Google Custom Search API Shutdown: What to Use Instead

What is actually happening

Google’s Custom Search JSON API page now opens by stating that the API is closed to new customers, and that existing customers have until January 1, 2027 to move. The published terms hold until then: 100 free queries a day, $5 per 1,000 above that, and a hard ceiling of 10,000 queries a day.

That ceiling is worth noting, because it tells you who this API was ever for. At 10,000 queries a day maximum, Custom Search was never infrastructure for a rank tracker or a monitoring product. It was a cheap way to bolt search onto an application, and a very large number of applications did exactly that.

What Google says to use instead

Two answers, and both are narrower than what they replace.

The API overview recommends Vertex AI Search for searching up to 50 domains, and a full web search solution for anything larger, which requires contacting Google.

The dedicated migration guide recommends Agent Search, described as Google-quality site-restricted search, deployed as a widget or through an API.

Together they describe one move. Google is replacing a general-purpose search API with a site-search product, and pointing anyone who wants the open web at a sales conversation.

What the official path costs you in work

The migration guide is not a swapped base URL. It describes:

1. Enabling Agent Search in the Google Cloud Console.
2. Creating an app configured for site search with AI mode.
3. Turning on Enterprise edition features, which the guide makes mandatory for website search.
4. Creating a data store and populating it with website content.
5. Specifying the URL patterns to index.

Authentication changes too. Custom Search used an API key; Agent Search primarily expects OAuth 2.0, though the searchLite method still accepts an API key. The guide also notes that features like summaries and domain verification can carry additional indexing costs.

None of that is unreasonable for a site-search product. It is a lot of machinery if all you wanted was ten results as JSON.

The fork: which half of Custom Search were you using?

Answer this before you compare anything, because everything below branches on it.

You searched your own site or a fixed list of domains. Google’s path fits. Vertex AI Search and Agent Search are built for exactly this, the indexing model is a genuine upgrade over a Programmable Search Engine, and the AI features are real. Follow the migration guide.

You searched the open web. Google has no like-for-like replacement for you at self-serve pricing. What you want is a SERP API: send a query, get the live Google results page back as structured JSON.

The open-web replacements, priced

Every rate below is per 1,000 queries at ten results, which is the shape Custom Search returned, and every one is the vendor’s real-time mode.

Option	Rate per 1,000	Index	Parsed AI Overview
Custom Search JSON API (until 2027)	$5.00, capped at 10,000/day	Programmable Search Engine	No
DataForSEO	~$2.00	Google	Yes, behind a flag that adds ~3s
cloro	$1.25 to $2.00, AI Overview included	Google	Yes
SerpApi	$2.00 to $4.00	Google plus 80+ engines	Yes, often a second billed call
Brave Search API	$5.00	Independent index, not Google	No
ScraperAPI	~$12.00 via its structured Google endpoint	Google	No

DataForSEO is the cheapest way to keep querying Google and the closest thing to a drop-in on price. SerpApi is the pick if Custom Search was one of several engines you queried, since it also covers Bing, Baidu, Yandex and a set of retail verticals. Brave Search API is the one to reach for if you would rather leave Google’s index entirely, and Anthropic built Claude’s web search on it. ScraperAPI and Oxylabs both return Google through a structured endpoint, though both are general scrapers carrying the cost of every other target they support.

At $5 per 1,000, Custom Search sat above most of this category and roughly eight times the cheapest option, so for a lot of teams the shutdown is an unwelcome deadline attached to a cheaper bill. The other change is the ceiling. Custom Search stopped at 10,000 queries a day whatever you paid, and per-call vendors have no equivalent limit, so a workload that was shaped around that cap can be rebuilt around what it actually needs.

What the results page looks like now

If your Custom Search integration is a few years old, the page it was querying has changed underneath it. In cloro’s own SERP feature census we scored 65,945 results and found an AI Overview on 87.2% of US results and People Also Ask on 90.6%. Custom Search never returned either.

So a straight port of ten blue links reproduces the 2020 page. A user in 2026 sees something else. If the reason you queried Google was to know what Google says, the AI Overview and its cited sources are now part of that answer, and it is worth picking a replacement that returns them rather than adding a second migration later.

Migrating to a SERP API

The move is small compared with the official path. From Custom Search:

GET https://www.googleapis.com/customsearch/v1
  ?key=<key>&cx=<engine id>&q=serp+api
  Results: items[] → { title, link, snippet, displayLink }

To a SERP API, using cloro’s as the worked example:

POST https://api.cloro.dev/v1/monitor/google
  Authorization: Bearer <key>
  { "query": "serp api", "country": "US",
    "include": { "aioverview": { "markdown": true } } }
  Results: result.organicResults[] → { position, title, link, snippet, displayedLink }

Custom Search	cloro
q	query
key query parameter	Authorization: Bearer header
cx (search engine id)	not needed, there is no engine to configure
items[]	result.organicResults[]
title	title
link	link
snippet	snippet
displayLink	displayedLink
—	position
—	result.aioverview, with cited sources

The cx parameter disappearing is the part worth pausing on. A Programmable Search Engine was a configured object you maintained in a console; a SERP API has no equivalent, which removes a piece of state but also removes the site restriction it gave you. If that restriction was the point, you are in the site-search half of the fork above.

Whichever vendor you pick, port a real query set and diff it against your current Custom Search results before committing. Most of this category offers a free tier large enough for that: DataForSEO, SerpApi and cloro all do.

Do not wait for a first-party replacement

This has happened before, recently, and the outcome is on the record. Microsoft retired the Bing Search API v7 on August 11, 2025; the endpoints now return HTTP 410 Gone, and the official replacement was an agent product rather than a SERP API. Teams that waited for a like-for-like successor did not get one, and migrated under time pressure instead. Our Bing Search API guide covers how that played out.

Google has given more notice and a longer runway. The pattern is the same: the first-party path leads to an AI product with a different shape, and the like-for-like path is third-party. Pick early enough to run both in parallel for a month.