApi (Google Search and AI Mode docs). SerpApi and SearchApi also take uule on both, and OpenWeb Ninja takes it for AI Mode. A common practice is to set the city you report on and keep it fixed between runs.

For organic results and the rest of the results page, see Best SERP APIs in 2026 and OpenWeb Ninja vs SerpApi. More SEO use cases: SEO and SERP research.

FAQ

How do I access Google AI Overviews through an API?

Is there an official Google API for AI Overviews?

Is there a free AI Overview API?

How do I get Google AI Mode answers as JSON?

Can I ask AI Mode a follow-up question through the API?

Are the AI Overview and AI Mode results live?

Can I track ChatGPT and Gemini answers with the same key?

Adam Ben-Ayoun

Written by

Adam Ben-Ayoun

CTO at OpenWeb Ninja

Adam leads engineering at OpenWeb Ninja, building the APIs and infrastructure that make public web data accessible to developers and AI agents.

Connect on LinkedIn

Get Google's AI Answers as JSON

50 free requests a month on each API, no credit card. AI Overviews, AI Mode, ChatGPT, Gemini and 50+ other APIs on one key.