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

Google Custom Search JSON API alternatives and migration

Google has closed the Custom Search JSON API to new customers, and existing customers have until January 1, 2027 to move to something else. If your product calls customsearch/v1 today, you need a replacement for both the data (web results for a query) and the request shape your code was built around.

What Google changed, and when

All dates below come from Google’s own pages, checked 2026-09-17.

Two details matter for planning. First, the help page notes that once “Search the entire web” is switched off on an existing engine, it cannot be switched back on, so do not toggle it while you evaluate options. Second, for existing customers the overview page still lists the pricing: 100 search queries per day free, then $5 per 1,000 queries, up to 10,000 queries per day. Your bill stays predictable until the deadline; your integration does not survive it.

What the Custom Search JSON API returned

The API returned JSON results from a Programmable Search Engine that you configured, identified by its engine ID (cx) and called with an API key. Per Google’s cse.list reference:

• num accepts integers from 1 to 10: at most ten results per request.
• start is the index of the first result, so start=11 begins the second page.
• The API never returns more than 100 results; start + num above 100 is an error.
• Filters include gl (boosts results from a country), hl (interface language), lr (restrict to a language, such as lang_ja), cr (restrict to documents from a country), dateRestrict (d[number], w[number], m[number], y[number]), safe (active or off), siteSearch with siteSearchFilter, exactTerms, excludeTerms, orTerms and fileType.

The Search response holds queries, searchInformation, spelling, promotions, context and items[], where each item has title, htmlTitle, link, displayLink, snippet, htmlSnippet, formattedUrl, pagemap, mime, fileFormat and a few more.

That is a list of links, not the page a searcher sees. Google’s Programmable Search Engine vs Google.com help page is explicit about the gap: an engine configured to search the entire web is limited to a subset of the Google Web Search corpus, is designed to emphasize results from your own sites, and lacks Google Web Search features such as Oneboxes, real-time results and universal search. If you used Custom Search as a proxy for “what ranks on Google”, your numbers were already an approximation.

What a full SERP response contains

A SERP API reads the Google results page itself. For comparison, this is what a Google Search request to this API (POST /v1/monitor/google) returns under result, per its API reference:

Modules Google does not show are omitted. For the data model in depth, see typed SERP data and what a SERP contains.

Your options

No option returns the same JSON from the same corpus. Pick by what you used the API for.

1. Google’s suggested path: Agent Search (formerly Vertex AI Search)

Google’s overview page calls Vertex AI Search “a favorable alternative for searching up to 50 domains” and asks customers who need full web search to contact Google. Google Cloud’s documentation now names the product Agent Search (formerly Vertex AI Search) and describes it as search within your public websites and within data stores containing your own data. If your engine searched your own site or a fixed list of domains, this is the closest fit. If you searched the whole web, it is not a like-for-like replacement.

2. Grounding with Google Search (Gemini API)

Grounding with Google Search connects a Gemini model to real-time web content and returns the model’s answer with citations and the search queries it ran. The output is a generated answer, not a ranked list of results, and the documentation points to usage requirements for displaying search suggestions. On the Gemini API pricing page (checked 2026-09-17), Gemini 3.x models list 5,000 free search requests per month shared across those models, then $14 per 1,000 requests; Gemini 2.5 models are billed per grounded prompt. Choose it if you were feeding search results to an LLM anyway. Do not choose it if you need positions, snippets or a stable result list.

3. A SERP API

SERP APIs return the Google results page as JSON. Prices below were read from each vendor’s own page on 2026-09-17; they change, so recheck before you commit.

For a broader field comparison, see best SERP APIs, cheapest SERP API and SerpApi compared.

4. An independent index

Brave’s Search API page (checked 2026-09-17) lists $5 per 1,000 requests with $5 in free credits every month and a capacity of 50 queries per second, and states that the results come from Brave’s own independent index rather than from Google or Bing. That is a valid choice when you need “web results” and not “Google’s results”; rankings will differ from Google’s.

Decision criteria

1. Did you search your own sites? Agent Search, or site search you host yourself.
2. Did you feed results into an LLM prompt? Grounding with Google Search, or a SERP API if you want to control which results are used.
3. Did you measure Google visibility (positions, SERP features, AI Overviews)? A SERP API. Google’s help page says Custom Search searches a subset of the web index and lacks features such as Oneboxes.
4. Do you just need a web index? An independent index such as Brave.

Parameter migration map

For a SERP API migration, most of your request logic carries over. This API’s Google Search request has two shapes:

• Standard shape: query plus country (or gl), with optional pages (1 to 10), hl, location or uule, and device.
• URL shape: url, a complete Google web search URL that may carry only q, gl, num, hl and uule. A URL with any other parameter (start, tbs, safe, tbm and so on) is refused with a 400. num is read as result depth, rounded up to whole pages of ten, up to ten pages. url cannot be combined with query or pages.

The operator syntax is covered in Google search operators, and every parameter in Google search parameters.

Change one structure instead of mapping it: Custom Search forced a loop of ten-result requests. Here, one request with pages: 5 returns the first five pages with a page number on every organic result. Replace pagination loops with a single request where you can.

Python

Using requests against the HTTP API:

import os
from datetime import date, timedelta
from urllib.parse import urlparse

import requests

API = "https://api.answerline.dev"
HEADERS = {"Authorization": f"Bearer {os.environ['ANSWERLINE_API_KEY']}"}
DATE_RESTRICT = {"d1": 1, "w1": 7, "m1": 30, "y1": 365}


def to_request(cse: dict) -> dict:
    """Translate Custom Search JSON API parameters into a Google Search request body."""
    q = cse["q"]
    if cse.get("exactTerms"):
        q += f' "{cse["exactTerms"]}"'
    if cse.get("excludeTerms"):
        q += "".join(f" -{term}" for term in cse["excludeTerms"].split())
    if cse.get("siteSearch"):
        q += f' {"-" if cse.get("siteSearchFilter") == "e" else ""}site:{cse["siteSearch"]}'
    if cse.get("fileType"):
        q += f" filetype:{cse['fileType']}"

    restrict = cse.get("dateRestrict")
    if restrict:
        if restrict not in DATE_RESTRICT:
            raise ValueError(f"no mapping for dateRestrict={restrict}")
        q += f" after:{date.today() - timedelta(days=DATE_RESTRICT[restrict])}"

    last = int(cse.get("start", 1)) + int(cse.get("num", 10)) - 1
    return {"query": q, "gl": cse.get("gl", "us"), "pages": min(10, -(-last // 10))}


def search(cse: dict) -> list[dict]:
    """Return Custom-Search-style items so existing callers keep working."""
    r = requests.post(f"{API}/v1/monitor/google", json=to_request(cse), headers=HEADERS, timeout=360)
    r.raise_for_status()
    first_page = (int(cse.get("start", 1)) - 1) // 10 + 1
    return [
        {
            "title": o["title"],
            "link": o["link"],
            "displayLink": urlparse(o["link"]).hostname,
            "snippet": o.get("snippet", ""),
        }
        for o in r.json()["result"].get("organicResults", [])
        if o["page"] >= first_page
    ]


print(search({"q": "annual report", "siteSearch": "example.com", "fileType": "pdf", "gl": "us"}))

TypeScript

const API = "https://api.answerline.dev";
const DATE_RESTRICT: Record<string, number> = { d1: 1, w1: 7, m1: 30, y1: 365 };

type CseParams = {
  q: string; gl?: string; num?: number; start?: number;
  dateRestrict?: string; exactTerms?: string; excludeTerms?: string;
  siteSearch?: string; siteSearchFilter?: "e" | "i"; fileType?: string;
};

function toRequest(cse: CseParams): Record<string, unknown> {
  let q = cse.q;
  if (cse.exactTerms) q += ` "${cse.exactTerms}"`;
  if (cse.excludeTerms) q += cse.excludeTerms.split(/\s+/).map((t) => ` -${t}`).join("");
  if (cse.siteSearch) q += ` ${cse.siteSearchFilter === "e" ? "-" : ""}site:${cse.siteSearch}`;
  if (cse.fileType) q += ` filetype:${cse.fileType}`;

  if (cse.dateRestrict) {
    const days = DATE_RESTRICT[cse.dateRestrict];
    if (!days) throw new Error(`no mapping for dateRestrict=${cse.dateRestrict}`);
    q += ` after:${new Date(Date.now() - days * 86_400_000).toISOString().slice(0, 10)}`;
  }

  const last = (cse.start ?? 1) + (cse.num ?? 10) - 1;
  return { query: q, gl: cse.gl ?? "us", pages: Math.min(10, Math.ceil(last / 10)) };
}

export async function search(cse: CseParams) {
  const res = await fetch(`${API}/v1/monitor/google`, {
    method: "POST",
    headers: { Authorization: `Bearer ${process.env.ANSWERLINE_API_KEY}`, "Content-Type": "application/json" },
    body: JSON.stringify(toRequest(cse)),
  });
  if (!res.ok) throw new Error(`${res.status}: ${await res.text()}`);
  const { result } = await res.json();
  const firstPage = Math.floor(((cse.start ?? 1) - 1) / 10) + 1;
  type Organic = { title: string; link: string; snippet?: string; page: number };
  return (result.organicResults ?? []).filter((o: Organic) => o.page >= firstPage).map((o: Organic) => ({
    title: o.title,
    link: o.link,
    displayLink: new URL(o.link).hostname,
    snippet: o.snippet ?? "",
  }));
}

Response mapping

Migration checklist

1. Inventory calls. List every call site with its parameters, especially cx engines restricted to specific sites; those may belong in Agent Search instead.
2. Decide per use. Site search, LLM grounding, Google visibility and generic web results lead to different products (see the criteria above).
3. Translate requests with a function like to_request, and log any parameter you cannot map instead of dropping it silently.
4. Collapse pagination. Replace start loops with pages where you want the top N pages.
5. Run both in parallel for a sample of your own queries before the deadline, and compare what matters to you. Expect differences: Custom Search results came from a subset of the index.
6. Budget. At the Starter tier, a Google Search task costs $0.00103, and each extra page $0.00071 more. pages: 3 is $0.00103 + $0.00071 x 2 = $0.00245. A query that carries siteSearch or fileType as site: or filetype: costs 5x: $0.01225 for the same three pages. See /docs/credits and cost planning.
7. Move bulk jobs to async. Batches of up to 500 tasks per call, with webhooks, avoid synchronous timeouts; see sync, async and webhooks and rank tracking with batches.
8. Handle errors. Invalid combinations (for example url with pages, or country and gl with different values) return 400 with details[] naming the field; see API errors and retries.

Pitfalls

• Off-by-one on start. Custom Search’s start is 1-based: start=11 is the first result of page 2, not page 1.
• Assuming gl means the same thing. In Custom Search it boosts results from a country; in a SERP request it sets the market the search is run in. Results will shift.
• Toggling “Search the entire web” off during evaluation. Google’s help page says it cannot be turned back on.
• Waiting for December. Parallel runs, quota planning and customer communication take longer than the code change.

For related reading, see is there an official Google Search API and, if you also used Microsoft’s service, Bing Search API alternatives.

Request fields and response examples are on the Google Search engine page.

Keep reading