What replaces the Google Custom Search JSON API?

Google's Custom Search JSON API is closed to new customers and ends on 1 January 2027; Google's own replacements cover up to 50 sites, and the full web by request. For raw web results, use Brave Search API, Exa, Perplexity Search API, Parallel or Tavily; SerpApi still returns Google's results.

Last checked: 20 October 2026. Disclosure: Grounder (grounder.net) publishes this page and routes searches to several of the providers below, including Brave, Exa, Parallel, Perplexity and Tavily. Grounder does not search Google.

Quick comparison

What Google has said

• Overview page (updated 18 February 2026): "The Custom Search JSON API is closed to new customers." "Existing Custom Search JSON API customers have until January 1, 2027 to transition to an alternative solution." Until then: 100 queries a day free, then $5 per 1,000, up to 10,000 a day. It adds: "Vertex AI Search is a favorable alternative for searching up to 50 domains," and asks anyone whose use case needs full web search to contact Google.
• Programmable Search Engine blog (20 January 2026): all new engines must use "Sites to search", which the post calls free and limits to 50 domains. Existing engines can keep "Search the entire web" until 1 January 2027. Users of more than 50 domains are asked to contact Google about its full web search solution, and every transition must finish by 1 January 2027.
• Partner-only API (secondary): Search Engine Journal (11 September 2026) reported a partner-only Web Search Service API with full web search, up to 20 results per request and a partner client ID. Google's pages for it returned 404 when we fetched them, so it is unconfirmed.

The options

Sites to search and Vertex AI Search (Google)

• What it is: Google's replacements. Sites to search powers a Search Element over up to 50 sites you list; Vertex AI Search is what Google suggests for JSON API users with up to 50 domains.
• Pricing: Sites to search is free. Google Cloud's pricing page for Vertex AI Search is titled "Agent Search pricing"; we could not read its rates.
• Strengths: Google ranking over sites you choose; no API change for an embedded box.
• Limits: 50 domains at most; no whole-web search for new engines. Google's own help page says engines set to search the whole web return a subset of the Google Web Search corpus and lack features such as Oneboxes and real-time results.
• Pick it when: you search your own sites, not the open web.

Grounding with Google Search (Gemini API)

• What it is: A Gemini tool: the model decides when to search and returns a cited answer, not results.
• Pricing: Gemini 3 and newer: 5,000 queries a month free on the paid tier, then $14 per 1,000. Gemini 2.5: $35 per 1,000 grounded prompts after 1,500 requests a day free.
• Strengths: Google's index with citations.
• Limits: Terms bar caching, framing, syndicating, analyzing or training on results and using links to build an index, with narrow storage exceptions, and require showing Search Suggestions.
• Pick it when: you build on Gemini and want an answer, not results.

Brave Search API

• What it is: Search on Brave's own index (no fallback to Google or Bing, per Brave); LLM Context returns extracted chunks.
• Pricing: $5 per 1,000 requests; $5 monthly free credit, which Brave's blog ties to attributing Brave.
• Strengths: 50 queries per second; search operators go inside q, so site restriction is a query string, not a structured filter.
• Limits: Terms bar storing results or training AI on them without a storage-rights plan. Local MCP server only.
• Pick it when: you want raw results from an independent index.

Exa

• What it is: Own-index semantic search returning highlights, text or summaries.
• Pricing: $7 per 1,000 requests (auto) or $4 (lite); $10 of free credit a month.
• Strengths: includeDomains and excludeDomains take up to 1,200 entries each, as hostnames, path prefixes or wildcard subdomains. Artificial Analysis scored auto 80.3 and lite 79.2 (measured 19 October 2026), with dynamic highlights.
• Limits: Dynamic Highlights is a preview (auto without it scored 73.7, measured 17 August 2026).
• Pick it when: you want site-restricted or whole-web search with page content.

Perplexity Search API

• What it is: Raw ranked results from Perplexity's index, with no generated answer.
• Pricing: $5 per 1,000 requests, or $1 with fast search; no free tier published.
• Strengths: search_domain_filter takes up to 20 domains as an allowlist or a denylist, not both at once. Artificial Analysis scored medium 80.0 (measured 27 August 2026).
• Limits: Its privacy page and FAQ disagree on default data retention.
• Pick it when: you want a domain allowlist plus country and language filters.

Parallel Search API

• What it is: Search on Parallel's own index: an objective in, ranked excerpts out.
• Pricing: $1 per 1,000 requests (turbo, fast) or $5 (basic, advanced); $5 of free credit a month.
• Strengths: source_policy includes or excludes domains and paths.
• Limits: Turbo is English and Japanese only.
• Pick it when: you want ranked excerpts from a full-web index at $1 per 1,000 requests.

Tavily

• What it is: A search and extraction API with crawl, map and research endpoints.
• Pricing: $0.008 per credit; a basic search is 1 credit ($8 per 1,000). 1,000 free credits a month.
• Strengths: include_domains takes up to 300 domains, exclude_domains up to 150.
• Limits: Artificial Analysis scored basic 65.6 (measured 17 August 2026), below Exa and Perplexity. Its privacy policy says Tavily may use query data to improve responses to future queries.
• Pick it when: you need domain-restricted search plus extraction.

SerpApi

• What it is: A scraper API that returns JSON for Google and many other engines. Serper is a cheaper Google-only scraper ($0.30 to $1.00 per 1,000 by pack, credits valid for 6 months).
• Pricing: From $25 a month for 1,000 searches to $1.96 per 1,000 at 54,000,000 a month (the largest plan); 250 free a month.
• Strengths: Broad engine coverage; official MCP server.
• Limits: Scraped, not licensed. Google sued SerpApi under the DMCA in December 2025; a court dismissed both claims on 20 July 2026, with leave to amend on part. Google filed an amended complaint on 10 August 2026 and SerpApi asked the court to dismiss it in late August; we found no ruling.
• Pick it when: you want Google's results plus other engines in one API.

Where Grounder fits

Grounder is a web search API and MCP server for AI agents. One search goes across 10 web search providers (Ozone, Tavily, Exa, Brave, Perplexity, You.com, Firecrawl, Octen, Parallel and Valyu), results are deduplicated, and each is marked with its provider and whether it is licensed. It returns ranked results, not a generated answer, and does not search Google, so its ranking will not match Google's.

• Site-restricted search, the common Custom Search use, maps to domain filters (include or exclude), which all 10 providers support. Grounder cannot index your private content.
• Licensed sources only is on by default for signed-up accounts: a search returns only licensed results and can return fewer than top_k. Turn it off on the account page, or per search with routing_settings.licensed_only set to false.
• New search keys are no longer made: agents connect over MCP, and the HTTP API accepts only keys made before 13 October 2026. Search records are kept indefinitely (/privacy). No storage rights are published for the providers' results.

Grounder's benchmark, run by Grounder, searched 887 questions on each provider's default search. Table as of 20 October 2026 05:22 UTC:

Grounder optimized answers only from licensed sources, so it gets fewer answers right than the all-sources row, which searches every source. Grounder optimized is what a signed-up account gets by default; the all-sources row is the same routing with nothing filtered out. Both Grounder rows are routed as Grounder routes on 19 October 2026, and that routing was tuned on these same searches. On SimpleQA the all-sources row is first (97.8%, against 96.0% for Octen and Perplexity); on SealQA Parallel (44.9%) is ahead of it (36.1%). With Licensed sources only on, Grounder scores below every single provider except Ozone on both benchmarks; all 12 rows are on /benchmarks.

• Pick it when: your agent speaks MCP and you want one connection to 10 search providers with licensed results flagged. Skip it if you need an HTTP key today, Google's own results, or search over your private content.

How we compared

Criteria: whole-web or site-restricted, raw results or answer, price and caps, data terms, time left. Sources are each vendor's own pages, checked 20 October 2026. Quality figures come from the Artificial Analysis Search Index (data page dated 19 October 2026; one answer model; rows measured on different dates, shown above) and from the benchmark at /benchmarks, which Grounder ran. From 8 to 13 October 2026 each question was searched once on every provider's default search (top 10 results, 60 seconds); Z.ai's GLM-5.3-Flash at high reasoning effort answered from those results alone, and OpenAI's SimpleQA grader graded the answers. Wider comparison: best search API for AI agents.

FAQ

Is the Google Custom Search JSON API shutting down, and what does it cost until then?

Yes. Google says it is closed to new customers and that existing customers have until 1 January 2027 to move; 73 days remain as of 20 October 2026. Until then it costs $5 per 1,000 queries after 100 free a day, up to 10,000 a day.

Can I still search the entire web with Programmable Search, and what does Google offer instead?

Not with a new engine: from 20 January 2026 new engines must use Sites to search, up to 50 domains. Google points JSON API users with up to 50 domains to Vertex AI Search and asks anyone who needs the whole web to contact it. Grounding with Google Search in the Gemini API is a separate product that returns a cited answer, not results.

Which alternatives return raw results for LLMs?

Brave Search API, Exa, Perplexity Search API, Parallel and Tavily return result lists rather than a generated answer by default, and each can restrict results to named domains.

Is a Google SERP scraper a safe replacement?

It keeps Google's results but depends on scraping them: Google sued SerpApi under the DMCA in December 2025 and the case continues on an amended complaint, and Serper publishes no legal-protection programme or data-retention commitment.

Will a replacement return the same results as Google?

No. Google's help page says that even engines set to search the whole web are unlikely to match Google Web Search, and every other provider has its own index and ranking. Test on your real queries before cutting over.

Next step

Your first 1,000 searches are free, up to 100 a day. No card needed. Connect an MCP client to https://api.grounder.net/mcp and sign in with Google, GitHub or Apple; the steps are in Docs.